2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
28 #include "byte-order.h"
30 #include "classifier.h"
34 #include "dynamic-string.h"
35 #include "fail-open.h"
39 #include "mac-learning.h"
40 #include "multipath.h"
46 #include "ofp-print.h"
48 #include "ofproto-sflow.h"
50 #include "openflow/nicira-ext.h"
51 #include "openflow/openflow.h"
52 #include "openvswitch/datapath-protocol.h"
56 #include "poll-loop.h"
61 #include "stream-ssl.h"
65 #include "unaligned.h"
70 VLOG_DEFINE_THIS_MODULE(ofproto);
72 COVERAGE_DEFINE(facet_changed_rule);
73 COVERAGE_DEFINE(facet_revalidate);
74 COVERAGE_DEFINE(odp_overflow);
75 COVERAGE_DEFINE(ofproto_agg_request);
76 COVERAGE_DEFINE(ofproto_costly_flags);
77 COVERAGE_DEFINE(ofproto_ctlr_action);
78 COVERAGE_DEFINE(ofproto_del_rule);
79 COVERAGE_DEFINE(ofproto_error);
80 COVERAGE_DEFINE(ofproto_expiration);
81 COVERAGE_DEFINE(ofproto_expired);
82 COVERAGE_DEFINE(ofproto_flows_req);
83 COVERAGE_DEFINE(ofproto_flush);
84 COVERAGE_DEFINE(ofproto_invalidated);
85 COVERAGE_DEFINE(ofproto_no_packet_in);
86 COVERAGE_DEFINE(ofproto_ofp2odp);
87 COVERAGE_DEFINE(ofproto_packet_in);
88 COVERAGE_DEFINE(ofproto_packet_out);
89 COVERAGE_DEFINE(ofproto_queue_req);
90 COVERAGE_DEFINE(ofproto_recv_openflow);
91 COVERAGE_DEFINE(ofproto_reinit_ports);
92 COVERAGE_DEFINE(ofproto_unexpected_rule);
93 COVERAGE_DEFINE(ofproto_uninstallable);
94 COVERAGE_DEFINE(ofproto_update_port);
96 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
97 * flow translation. */
98 #define MAX_RESUBMIT_RECURSION 16
103 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
104 struct netdev *netdev;
105 struct ofp_phy_port opp;
107 struct cfm *cfm; /* Connectivity Fault Management, if any. */
110 static void ofport_free(struct ofport *);
111 static void ofport_run(struct ofproto *, struct ofport *);
112 static void ofport_wait(struct ofport *);
114 struct action_xlate_ctx {
115 /* action_xlate_ctx_init() initializes these members. */
118 struct ofproto *ofproto;
120 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
121 * this flow when actions change header fields. */
124 /* The packet corresponding to 'flow', or a null pointer if we are
125 * revalidating without a packet to refer to. */
126 const struct ofpbuf *packet;
128 /* If nonnull, called just before executing a resubmit action.
130 * This is normally null so the client has to set it manually after
131 * calling action_xlate_ctx_init(). */
132 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
134 /* If true, the speciality of 'flow' should be checked before executing
135 * its actions. If special_cb returns false on 'flow' rendered
136 * uninstallable and no actions will be executed. */
139 /* xlate_actions() initializes and uses these members. The client might want
140 * to look at them after it returns. */
142 struct ofpbuf *odp_actions; /* Datapath actions. */
143 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
144 bool may_set_up_flow; /* True ordinarily; false if the actions must
145 * be reassessed for every packet. */
146 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
148 /* xlate_actions() initializes and uses these members, but the client has no
149 * reason to look at them. */
151 int recurse; /* Recursion level, via xlate_table_action. */
152 int last_pop_priority; /* Offset in 'odp_actions' just past most
153 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
156 static void action_xlate_ctx_init(struct action_xlate_ctx *,
157 struct ofproto *, const struct flow *,
158 const struct ofpbuf *);
159 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
160 const union ofp_action *in, size_t n_in);
162 /* An OpenFlow flow. */
164 long long int used; /* Time last used; time created if not used. */
165 long long int created; /* Creation time. */
169 * - Do include packets and bytes from facets that have been deleted or
170 * whose own statistics have been folded into the rule.
172 * - Do include packets and bytes sent "by hand" that were accounted to
173 * the rule without any facet being involved (this is a rare corner
174 * case in rule_execute()).
176 * - Do not include packet or bytes that can be obtained from any facet's
177 * packet_count or byte_count member or that can be obtained from the
178 * datapath by, e.g., dpif_flow_get() for any facet.
180 uint64_t packet_count; /* Number of packets received. */
181 uint64_t byte_count; /* Number of bytes received. */
183 ovs_be64 flow_cookie; /* Controller-issued identifier. */
185 struct cls_rule cr; /* In owning ofproto's classifier. */
186 uint16_t idle_timeout; /* In seconds from time of last use. */
187 uint16_t hard_timeout; /* In seconds from time of creation. */
188 bool send_flow_removed; /* Send a flow removed message? */
189 int n_actions; /* Number of elements in actions[]. */
190 union ofp_action *actions; /* OpenFlow actions. */
191 struct list facets; /* List of "struct facet"s. */
194 static struct rule *rule_from_cls_rule(const struct cls_rule *);
195 static bool rule_is_hidden(const struct rule *);
197 static struct rule *rule_create(const struct cls_rule *,
198 const union ofp_action *, size_t n_actions,
199 uint16_t idle_timeout, uint16_t hard_timeout,
200 ovs_be64 flow_cookie, bool send_flow_removed);
201 static void rule_destroy(struct ofproto *, struct rule *);
202 static void rule_free(struct rule *);
204 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
205 static void rule_insert(struct ofproto *, struct rule *);
206 static void rule_remove(struct ofproto *, struct rule *);
208 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
209 static void rule_get_stats(const struct rule *, uint64_t *packets,
212 /* An exact-match instantiation of an OpenFlow flow. */
214 long long int used; /* Time last used; time created if not used. */
218 * - Do include packets and bytes sent "by hand", e.g. with
221 * - Do include packets and bytes that were obtained from the datapath
222 * when a flow was deleted (e.g. dpif_flow_del()) or when its
223 * statistics were reset (e.g. dpif_flow_put() with
224 * DPIF_FP_ZERO_STATS).
226 * - Do not include any packets or bytes that can currently be obtained
227 * from the datapath by, e.g., dpif_flow_get().
229 uint64_t packet_count; /* Number of packets received. */
230 uint64_t byte_count; /* Number of bytes received. */
232 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
233 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
235 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
236 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
237 long long int rs_used; /* Used time pushed to resubmit children. */
239 /* Number of bytes passed to account_cb. This may include bytes that can
240 * currently obtained from the datapath (thus, it can be greater than
242 uint64_t accounted_bytes;
244 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
245 struct list list_node; /* In owning rule's 'facets' list. */
246 struct rule *rule; /* Owning rule. */
247 struct flow flow; /* Exact-match flow. */
248 bool installed; /* Installed in datapath? */
249 bool may_install; /* True ordinarily; false if actions must
250 * be reassessed for every packet. */
251 size_t actions_len; /* Number of bytes in actions[]. */
252 struct nlattr *actions; /* Datapath actions. */
253 tag_type tags; /* Tags (set only by hooks). */
254 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
257 static struct facet *facet_create(struct ofproto *, struct rule *,
259 const struct ofpbuf *packet);
260 static void facet_remove(struct ofproto *, struct facet *);
261 static void facet_free(struct facet *);
263 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
264 static bool facet_revalidate(struct ofproto *, struct facet *);
266 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
267 static void facet_uninstall(struct ofproto *, struct facet *);
268 static void facet_flush_stats(struct ofproto *, struct facet *);
270 static void facet_make_actions(struct ofproto *, struct facet *,
271 const struct ofpbuf *packet);
272 static void facet_update_stats(struct ofproto *, struct facet *,
273 const struct dpif_flow_stats *);
274 static void facet_push_stats(struct ofproto *, struct facet *);
276 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
277 const struct flow *, bool clone);
281 uint64_t datapath_id; /* Datapath ID. */
282 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
283 char *mfr_desc; /* Manufacturer. */
284 char *hw_desc; /* Hardware. */
285 char *sw_desc; /* Software version. */
286 char *serial_desc; /* Serial number. */
287 char *dp_desc; /* Datapath description. */
291 struct netdev_monitor *netdev_monitor;
292 struct hmap ports; /* Contains "struct ofport"s. */
293 struct shash port_by_name;
297 struct netflow *netflow;
298 struct ofproto_sflow *sflow;
301 struct classifier cls;
302 struct timer next_expiration;
306 bool need_revalidate;
307 struct tag_set revalidate_set;
309 /* OpenFlow connections. */
310 struct connmgr *connmgr;
312 /* Hooks for ovs-vswitchd. */
313 const struct ofhooks *ofhooks;
316 /* Used by default ofhooks. */
317 struct mac_learning *ml;
320 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
321 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
323 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
325 static const struct ofhooks default_ofhooks;
327 static uint64_t pick_datapath_id(const struct ofproto *);
328 static uint64_t pick_fallback_dpid(void);
330 static void ofproto_flush_flows__(struct ofproto *);
331 static int ofproto_expire(struct ofproto *);
332 static void flow_push_stats(struct ofproto *, const struct rule *,
333 struct flow *, uint64_t packets, uint64_t bytes,
336 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
338 static void handle_openflow(struct ofconn *, struct ofpbuf *);
340 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
341 static void update_port(struct ofproto *, const char *devname);
342 static int init_ports(struct ofproto *);
343 static void reinit_ports(struct ofproto *);
345 static void ofproto_unixctl_init(void);
348 ofproto_create(const char *datapath, const char *datapath_type,
349 const struct ofhooks *ofhooks, void *aux,
350 struct ofproto **ofprotop)
352 char local_name[IF_NAMESIZE];
359 ofproto_unixctl_init();
361 /* Connect to datapath and start listening for messages. */
362 error = dpif_create_and_open(datapath, datapath_type, &dpif);
364 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
367 error = dpif_recv_set_mask(dpif,
368 ((1u << DPIF_UC_MISS) |
369 (1u << DPIF_UC_ACTION) |
370 (1u << DPIF_UC_SAMPLE)));
372 VLOG_ERR("failed to listen on datapath %s: %s",
373 datapath, strerror(error));
377 dpif_flow_flush(dpif);
378 dpif_recv_purge(dpif);
380 error = dpif_port_get_name(dpif, ODPP_LOCAL,
381 local_name, sizeof local_name);
383 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
384 datapath, strerror(error));
388 /* Initialize settings. */
389 p = xzalloc(sizeof *p);
390 p->fallback_dpid = pick_fallback_dpid();
391 p->datapath_id = p->fallback_dpid;
392 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
393 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
394 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
395 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
396 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
398 /* Initialize datapath. */
400 p->netdev_monitor = netdev_monitor_create();
401 hmap_init(&p->ports);
402 shash_init(&p->port_by_name);
403 p->max_ports = dpif_get_max_ports(dpif);
405 /* Initialize submodules. */
409 /* Initialize flow table. */
410 classifier_init(&p->cls);
411 timer_set_duration(&p->next_expiration, 1000);
413 /* Initialize facet table. */
414 hmap_init(&p->facets);
415 p->need_revalidate = false;
416 tag_set_init(&p->revalidate_set);
418 /* Initialize hooks. */
420 p->ofhooks = ofhooks;
424 p->ofhooks = &default_ofhooks;
426 p->ml = mac_learning_create();
429 /* Pick final datapath ID. */
430 p->datapath_id = pick_datapath_id(p);
431 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
433 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
435 /* Initialize OpenFlow connections. */
436 p->connmgr = connmgr_create(p, datapath, local_name);
443 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
445 uint64_t old_dpid = p->datapath_id;
446 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
447 if (p->datapath_id != old_dpid) {
448 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
450 /* Force all active connections to reconnect, since there is no way to
451 * notify a controller that the datapath ID has changed. */
452 ofproto_reconnect_controllers(p);
457 ofproto_set_controllers(struct ofproto *p,
458 const struct ofproto_controller *controllers,
459 size_t n_controllers)
461 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
465 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
467 connmgr_set_fail_mode(p->connmgr, fail_mode);
470 /* Drops the connections between 'ofproto' and all of its controllers, forcing
471 * them to reconnect. */
473 ofproto_reconnect_controllers(struct ofproto *ofproto)
475 connmgr_reconnect(ofproto->connmgr);
478 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
479 * in-band control should guarantee access, in the same way that in-band
480 * control guarantees access to OpenFlow controllers. */
482 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
483 const struct sockaddr_in *extras, size_t n)
485 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
488 /* Sets the OpenFlow queue used by flows set up by in-band control on
489 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
490 * flows will use the default queue. */
492 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
494 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
498 ofproto_set_desc(struct ofproto *p,
499 const char *mfr_desc, const char *hw_desc,
500 const char *sw_desc, const char *serial_desc,
503 struct ofp_desc_stats *ods;
506 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
507 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
508 sizeof ods->mfr_desc);
511 p->mfr_desc = xstrdup(mfr_desc);
514 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
515 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
516 sizeof ods->hw_desc);
519 p->hw_desc = xstrdup(hw_desc);
522 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
523 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
524 sizeof ods->sw_desc);
527 p->sw_desc = xstrdup(sw_desc);
530 if (strlen(serial_desc) >= sizeof ods->serial_num) {
531 VLOG_WARN("truncating serial_desc, must be less than %zu "
533 sizeof ods->serial_num);
535 free(p->serial_desc);
536 p->serial_desc = xstrdup(serial_desc);
539 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
540 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
541 sizeof ods->dp_desc);
544 p->dp_desc = xstrdup(dp_desc);
549 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
551 return connmgr_set_snoops(ofproto->connmgr, snoops);
555 ofproto_set_netflow(struct ofproto *ofproto,
556 const struct netflow_options *nf_options)
558 if (nf_options && !sset_is_empty(&nf_options->collectors)) {
559 if (!ofproto->netflow) {
560 ofproto->netflow = netflow_create();
562 return netflow_set_options(ofproto->netflow, nf_options);
564 netflow_destroy(ofproto->netflow);
565 ofproto->netflow = NULL;
571 ofproto_set_sflow(struct ofproto *ofproto,
572 const struct ofproto_sflow_options *oso)
574 struct ofproto_sflow *os = ofproto->sflow;
577 struct ofport *ofport;
579 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
580 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
581 ofproto_sflow_add_port(os, ofport->odp_port,
582 netdev_get_name(ofport->netdev));
585 ofproto_sflow_set_options(os, oso);
587 ofproto_sflow_destroy(os);
588 ofproto->sflow = NULL;
592 /* Connectivity Fault Management configuration. */
594 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
596 ofproto_port_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
598 struct ofport *ofport = get_port(ofproto, port_no);
599 if (ofport && ofport->cfm){
600 cfm_destroy(ofport->cfm);
605 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
606 * basic configuration from the configuration members in 'cfm', and the set of
607 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
608 * Ignores the statistics members of 'cfm'.
610 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
612 ofproto_port_set_cfm(struct ofproto *ofproto, uint32_t port_no,
613 const struct cfm *cfm,
614 const uint16_t *remote_mps, size_t n_remote_mps)
616 struct ofport *ofport;
618 ofport = get_port(ofproto, port_no);
620 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
621 dpif_name(ofproto->dpif), port_no);
626 ofport->cfm = cfm_create();
629 ofport->cfm->mpid = cfm->mpid;
630 ofport->cfm->interval = cfm->interval;
631 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
633 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
635 if (!cfm_configure(ofport->cfm)) {
636 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
637 dpif_name(ofproto->dpif), port_no,
638 netdev_get_name(ofport->netdev));
639 cfm_destroy(ofport->cfm);
644 /* Returns the connectivity fault management object associated with 'port_no'
645 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
646 * 'port_no' or if that port does not have CFM configured. The caller must not
647 * modify or destroy the returned object. */
649 ofproto_port_get_cfm(struct ofproto *ofproto, uint32_t port_no)
651 struct ofport *ofport = get_port(ofproto, port_no);
652 return ofport ? ofport->cfm : NULL;
656 ofproto_get_datapath_id(const struct ofproto *ofproto)
658 return ofproto->datapath_id;
661 enum ofproto_fail_mode
662 ofproto_get_fail_mode(const struct ofproto *p)
664 return connmgr_get_fail_mode(p->connmgr);
668 ofproto_has_snoops(const struct ofproto *ofproto)
670 return connmgr_has_snoops(ofproto->connmgr);
674 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
676 connmgr_get_snoops(ofproto->connmgr, snoops);
680 ofproto_destroy__(struct ofproto *p, bool delete)
682 struct ofport *ofport, *next_ofport;
688 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
690 ofproto_flush_flows__(p);
691 connmgr_destroy(p->connmgr);
692 classifier_destroy(&p->cls);
693 hmap_destroy(&p->facets);
696 int error = dpif_delete(p->dpif);
697 if (error && error != ENOENT) {
698 VLOG_ERR("bridge %s: failed to destroy (%s)",
699 dpif_name(p->dpif), strerror(error));
704 netdev_monitor_destroy(p->netdev_monitor);
705 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
706 hmap_remove(&p->ports, &ofport->hmap_node);
709 shash_destroy(&p->port_by_name);
711 netflow_destroy(p->netflow);
712 ofproto_sflow_destroy(p->sflow);
714 mac_learning_destroy(p->ml);
719 free(p->serial_desc);
722 hmap_destroy(&p->ports);
728 ofproto_destroy(struct ofproto *p)
730 ofproto_destroy__(p, false);
734 ofproto_destroy_and_delete(struct ofproto *p)
736 ofproto_destroy__(p, true);
740 ofproto_run(struct ofproto *p)
742 int error = ofproto_run1(p);
744 error = ofproto_run2(p, false);
750 process_port_change(struct ofproto *ofproto, int error, char *devname)
752 if (error == ENOBUFS) {
753 reinit_ports(ofproto);
755 update_port(ofproto, devname);
761 ofproto_run1(struct ofproto *p)
763 struct ofport *ofport;
768 if (shash_is_empty(&p->port_by_name)) {
772 for (i = 0; i < 50; i++) {
773 struct dpif_upcall packet;
775 error = dpif_recv(p->dpif, &packet);
777 if (error == ENODEV) {
778 /* Someone destroyed the datapath behind our back. The caller
779 * better destroy us and give up, because we're just going to
780 * spin from here on out. */
781 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
782 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
789 handle_upcall(p, &packet);
792 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
793 process_port_change(p, error, devname);
795 while ((error = netdev_monitor_poll(p->netdev_monitor,
796 &devname)) != EAGAIN) {
797 process_port_change(p, error, devname);
800 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
801 ofport_run(p, ofport);
804 connmgr_run(p->connmgr, handle_openflow);
806 if (timer_expired(&p->next_expiration)) {
807 int delay = ofproto_expire(p);
808 timer_set_duration(&p->next_expiration, delay);
809 COVERAGE_INC(ofproto_expiration);
813 netflow_run(p->netflow);
816 ofproto_sflow_run(p->sflow);
823 ofproto_run2(struct ofproto *p, bool revalidate_all)
825 /* Figure out what we need to revalidate now, if anything. */
826 struct tag_set revalidate_set = p->revalidate_set;
827 if (p->need_revalidate) {
828 revalidate_all = true;
831 /* Clear the revalidation flags. */
832 tag_set_init(&p->revalidate_set);
833 p->need_revalidate = false;
835 /* Now revalidate if there's anything to do. */
836 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
837 struct facet *facet, *next;
839 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
841 || tag_set_intersects(&revalidate_set, facet->tags)) {
842 facet_revalidate(p, facet);
851 ofproto_wait(struct ofproto *p)
853 struct ofport *ofport;
855 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
858 dpif_recv_wait(p->dpif);
859 dpif_port_poll_wait(p->dpif);
860 netdev_monitor_poll_wait(p->netdev_monitor);
862 ofproto_sflow_wait(p->sflow);
864 if (!tag_set_is_empty(&p->revalidate_set)) {
865 poll_immediate_wake();
867 if (p->need_revalidate) {
868 /* Shouldn't happen, but if it does just go around again. */
869 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
870 poll_immediate_wake();
872 timer_wait(&p->next_expiration);
874 connmgr_wait(p->connmgr);
878 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
880 tag_set_add(&ofproto->revalidate_set, tag);
884 ofproto_get_revalidate_set(struct ofproto *ofproto)
886 return &ofproto->revalidate_set;
890 ofproto_is_alive(const struct ofproto *p)
892 return connmgr_has_controllers(p->connmgr);
896 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
899 connmgr_get_controller_info(ofproto->connmgr, info);
903 ofproto_free_ofproto_controller_info(struct shash *info)
905 struct shash_node *node;
907 SHASH_FOR_EACH (node, info) {
908 struct ofproto_controller_info *cinfo = node->data;
909 while (cinfo->pairs.n) {
910 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
917 /* Makes a deep copy of 'old' into 'port'. */
919 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
921 port->name = xstrdup(old->name);
922 port->type = xstrdup(old->type);
923 port->ofp_port = old->ofp_port;
926 /* Frees memory allocated to members of 'ofproto_port'.
928 * Do not call this function on an ofproto_port obtained from
929 * ofproto_port_dump_next(): that function retains ownership of the data in the
932 ofproto_port_destroy(struct ofproto_port *ofproto_port)
934 free(ofproto_port->name);
935 free(ofproto_port->type);
938 /* Converts a dpif_port into an ofproto_port.
940 * This only makes a shallow copy, so make sure that the dpif_port doesn't get
941 * freed while the ofproto_port is still in use. You can choose to free the
942 * ofproto_port instead of the dpif_port. */
944 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
945 struct dpif_port *dpif_port)
947 ofproto_port->name = dpif_port->name;
948 ofproto_port->type = dpif_port->type;
949 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
952 /* Initializes 'dump' to begin dumping the ports in an ofproto.
954 * This function provides no status indication. An error status for the entire
955 * dump operation is provided when it is completed by calling
956 * ofproto_port_dump_done().
959 ofproto_port_dump_start(struct ofproto_port_dump *dump,
960 const struct ofproto *ofproto)
962 struct dpif_port_dump *dpif_dump;
964 dump->state = dpif_dump = xmalloc(sizeof *dpif_dump);
965 dpif_port_dump_start(dpif_dump, ofproto->dpif);
968 /* Attempts to retrieve another port from 'dump', which must have been created
969 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
970 * 'port' and returns true. On failure, returns false.
972 * Failure might indicate an actual error or merely that the last port has been
973 * dumped. An error status for the entire dump operation is provided when it
974 * is completed by calling ofproto_port_dump_done().
976 * The ofproto owns the data stored in 'port'. It will remain valid until at
977 * least the next time 'dump' is passed to ofproto_port_dump_next() or
978 * ofproto_port_dump_done(). */
980 ofproto_port_dump_next(struct ofproto_port_dump *dump,
981 struct ofproto_port *port)
983 struct dpif_port_dump *dpif_dump = dump->state;
984 struct dpif_port dpif_port;
987 ok = dpif_port_dump_next(dpif_dump, &dpif_port);
989 ofproto_port_from_dpif_port(port, &dpif_port);
994 /* Completes port table dump operation 'dump', which must have been created
995 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
996 * error-free, otherwise a positive errno value describing the problem. */
998 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1000 struct dpif_port_dump *dpif_dump = dump->state;
1001 int error = dpif_port_dump_done(dpif_dump);
1006 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1007 * and sets '*ofp_portp' to the new port's port OpenFlow number (if 'ofp_portp'
1008 * is non-null). On failure, returns a positive errno value and sets
1009 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1011 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1012 uint16_t *ofp_portp)
1017 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1019 *ofp_portp = error ? OFPP_NONE : odp_port_to_ofp_port(odp_port);
1024 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1025 * initializes '*port' appropriately; on failure, returns a positive errno
1028 * The caller owns the data in 'port' and must free it with
1029 * ofproto_port_destroy() when it is no longer needed. */
1031 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1032 struct ofproto_port *port)
1034 struct dpif_port dpif_port;
1037 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1039 ofproto_port_from_dpif_port(port, &dpif_port);
1044 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1046 * This is almost the same as calling dpif_port_del() directly on the
1047 * datapath, but it also makes 'ofproto' close its open netdev for the port
1048 * (if any). This makes it possible to create a new netdev of a different
1049 * type under the same name, which otherwise the netdev library would refuse
1050 * to do because of the conflict. (The netdev would eventually get closed on
1051 * the next trip through ofproto_run(), but this interface is more direct.)
1053 * Returns 0 if successful, otherwise a positive errno. */
1055 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1057 uint32_t odp_port = ofp_port_to_odp_port(ofp_port);
1058 struct ofport *ofport = get_port(ofproto, odp_port);
1059 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1062 error = dpif_port_del(ofproto->dpif, odp_port);
1064 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1065 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1066 } else if (ofport) {
1067 /* 'name' is the netdev's name and update_port() is going to close the
1068 * netdev. Just in case update_port() refers to 'name' after it
1069 * destroys 'ofport', make a copy of it around the update_port()
1071 char *devname = xstrdup(name);
1072 update_port(ofproto, devname);
1078 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1079 * true if 'odp_port' exists and should be included, false otherwise. */
1081 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1083 struct ofport *ofport = get_port(ofproto, odp_port);
1084 return ofport && !(ofport->opp.config & htonl(OFPPC_NO_FLOOD));
1087 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1088 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1089 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1091 * Returns 0 if successful, otherwise a positive errno value. */
1093 ofproto_send_packet(struct ofproto *ofproto,
1094 uint32_t port_no, uint16_t vlan_tci,
1095 const struct ofpbuf *packet)
1097 struct ofpbuf odp_actions;
1100 ofpbuf_init(&odp_actions, 32);
1101 if (vlan_tci != 0) {
1102 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1103 ntohs(vlan_tci & ~VLAN_CFI));
1105 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1106 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1108 ofpbuf_uninit(&odp_actions);
1111 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1112 dpif_name(ofproto->dpif), port_no, strerror(error));
1117 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1118 * performs the 'n_actions' actions in 'actions'. The new flow will not
1121 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1122 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1123 * controllers; otherwise, it will be hidden.
1125 * The caller retains ownership of 'cls_rule' and 'actions'. */
1127 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1128 const union ofp_action *actions, size_t n_actions)
1131 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1132 rule_insert(p, rule);
1136 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1140 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1143 rule_remove(ofproto, rule);
1148 ofproto_flush_flows__(struct ofproto *ofproto)
1150 struct facet *facet, *next_facet;
1151 struct rule *rule, *next_rule;
1152 struct cls_cursor cursor;
1154 COVERAGE_INC(ofproto_flush);
1156 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1157 /* Mark the facet as not installed so that facet_remove() doesn't
1158 * bother trying to uninstall it. There is no point in uninstalling it
1159 * individually since we are about to blow away all the facets with
1160 * dpif_flow_flush(). */
1161 facet->installed = false;
1162 facet->dp_packet_count = 0;
1163 facet->dp_byte_count = 0;
1164 facet_remove(ofproto, facet);
1167 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1168 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1169 rule_remove(ofproto, rule);
1172 dpif_flow_flush(ofproto->dpif);
1176 ofproto_flush_flows(struct ofproto *ofproto)
1178 ofproto_flush_flows__(ofproto);
1179 connmgr_flushed(ofproto->connmgr);
1183 reinit_ports(struct ofproto *p)
1185 struct dpif_port_dump dump;
1186 struct sset devnames;
1187 struct ofport *ofport;
1188 struct dpif_port dpif_port;
1189 const char *devname;
1191 COVERAGE_INC(ofproto_reinit_ports);
1193 sset_init(&devnames);
1194 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1195 sset_add(&devnames, netdev_get_name(ofport->netdev));
1197 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1198 sset_add(&devnames, dpif_port.name);
1201 SSET_FOR_EACH (devname, &devnames) {
1202 update_port(p, devname);
1204 sset_destroy(&devnames);
1207 /* Opens and returns a netdev for 'dpif_port', or a null pointer if the netdev
1208 * cannot be opened. On success, also fills in 'opp'. */
1209 static struct netdev *
1210 ofport_open(const struct dpif_port *dpif_port, struct ofp_phy_port *opp)
1212 uint32_t curr, advertised, supported, peer;
1213 struct netdev_options netdev_options;
1214 enum netdev_flags flags;
1215 struct netdev *netdev;
1218 memset(&netdev_options, 0, sizeof netdev_options);
1219 netdev_options.name = dpif_port->name;
1220 netdev_options.type = dpif_port->type;
1221 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1223 error = netdev_open(&netdev_options, &netdev);
1225 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1226 "cannot be opened (%s)",
1227 dpif_port->name, dpif_port->port_no,
1228 dpif_port->name, strerror(error));
1232 netdev_get_flags(netdev, &flags);
1233 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1235 opp->port_no = htons(odp_port_to_ofp_port(dpif_port->port_no));
1236 netdev_get_etheraddr(netdev, opp->hw_addr);
1237 ovs_strzcpy(opp->name, dpif_port->name, sizeof opp->name);
1238 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1239 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1240 opp->curr = htonl(curr);
1241 opp->advertised = htonl(advertised);
1242 opp->supported = htonl(supported);
1243 opp->peer = htonl(peer);
1249 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1251 if (get_port(p, dpif_port->port_no)) {
1252 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1253 dpif_port->port_no);
1255 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1256 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1264 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1265 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1268 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1270 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1271 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1272 && a->state == b->state
1273 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1274 && a->curr == b->curr
1275 && a->advertised == b->advertised
1276 && a->supported == b->supported
1277 && a->peer == b->peer);
1280 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1281 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1282 * one with the same name or port number). */
1284 ofport_install(struct ofproto *p,
1285 struct netdev *netdev, const struct ofp_phy_port *opp)
1287 const char *netdev_name = netdev_get_name(netdev);
1288 struct ofport *ofport;
1290 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1292 /* Create ofport. */
1293 ofport = xmalloc(sizeof *ofport);
1294 ofport->netdev = netdev;
1296 ofport->odp_port = ofp_port_to_odp_port(ntohs(opp->port_no));
1299 /* Add port to 'p'. */
1300 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1301 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1302 shash_add(&p->port_by_name, netdev_name, ofport);
1304 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1308 /* Removes 'ofport' from 'p' and destroys it. */
1310 ofport_remove(struct ofproto *p, struct ofport *ofport)
1312 connmgr_send_port_status(p->connmgr, &ofport->opp, OFPPR_DELETE);
1314 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1315 hmap_remove(&p->ports, &ofport->hmap_node);
1316 shash_delete(&p->port_by_name,
1317 shash_find(&p->port_by_name,
1318 netdev_get_name(ofport->netdev)));
1320 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1323 ofport_free(ofport);
1326 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1329 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1331 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1333 ofport_remove(ofproto, port);
1337 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1339 * Does not handle a name or port number change. The caller must implement
1340 * such a change as a delete followed by an add. */
1342 ofport_modified(struct ofproto *ofproto, struct ofport *port,
1343 struct netdev *netdev, struct ofp_phy_port *opp)
1345 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1346 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1347 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1348 port->opp.state = opp->state;
1349 port->opp.curr = opp->curr;
1350 port->opp.advertised = opp->advertised;
1351 port->opp.supported = opp->supported;
1352 port->opp.peer = opp->peer;
1354 netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
1355 netdev_monitor_add(ofproto->netdev_monitor, netdev);
1357 netdev_close(port->netdev);
1358 port->netdev = netdev;
1360 connmgr_send_port_status(ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1364 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1367 cfm_run(ofport->cfm);
1369 if (cfm_should_send_ccm(ofport->cfm)) {
1370 struct ofpbuf packet;
1373 ofpbuf_init(&packet, 0);
1374 ccm = eth_compose(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1375 ETH_TYPE_CFM, sizeof *ccm);
1376 cfm_compose_ccm(ofport->cfm, ccm);
1377 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1378 ofpbuf_uninit(&packet);
1384 ofport_wait(struct ofport *ofport)
1387 cfm_wait(ofport->cfm);
1392 ofport_free(struct ofport *ofport)
1395 cfm_destroy(ofport->cfm);
1396 netdev_close(ofport->netdev);
1401 static struct ofport *
1402 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1404 struct ofport *port;
1406 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1407 hash_int(odp_port, 0), &ofproto->ports) {
1408 if (port->odp_port == odp_port) {
1416 update_port(struct ofproto *ofproto, const char *name)
1418 struct dpif_port dpif_port;
1419 struct ofp_phy_port opp;
1420 struct netdev *netdev;
1421 struct ofport *port;
1423 COVERAGE_INC(ofproto_update_port);
1425 /* Fetch 'name''s location and properties from the datapath. */
1426 netdev = (!dpif_port_query_by_name(ofproto->dpif, name, &dpif_port)
1427 ? ofport_open(&dpif_port, &opp)
1430 port = get_port(ofproto, dpif_port.port_no);
1431 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1432 /* 'name' hasn't changed location. Any properties changed? */
1433 if (!ofport_equal(&port->opp, &opp)) {
1434 ofport_modified(ofproto, port, netdev, &opp);
1436 netdev_close(netdev);
1439 /* If 'port' is nonnull then its name differs from 'name' and thus
1440 * we should delete it. If we think there's a port named 'name'
1441 * then its port number must be wrong now so delete it too. */
1443 ofport_remove(ofproto, port);
1445 ofport_remove_with_name(ofproto, name);
1446 ofport_install(ofproto, netdev, &opp);
1449 /* Any port named 'name' is gone now. */
1450 ofport_remove_with_name(ofproto, name);
1452 dpif_port_destroy(&dpif_port);
1456 init_ports(struct ofproto *p)
1458 struct dpif_port_dump dump;
1459 struct dpif_port dpif_port;
1461 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1462 if (!ofport_conflicts(p, &dpif_port)) {
1463 struct ofp_phy_port opp;
1464 struct netdev *netdev;
1466 netdev = ofport_open(&dpif_port, &opp);
1468 ofport_install(p, netdev, &opp);
1476 /* Returns true if 'rule' should be hidden from the controller.
1478 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1479 * (e.g. by in-band control) and are intentionally hidden from the
1482 rule_is_hidden(const struct rule *rule)
1484 return rule->cr.priority > UINT16_MAX;
1487 /* Creates and returns a new rule initialized as specified.
1489 * The caller is responsible for inserting the rule into the classifier (with
1490 * rule_insert()). */
1491 static struct rule *
1492 rule_create(const struct cls_rule *cls_rule,
1493 const union ofp_action *actions, size_t n_actions,
1494 uint16_t idle_timeout, uint16_t hard_timeout,
1495 ovs_be64 flow_cookie, bool send_flow_removed)
1497 struct rule *rule = xzalloc(sizeof *rule);
1498 rule->cr = *cls_rule;
1499 rule->idle_timeout = idle_timeout;
1500 rule->hard_timeout = hard_timeout;
1501 rule->flow_cookie = flow_cookie;
1502 rule->used = rule->created = time_msec();
1503 rule->send_flow_removed = send_flow_removed;
1504 list_init(&rule->facets);
1505 if (n_actions > 0) {
1506 rule->n_actions = n_actions;
1507 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1513 static struct rule *
1514 rule_from_cls_rule(const struct cls_rule *cls_rule)
1516 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1520 rule_free(struct rule *rule)
1522 free(rule->actions);
1526 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1527 * destroying any that no longer has a rule (which is probably all of them).
1529 * The caller must have already removed 'rule' from the classifier. */
1531 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1533 struct facet *facet, *next_facet;
1534 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1535 facet_revalidate(ofproto, facet);
1540 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1541 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1544 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1546 const union ofp_action *oa;
1547 struct actions_iterator i;
1549 if (out_port == htons(OFPP_NONE)) {
1552 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1553 oa = actions_next(&i)) {
1554 if (action_outputs_to_port(oa, out_port)) {
1561 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1562 * 'packet', which arrived on 'in_port'.
1564 * Takes ownership of 'packet'. */
1566 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1567 const struct nlattr *odp_actions, size_t actions_len,
1568 struct ofpbuf *packet)
1570 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1571 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1572 /* As an optimization, avoid a round-trip from userspace to kernel to
1573 * userspace. This also avoids possibly filling up kernel packet
1574 * buffers along the way. */
1575 struct dpif_upcall upcall;
1577 upcall.type = DPIF_UC_ACTION;
1578 upcall.packet = packet;
1581 upcall.userdata = nl_attr_get_u64(odp_actions);
1582 upcall.sample_pool = 0;
1583 upcall.actions = NULL;
1584 upcall.actions_len = 0;
1586 send_packet_in(ofproto, &upcall, flow, false);
1592 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1593 ofpbuf_delete(packet);
1598 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1599 * statistics appropriately. 'packet' must have at least sizeof(struct
1600 * ofp_packet_in) bytes of headroom.
1602 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1603 * applying flow_extract() to 'packet' would yield the same flow as
1606 * 'facet' must have accurately composed ODP actions; that is, it must not be
1607 * in need of revalidation.
1609 * Takes ownership of 'packet'. */
1611 facet_execute(struct ofproto *ofproto, struct facet *facet,
1612 struct ofpbuf *packet)
1614 struct dpif_flow_stats stats;
1616 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1618 flow_extract_stats(&facet->flow, packet, &stats);
1619 stats.used = time_msec();
1620 if (execute_odp_actions(ofproto, &facet->flow,
1621 facet->actions, facet->actions_len, packet)) {
1622 facet_update_stats(ofproto, facet, &stats);
1626 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1627 * statistics (or the statistics for one of its facets) appropriately.
1628 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1630 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1631 * with statistics for 'packet' either way.
1633 * Takes ownership of 'packet'. */
1635 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1636 struct ofpbuf *packet)
1638 struct action_xlate_ctx ctx;
1639 struct ofpbuf *odp_actions;
1640 struct facet *facet;
1644 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1646 flow_extract(packet, 0, in_port, &flow);
1648 /* First look for a related facet. If we find one, account it to that. */
1649 facet = facet_lookup_valid(ofproto, &flow);
1650 if (facet && facet->rule == rule) {
1651 facet_execute(ofproto, facet, packet);
1655 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1656 * create a new facet for it and use that. */
1657 if (rule_lookup(ofproto, &flow) == rule) {
1658 facet = facet_create(ofproto, rule, &flow, packet);
1659 facet_execute(ofproto, facet, packet);
1660 facet_install(ofproto, facet, true);
1664 /* We can't account anything to a facet. If we were to try, then that
1665 * facet would have a non-matching rule, busting our invariants. */
1666 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1667 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1668 size = packet->size;
1669 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1670 odp_actions->size, packet)) {
1671 rule->used = time_msec();
1672 rule->packet_count++;
1673 rule->byte_count += size;
1674 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1676 ofpbuf_delete(odp_actions);
1679 /* Inserts 'rule' into 'p''s flow table. */
1681 rule_insert(struct ofproto *p, struct rule *rule)
1683 struct rule *displaced_rule;
1685 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1686 if (displaced_rule) {
1687 rule_destroy(p, displaced_rule);
1689 p->need_revalidate = true;
1692 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1693 * 'flow' and an example 'packet' within that flow.
1695 * The caller must already have determined that no facet with an identical
1696 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1697 * 'ofproto''s classifier table. */
1698 static struct facet *
1699 facet_create(struct ofproto *ofproto, struct rule *rule,
1700 const struct flow *flow, const struct ofpbuf *packet)
1702 struct facet *facet;
1704 facet = xzalloc(sizeof *facet);
1705 facet->used = time_msec();
1706 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1707 list_push_back(&rule->facets, &facet->list_node);
1709 facet->flow = *flow;
1710 netflow_flow_init(&facet->nf_flow);
1711 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1713 facet_make_actions(ofproto, facet, packet);
1719 facet_free(struct facet *facet)
1721 free(facet->actions);
1725 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1727 * - Removes 'rule' from the classifier.
1729 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1730 * destroys them), via rule_destroy().
1733 rule_remove(struct ofproto *ofproto, struct rule *rule)
1735 COVERAGE_INC(ofproto_del_rule);
1736 ofproto->need_revalidate = true;
1737 classifier_remove(&ofproto->cls, &rule->cr);
1738 rule_destroy(ofproto, rule);
1741 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1743 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1744 * rule's statistics, via facet_uninstall().
1746 * - Removes 'facet' from its rule and from ofproto->facets.
1749 facet_remove(struct ofproto *ofproto, struct facet *facet)
1751 facet_uninstall(ofproto, facet);
1752 facet_flush_stats(ofproto, facet);
1753 hmap_remove(&ofproto->facets, &facet->hmap_node);
1754 list_remove(&facet->list_node);
1758 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1760 facet_make_actions(struct ofproto *p, struct facet *facet,
1761 const struct ofpbuf *packet)
1763 const struct rule *rule = facet->rule;
1764 struct ofpbuf *odp_actions;
1765 struct action_xlate_ctx ctx;
1767 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1768 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1769 facet->tags = ctx.tags;
1770 facet->may_install = ctx.may_set_up_flow;
1771 facet->nf_flow.output_iface = ctx.nf_output_iface;
1773 if (facet->actions_len != odp_actions->size
1774 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1775 free(facet->actions);
1776 facet->actions_len = odp_actions->size;
1777 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1780 ofpbuf_delete(odp_actions);
1784 facet_put__(struct ofproto *ofproto, struct facet *facet,
1785 const struct nlattr *actions, size_t actions_len,
1786 struct dpif_flow_stats *stats)
1788 struct odputil_keybuf keybuf;
1789 enum dpif_flow_put_flags flags;
1792 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1794 flags |= DPIF_FP_ZERO_STATS;
1795 facet->dp_packet_count = 0;
1796 facet->dp_byte_count = 0;
1799 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1800 odp_flow_key_from_flow(&key, &facet->flow);
1802 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1803 actions, actions_len, stats);
1806 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1807 * 'zero_stats' is true, clears any existing statistics from the datapath for
1810 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1812 struct dpif_flow_stats stats;
1814 if (facet->may_install
1815 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1816 zero_stats ? &stats : NULL)) {
1817 facet->installed = true;
1821 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1822 * to the accounting hook function in the ofhooks structure. */
1824 facet_account(struct ofproto *ofproto,
1825 struct facet *facet, uint64_t extra_bytes)
1827 uint64_t total_bytes = facet->byte_count + extra_bytes;
1829 if (ofproto->ofhooks->account_flow_cb
1830 && total_bytes > facet->accounted_bytes)
1832 ofproto->ofhooks->account_flow_cb(
1833 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1834 total_bytes - facet->accounted_bytes, ofproto->aux);
1835 facet->accounted_bytes = total_bytes;
1839 /* If 'rule' is installed in the datapath, uninstalls it. */
1841 facet_uninstall(struct ofproto *p, struct facet *facet)
1843 if (facet->installed) {
1844 struct odputil_keybuf keybuf;
1845 struct dpif_flow_stats stats;
1848 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1849 odp_flow_key_from_flow(&key, &facet->flow);
1851 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1852 facet_update_stats(p, facet, &stats);
1854 facet->installed = false;
1855 facet->dp_packet_count = 0;
1856 facet->dp_byte_count = 0;
1858 assert(facet->dp_packet_count == 0);
1859 assert(facet->dp_byte_count == 0);
1863 /* Returns true if the only action for 'facet' is to send to the controller.
1864 * (We don't report NetFlow expiration messages for such facets because they
1865 * are just part of the control logic for the network, not real traffic). */
1867 facet_is_controller_flow(struct facet *facet)
1870 && facet->rule->n_actions == 1
1871 && action_outputs_to_port(&facet->rule->actions[0],
1872 htons(OFPP_CONTROLLER)));
1875 /* Folds all of 'facet''s statistics into its rule. Also updates the
1876 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1877 * 'facet''s statistics in the datapath should have been zeroed and folded into
1878 * its packet and byte counts before this function is called. */
1880 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1882 assert(!facet->dp_byte_count);
1883 assert(!facet->dp_packet_count);
1885 facet_push_stats(ofproto, facet);
1886 facet_account(ofproto, facet, 0);
1888 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1889 struct ofexpired expired;
1890 expired.flow = facet->flow;
1891 expired.packet_count = facet->packet_count;
1892 expired.byte_count = facet->byte_count;
1893 expired.used = facet->used;
1894 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1897 facet->rule->packet_count += facet->packet_count;
1898 facet->rule->byte_count += facet->byte_count;
1900 /* Reset counters to prevent double counting if 'facet' ever gets
1902 facet->packet_count = 0;
1903 facet->byte_count = 0;
1904 facet->rs_packet_count = 0;
1905 facet->rs_byte_count = 0;
1906 facet->accounted_bytes = 0;
1908 netflow_flow_clear(&facet->nf_flow);
1911 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1912 * Returns it if found, otherwise a null pointer.
1914 * The returned facet might need revalidation; use facet_lookup_valid()
1915 * instead if that is important. */
1916 static struct facet *
1917 facet_find(struct ofproto *ofproto, const struct flow *flow)
1919 struct facet *facet;
1921 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1923 if (flow_equal(flow, &facet->flow)) {
1931 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1932 * Returns it if found, otherwise a null pointer.
1934 * The returned facet is guaranteed to be valid. */
1935 static struct facet *
1936 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1938 struct facet *facet = facet_find(ofproto, flow);
1940 /* The facet we found might not be valid, since we could be in need of
1941 * revalidation. If it is not valid, don't return it. */
1943 && ofproto->need_revalidate
1944 && !facet_revalidate(ofproto, facet)) {
1945 COVERAGE_INC(ofproto_invalidated);
1952 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1954 * - If the rule found is different from 'facet''s current rule, moves
1955 * 'facet' to the new rule and recompiles its actions.
1957 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1958 * where it is and recompiles its actions anyway.
1960 * - If there is none, destroys 'facet'.
1962 * Returns true if 'facet' still exists, false if it has been destroyed. */
1964 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1966 struct action_xlate_ctx ctx;
1967 struct ofpbuf *odp_actions;
1968 struct rule *new_rule;
1969 bool actions_changed;
1971 COVERAGE_INC(facet_revalidate);
1973 /* Determine the new rule. */
1974 new_rule = rule_lookup(ofproto, &facet->flow);
1976 /* No new rule, so delete the facet. */
1977 facet_remove(ofproto, facet);
1981 /* Calculate new ODP actions.
1983 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1984 * emit a NetFlow expiration and, if so, we need to have the old state
1985 * around to properly compose it. */
1986 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1987 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1988 actions_changed = (facet->actions_len != odp_actions->size
1989 || memcmp(facet->actions, odp_actions->data,
1990 facet->actions_len));
1992 /* If the ODP actions changed or the installability changed, then we need
1993 * to talk to the datapath. */
1994 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
1995 if (ctx.may_set_up_flow) {
1996 struct dpif_flow_stats stats;
1998 facet_put__(ofproto, facet,
1999 odp_actions->data, odp_actions->size, &stats);
2000 facet_update_stats(ofproto, facet, &stats);
2002 facet_uninstall(ofproto, facet);
2005 /* The datapath flow is gone or has zeroed stats, so push stats out of
2006 * 'facet' into 'rule'. */
2007 facet_flush_stats(ofproto, facet);
2010 /* Update 'facet' now that we've taken care of all the old state. */
2011 facet->tags = ctx.tags;
2012 facet->nf_flow.output_iface = ctx.nf_output_iface;
2013 facet->may_install = ctx.may_set_up_flow;
2014 if (actions_changed) {
2015 free(facet->actions);
2016 facet->actions_len = odp_actions->size;
2017 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2019 if (facet->rule != new_rule) {
2020 COVERAGE_INC(facet_changed_rule);
2021 list_remove(&facet->list_node);
2022 list_push_back(&new_rule->facets, &facet->list_node);
2023 facet->rule = new_rule;
2024 facet->used = new_rule->created;
2025 facet->rs_used = facet->used;
2028 ofpbuf_delete(odp_actions);
2034 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2037 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2039 COVERAGE_INC(ofproto_error);
2040 ofconn_send_reply(ofconn, buf);
2045 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2047 ofconn_send_reply(ofconn, make_echo_reply(oh));
2052 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2054 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2055 struct ofp_switch_features *osf;
2057 struct ofport *port;
2059 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2060 osf->datapath_id = htonll(ofproto->datapath_id);
2061 osf->n_buffers = htonl(pktbuf_capacity());
2063 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2064 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2065 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2066 (1u << OFPAT_SET_VLAN_VID) |
2067 (1u << OFPAT_SET_VLAN_PCP) |
2068 (1u << OFPAT_STRIP_VLAN) |
2069 (1u << OFPAT_SET_DL_SRC) |
2070 (1u << OFPAT_SET_DL_DST) |
2071 (1u << OFPAT_SET_NW_SRC) |
2072 (1u << OFPAT_SET_NW_DST) |
2073 (1u << OFPAT_SET_NW_TOS) |
2074 (1u << OFPAT_SET_TP_SRC) |
2075 (1u << OFPAT_SET_TP_DST) |
2076 (1u << OFPAT_ENQUEUE));
2078 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2079 ofpbuf_put(buf, &port->opp, sizeof port->opp);
2082 ofconn_send_reply(ofconn, buf);
2087 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2089 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2091 struct ofp_switch_config *osc;
2095 /* Figure out flags. */
2096 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2097 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2100 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2101 osc->flags = htons(flags);
2102 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2103 ofconn_send_reply(ofconn, buf);
2109 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2111 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2112 uint16_t flags = ntohs(osc->flags);
2114 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2115 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2116 switch (flags & OFPC_FRAG_MASK) {
2117 case OFPC_FRAG_NORMAL:
2118 dpif_set_drop_frags(ofproto->dpif, false);
2120 case OFPC_FRAG_DROP:
2121 dpif_set_drop_frags(ofproto->dpif, true);
2124 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2130 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2135 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2136 struct action_xlate_ctx *ctx);
2139 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2141 const struct ofport *ofport = get_port(ctx->ofproto, port);
2144 if (ofport->opp.config & htonl(OFPPC_NO_FWD)) {
2145 /* Forwarding disabled on port. */
2150 * We don't have an ofport record for this port, but it doesn't hurt to
2151 * allow forwarding to it anyhow. Maybe such a port will appear later
2152 * and we're pre-populating the flow table.
2156 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2157 ctx->nf_output_iface = port;
2160 static struct rule *
2161 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2163 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2167 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2169 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2170 uint16_t old_in_port;
2173 /* Look up a flow with 'in_port' as the input port. Then restore the
2174 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2175 * have surprising behavior). */
2176 old_in_port = ctx->flow.in_port;
2177 ctx->flow.in_port = in_port;
2178 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2179 ctx->flow.in_port = old_in_port;
2181 if (ctx->resubmit_hook) {
2182 ctx->resubmit_hook(ctx, rule);
2187 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2191 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2193 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2194 MAX_RESUBMIT_RECURSION);
2199 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, ovs_be32 mask,
2200 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2202 struct ofport *ofport;
2204 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2205 uint16_t odp_port = ofport->odp_port;
2206 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2207 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2210 *nf_output_iface = NF_OUT_FLOOD;
2214 xlate_output_action__(struct action_xlate_ctx *ctx,
2215 uint16_t port, uint16_t max_len)
2218 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2220 ctx->nf_output_iface = NF_OUT_DROP;
2224 add_output_action(ctx, ctx->flow.in_port);
2227 xlate_table_action(ctx, ctx->flow.in_port);
2230 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2231 ctx->odp_actions, &ctx->tags,
2232 &ctx->nf_output_iface,
2233 ctx->ofproto->aux)) {
2234 COVERAGE_INC(ofproto_uninstallable);
2235 ctx->may_set_up_flow = false;
2239 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
2240 &ctx->nf_output_iface, ctx->odp_actions);
2243 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
2244 &ctx->nf_output_iface, ctx->odp_actions);
2246 case OFPP_CONTROLLER:
2247 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2250 add_output_action(ctx, ODPP_LOCAL);
2253 odp_port = ofp_port_to_odp_port(port);
2254 if (odp_port != ctx->flow.in_port) {
2255 add_output_action(ctx, odp_port);
2260 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2261 ctx->nf_output_iface = NF_OUT_FLOOD;
2262 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2263 ctx->nf_output_iface = prev_nf_output_iface;
2264 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2265 ctx->nf_output_iface != NF_OUT_FLOOD) {
2266 ctx->nf_output_iface = NF_OUT_MULTI;
2271 xlate_output_action(struct action_xlate_ctx *ctx,
2272 const struct ofp_action_output *oao)
2274 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2277 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2278 * optimization, because we're going to add another action that sets the
2279 * priority immediately after, or because there are no actions following the
2282 remove_pop_action(struct action_xlate_ctx *ctx)
2284 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2285 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2286 ctx->last_pop_priority = -1;
2291 add_pop_action(struct action_xlate_ctx *ctx)
2293 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2294 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2295 ctx->last_pop_priority = ctx->odp_actions->size;
2300 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2301 const struct ofp_action_enqueue *oae)
2303 uint16_t ofp_port, odp_port;
2307 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2310 /* Fall back to ordinary output action. */
2311 xlate_output_action__(ctx, ntohs(oae->port), 0);
2315 /* Figure out ODP output port. */
2316 ofp_port = ntohs(oae->port);
2317 if (ofp_port != OFPP_IN_PORT) {
2318 odp_port = ofp_port_to_odp_port(ofp_port);
2320 odp_port = ctx->flow.in_port;
2323 /* Add ODP actions. */
2324 remove_pop_action(ctx);
2325 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2326 add_output_action(ctx, odp_port);
2327 add_pop_action(ctx);
2329 /* Update NetFlow output port. */
2330 if (ctx->nf_output_iface == NF_OUT_DROP) {
2331 ctx->nf_output_iface = odp_port;
2332 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2333 ctx->nf_output_iface = NF_OUT_MULTI;
2338 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2339 const struct nx_action_set_queue *nasq)
2344 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2347 /* Couldn't translate queue to a priority, so ignore. A warning
2348 * has already been logged. */
2352 remove_pop_action(ctx);
2353 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2357 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2359 ovs_be16 tci = ctx->flow.vlan_tci;
2360 if (!(tci & htons(VLAN_CFI))) {
2361 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2363 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2364 tci & ~htons(VLAN_CFI));
2368 struct xlate_reg_state {
2374 save_reg_state(const struct action_xlate_ctx *ctx,
2375 struct xlate_reg_state *state)
2377 state->vlan_tci = ctx->flow.vlan_tci;
2378 state->tun_id = ctx->flow.tun_id;
2382 update_reg_state(struct action_xlate_ctx *ctx,
2383 const struct xlate_reg_state *state)
2385 if (ctx->flow.vlan_tci != state->vlan_tci) {
2386 xlate_set_dl_tci(ctx);
2388 if (ctx->flow.tun_id != state->tun_id) {
2389 nl_msg_put_be64(ctx->odp_actions,
2390 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2395 xlate_nicira_action(struct action_xlate_ctx *ctx,
2396 const struct nx_action_header *nah)
2398 const struct nx_action_resubmit *nar;
2399 const struct nx_action_set_tunnel *nast;
2400 const struct nx_action_set_queue *nasq;
2401 const struct nx_action_multipath *nam;
2402 const struct nx_action_autopath *naa;
2403 enum nx_action_subtype subtype = ntohs(nah->subtype);
2404 const struct ofhooks *ofhooks = ctx->ofproto->ofhooks;
2405 struct xlate_reg_state state;
2406 uint16_t autopath_port;
2409 assert(nah->vendor == htonl(NX_VENDOR_ID));
2411 case NXAST_RESUBMIT:
2412 nar = (const struct nx_action_resubmit *) nah;
2413 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2416 case NXAST_SET_TUNNEL:
2417 nast = (const struct nx_action_set_tunnel *) nah;
2418 tun_id = htonll(ntohl(nast->tun_id));
2419 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2420 ctx->flow.tun_id = tun_id;
2423 case NXAST_DROP_SPOOFED_ARP:
2424 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2425 nl_msg_put_flag(ctx->odp_actions,
2426 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2430 case NXAST_SET_QUEUE:
2431 nasq = (const struct nx_action_set_queue *) nah;
2432 xlate_set_queue_action(ctx, nasq);
2435 case NXAST_POP_QUEUE:
2436 add_pop_action(ctx);
2439 case NXAST_REG_MOVE:
2440 save_reg_state(ctx, &state);
2441 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2443 update_reg_state(ctx, &state);
2446 case NXAST_REG_LOAD:
2447 save_reg_state(ctx, &state);
2448 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2450 update_reg_state(ctx, &state);
2454 /* Nothing to do. */
2457 case NXAST_SET_TUNNEL64:
2458 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2459 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2460 ctx->flow.tun_id = tun_id;
2463 case NXAST_MULTIPATH:
2464 nam = (const struct nx_action_multipath *) nah;
2465 multipath_execute(nam, &ctx->flow);
2468 case NXAST_AUTOPATH:
2469 naa = (const struct nx_action_autopath *) nah;
2470 autopath_port = (ofhooks->autopath_cb
2471 ? ofhooks->autopath_cb(&ctx->flow, ntohl(naa->id),
2472 &ctx->tags, ctx->ofproto->aux)
2474 autopath_execute(naa, &ctx->flow, autopath_port);
2477 /* If you add a new action here that modifies flow data, don't forget to
2478 * update the flow key in ctx->flow at the same time. */
2480 case NXAST_SNAT__OBSOLETE:
2482 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2488 do_xlate_actions(const union ofp_action *in, size_t n_in,
2489 struct action_xlate_ctx *ctx)
2491 struct actions_iterator iter;
2492 const union ofp_action *ia;
2493 const struct ofport *port;
2495 port = get_port(ctx->ofproto, ctx->flow.in_port);
2496 if (port && port->opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2497 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2498 ? htonl(OFPPC_NO_RECV_STP)
2499 : htonl(OFPPC_NO_RECV))) {
2500 /* Drop this flow. */
2504 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2505 enum ofp_action_type type = ntohs(ia->type);
2506 const struct ofp_action_dl_addr *oada;
2510 xlate_output_action(ctx, &ia->output);
2513 case OFPAT_SET_VLAN_VID:
2514 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2515 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2516 xlate_set_dl_tci(ctx);
2519 case OFPAT_SET_VLAN_PCP:
2520 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2521 ctx->flow.vlan_tci |= htons(
2522 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2523 xlate_set_dl_tci(ctx);
2526 case OFPAT_STRIP_VLAN:
2527 ctx->flow.vlan_tci = htons(0);
2528 xlate_set_dl_tci(ctx);
2531 case OFPAT_SET_DL_SRC:
2532 oada = ((struct ofp_action_dl_addr *) ia);
2533 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2534 oada->dl_addr, ETH_ADDR_LEN);
2535 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2538 case OFPAT_SET_DL_DST:
2539 oada = ((struct ofp_action_dl_addr *) ia);
2540 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2541 oada->dl_addr, ETH_ADDR_LEN);
2542 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2545 case OFPAT_SET_NW_SRC:
2546 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2547 ia->nw_addr.nw_addr);
2548 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2551 case OFPAT_SET_NW_DST:
2552 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2553 ia->nw_addr.nw_addr);
2554 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2557 case OFPAT_SET_NW_TOS:
2558 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2560 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2563 case OFPAT_SET_TP_SRC:
2564 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2565 ia->tp_port.tp_port);
2566 ctx->flow.tp_src = ia->tp_port.tp_port;
2569 case OFPAT_SET_TP_DST:
2570 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2571 ia->tp_port.tp_port);
2572 ctx->flow.tp_dst = ia->tp_port.tp_port;
2576 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2580 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2584 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2591 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2592 struct ofproto *ofproto, const struct flow *flow,
2593 const struct ofpbuf *packet)
2595 ctx->ofproto = ofproto;
2597 ctx->packet = packet;
2598 ctx->resubmit_hook = NULL;
2599 ctx->check_special = true;
2603 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2604 const struct ofpbuf *packet)
2606 struct ofport *ofport;
2608 ofport = get_port(ofproto, flow->in_port);
2609 if (ofport && ofport->cfm) {
2610 cfm_process_heartbeat(ofport->cfm, packet);
2614 static struct ofpbuf *
2615 xlate_actions(struct action_xlate_ctx *ctx,
2616 const union ofp_action *in, size_t n_in)
2618 COVERAGE_INC(ofproto_ofp2odp);
2620 ctx->odp_actions = ofpbuf_new(512);
2622 ctx->may_set_up_flow = true;
2623 ctx->nf_output_iface = NF_OUT_DROP;
2625 ctx->last_pop_priority = -1;
2627 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2629 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2631 ctx->may_set_up_flow = false;
2632 } else if (ctx->check_special
2633 && ctx->ofproto->ofhooks->special_cb
2634 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2635 ctx->ofproto->aux)) {
2636 ctx->may_set_up_flow = false;
2638 do_xlate_actions(in, n_in, ctx);
2641 remove_pop_action(ctx);
2643 /* Check with in-band control to see if we're allowed to set up this
2645 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2646 ctx->odp_actions->data,
2647 ctx->odp_actions->size)) {
2648 ctx->may_set_up_flow = false;
2651 return ctx->odp_actions;
2654 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2655 * error message code (composed with ofp_mkerr()) for the caller to propagate
2656 * upward. Otherwise, returns 0.
2658 * The log message mentions 'msg_type'. */
2660 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2662 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2663 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2664 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2665 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2668 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2675 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2677 struct ofproto *p = ofconn_get_ofproto(ofconn);
2678 struct ofp_packet_out *opo;
2679 struct ofpbuf payload, *buffer;
2680 union ofp_action *ofp_actions;
2681 struct action_xlate_ctx ctx;
2682 struct ofpbuf *odp_actions;
2683 struct ofpbuf request;
2685 size_t n_ofp_actions;
2689 COVERAGE_INC(ofproto_packet_out);
2691 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2696 /* Get ofp_packet_out. */
2697 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2698 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2701 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2702 &ofp_actions, &n_ofp_actions);
2708 if (opo->buffer_id != htonl(UINT32_MAX)) {
2709 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2711 if (error || !buffer) {
2720 /* Extract flow, check actions. */
2721 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2723 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2729 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2730 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2731 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2732 ofpbuf_delete(odp_actions);
2735 ofpbuf_delete(buffer);
2740 update_port_config(struct ofproto *p, struct ofport *port,
2741 ovs_be32 config, ovs_be32 mask)
2743 mask &= config ^ port->opp.config;
2744 if (mask & htonl(OFPPC_PORT_DOWN)) {
2745 if (config & htonl(OFPPC_PORT_DOWN)) {
2746 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2748 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2751 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2752 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2753 if (mask & htonl(REVALIDATE_BITS)) {
2754 COVERAGE_INC(ofproto_costly_flags);
2755 port->opp.config ^= mask & htonl(REVALIDATE_BITS);
2756 p->need_revalidate = true;
2758 #undef REVALIDATE_BITS
2759 if (mask & htonl(OFPPC_NO_PACKET_IN)) {
2760 port->opp.config ^= htonl(OFPPC_NO_PACKET_IN);
2765 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2767 struct ofproto *p = ofconn_get_ofproto(ofconn);
2768 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2769 struct ofport *port;
2772 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2777 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2779 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2780 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2781 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2783 update_port_config(p, port, opm->config, opm->mask);
2784 if (opm->advertise) {
2785 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2791 static struct ofpbuf *
2792 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2794 struct ofp_stats_reply *osr;
2797 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2798 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2800 osr->flags = htons(0);
2804 static struct ofpbuf *
2805 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2807 const struct ofp_stats_request *osr
2808 = (const struct ofp_stats_request *) request;
2809 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2813 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2814 struct ofpbuf **msgp)
2816 struct ofpbuf *msg = *msgp;
2817 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2818 if (nbytes + msg->size > UINT16_MAX) {
2819 struct ofp_stats_reply *reply = msg->data;
2820 reply->flags = htons(OFPSF_REPLY_MORE);
2821 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2822 ofconn_send_reply(ofconn, msg);
2824 return ofpbuf_put_uninit(*msgp, nbytes);
2827 static struct ofpbuf *
2828 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2830 struct nicira_stats_msg *nsm;
2833 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2834 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2835 nsm->type = htons(OFPST_VENDOR);
2836 nsm->flags = htons(0);
2837 nsm->vendor = htonl(NX_VENDOR_ID);
2838 nsm->subtype = subtype;
2842 static struct ofpbuf *
2843 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2845 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2849 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2850 struct ofpbuf **msgp)
2852 struct ofpbuf *msg = *msgp;
2853 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2854 if (nbytes + msg->size > UINT16_MAX) {
2855 struct nicira_stats_msg *reply = msg->data;
2856 reply->flags = htons(OFPSF_REPLY_MORE);
2857 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2858 ofconn_send_reply(ofconn, msg);
2860 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2864 handle_desc_stats_request(struct ofconn *ofconn,
2865 const struct ofp_header *request)
2867 struct ofproto *p = ofconn_get_ofproto(ofconn);
2868 struct ofp_desc_stats *ods;
2871 msg = start_ofp_stats_reply(request, sizeof *ods);
2872 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2873 memset(ods, 0, sizeof *ods);
2874 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2875 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2876 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2877 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2878 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2879 ofconn_send_reply(ofconn, msg);
2885 handle_table_stats_request(struct ofconn *ofconn,
2886 const struct ofp_header *request)
2888 struct ofproto *p = ofconn_get_ofproto(ofconn);
2889 struct ofp_table_stats *ots;
2892 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2894 /* Classifier table. */
2895 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2896 memset(ots, 0, sizeof *ots);
2897 strcpy(ots->name, "classifier");
2898 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2899 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2900 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2901 ots->active_count = htonl(classifier_count(&p->cls));
2902 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2903 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2905 ofconn_send_reply(ofconn, msg);
2910 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2911 struct ofpbuf **msgp)
2913 struct netdev_stats stats;
2914 struct ofp_port_stats *ops;
2916 /* Intentionally ignore return value, since errors will set
2917 * 'stats' to all-1s, which is correct for OpenFlow, and
2918 * netdev_get_stats() will log errors. */
2919 netdev_get_stats(port->netdev, &stats);
2921 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2922 ops->port_no = port->opp.port_no;
2923 memset(ops->pad, 0, sizeof ops->pad);
2924 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2925 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2926 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2927 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2928 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2929 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2930 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2931 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2932 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2933 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2934 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2935 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2939 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2941 struct ofproto *p = ofconn_get_ofproto(ofconn);
2942 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2943 struct ofp_port_stats *ops;
2945 struct ofport *port;
2947 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2948 if (psr->port_no != htons(OFPP_NONE)) {
2949 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2951 append_port_stat(port, ofconn, &msg);
2954 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2955 append_port_stat(port, ofconn, &msg);
2959 ofconn_send_reply(ofconn, msg);
2964 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2966 long long int msecs = time_msec() - start;
2967 *sec = msecs / 1000;
2968 *nsec = (msecs % 1000) * (1000 * 1000);
2972 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2976 calc_flow_duration__(start, &sec, &nsec);
2977 *sec_be = htonl(sec);
2978 *nsec_be = htonl(nsec);
2982 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2983 ovs_be16 out_port, struct ofpbuf **replyp)
2985 struct ofp_flow_stats *ofs;
2986 uint64_t packet_count, byte_count;
2988 size_t act_len, len;
2990 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2994 act_len = sizeof *rule->actions * rule->n_actions;
2995 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2997 rule_get_stats(rule, &packet_count, &byte_count);
2999 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3000 ofs->length = htons(len);
3003 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3004 &ofs->match, rule->flow_cookie, &cookie);
3005 put_32aligned_be64(&ofs->cookie, cookie);
3006 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3007 ofs->priority = htons(rule->cr.priority);
3008 ofs->idle_timeout = htons(rule->idle_timeout);
3009 ofs->hard_timeout = htons(rule->hard_timeout);
3010 memset(ofs->pad2, 0, sizeof ofs->pad2);
3011 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3012 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3013 if (rule->n_actions > 0) {
3014 memcpy(ofs->actions, rule->actions, act_len);
3019 is_valid_table(uint8_t table_id)
3021 if (table_id == 0 || table_id == 0xff) {
3024 /* It would probably be better to reply with an error but there doesn't
3025 * seem to be any appropriate value, so that might just be
3027 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3034 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3036 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3037 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3038 struct ofpbuf *reply;
3040 COVERAGE_INC(ofproto_flows_req);
3041 reply = start_ofp_stats_reply(oh, 1024);
3042 if (is_valid_table(fsr->table_id)) {
3043 struct cls_cursor cursor;
3044 struct cls_rule target;
3047 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3049 cls_cursor_init(&cursor, &ofproto->cls, &target);
3050 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3051 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3054 ofconn_send_reply(ofconn, reply);
3060 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3061 ovs_be16 out_port, struct ofpbuf **replyp)
3063 struct nx_flow_stats *nfs;
3064 uint64_t packet_count, byte_count;
3065 size_t act_len, start_len;
3066 struct ofpbuf *reply;
3068 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3072 rule_get_stats(rule, &packet_count, &byte_count);
3074 act_len = sizeof *rule->actions * rule->n_actions;
3076 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3077 start_len = (*replyp)->size;
3080 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3083 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3084 nfs->cookie = rule->flow_cookie;
3085 nfs->priority = htons(rule->cr.priority);
3086 nfs->idle_timeout = htons(rule->idle_timeout);
3087 nfs->hard_timeout = htons(rule->hard_timeout);
3088 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3089 memset(nfs->pad2, 0, sizeof nfs->pad2);
3090 nfs->packet_count = htonll(packet_count);
3091 nfs->byte_count = htonll(byte_count);
3092 if (rule->n_actions > 0) {
3093 ofpbuf_put(reply, rule->actions, act_len);
3095 nfs->length = htons(reply->size - start_len);
3099 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3101 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3102 struct nx_flow_stats_request *nfsr;
3103 struct cls_rule target;
3104 struct ofpbuf *reply;
3108 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3110 /* Dissect the message. */
3111 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3112 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3117 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3120 COVERAGE_INC(ofproto_flows_req);
3121 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3122 if (is_valid_table(nfsr->table_id)) {
3123 struct cls_cursor cursor;
3126 cls_cursor_init(&cursor, &ofproto->cls, &target);
3127 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3128 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3131 ofconn_send_reply(ofconn, reply);
3137 flow_stats_ds(struct rule *rule, struct ds *results)
3139 uint64_t packet_count, byte_count;
3140 size_t act_len = sizeof *rule->actions * rule->n_actions;
3142 rule_get_stats(rule, &packet_count, &byte_count);
3144 ds_put_format(results, "duration=%llds, ",
3145 (time_msec() - rule->created) / 1000);
3146 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3147 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3148 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3149 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3150 cls_rule_format(&rule->cr, results);
3151 ds_put_char(results, ',');
3153 ofp_print_actions(results, &rule->actions->header, act_len);
3155 ds_put_cstr(results, "drop");
3157 ds_put_cstr(results, "\n");
3160 /* Adds a pretty-printed description of all flows to 'results', including
3161 * hidden flows (e.g., set up by in-band control). */
3163 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3165 struct cls_cursor cursor;
3168 cls_cursor_init(&cursor, &p->cls, NULL);
3169 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3170 flow_stats_ds(rule, results);
3174 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3175 * '*engine_type' and '*engine_id', respectively. */
3177 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3178 uint8_t *engine_type, uint8_t *engine_id)
3180 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3184 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3185 ovs_be16 out_port, uint8_t table_id,
3186 struct ofp_aggregate_stats_reply *oasr)
3188 uint64_t total_packets = 0;
3189 uint64_t total_bytes = 0;
3192 COVERAGE_INC(ofproto_agg_request);
3194 if (is_valid_table(table_id)) {
3195 struct cls_cursor cursor;
3198 cls_cursor_init(&cursor, &ofproto->cls, target);
3199 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3200 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3201 uint64_t packet_count;
3202 uint64_t byte_count;
3204 rule_get_stats(rule, &packet_count, &byte_count);
3206 total_packets += packet_count;
3207 total_bytes += byte_count;
3213 oasr->flow_count = htonl(n_flows);
3214 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3215 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3216 memset(oasr->pad, 0, sizeof oasr->pad);
3220 handle_aggregate_stats_request(struct ofconn *ofconn,
3221 const struct ofp_header *oh)
3223 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3224 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3225 struct ofp_aggregate_stats_reply *reply;
3226 struct cls_rule target;
3229 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3232 msg = start_ofp_stats_reply(oh, sizeof *reply);
3233 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3234 query_aggregate_stats(ofproto, &target, request->out_port,
3235 request->table_id, reply);
3236 ofconn_send_reply(ofconn, msg);
3241 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3243 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3244 struct nx_aggregate_stats_request *request;
3245 struct ofp_aggregate_stats_reply *reply;
3246 struct cls_rule target;
3251 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3253 /* Dissect the message. */
3254 request = ofpbuf_pull(&b, sizeof *request);
3255 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3260 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3264 COVERAGE_INC(ofproto_flows_req);
3265 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3266 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3267 query_aggregate_stats(ofproto, &target, request->out_port,
3268 request->table_id, reply);
3269 ofconn_send_reply(ofconn, buf);
3274 struct queue_stats_cbdata {
3275 struct ofconn *ofconn;
3276 struct ofport *ofport;
3281 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3282 const struct netdev_queue_stats *stats)
3284 struct ofp_queue_stats *reply;
3286 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3287 reply->port_no = cbdata->ofport->opp.port_no;
3288 memset(reply->pad, 0, sizeof reply->pad);
3289 reply->queue_id = htonl(queue_id);
3290 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3291 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3292 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3296 handle_queue_stats_dump_cb(uint32_t queue_id,
3297 struct netdev_queue_stats *stats,
3300 struct queue_stats_cbdata *cbdata = cbdata_;
3302 put_queue_stats(cbdata, queue_id, stats);
3306 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3307 struct queue_stats_cbdata *cbdata)
3309 cbdata->ofport = port;
3310 if (queue_id == OFPQ_ALL) {
3311 netdev_dump_queue_stats(port->netdev,
3312 handle_queue_stats_dump_cb, cbdata);
3314 struct netdev_queue_stats stats;
3316 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3317 put_queue_stats(cbdata, queue_id, &stats);
3323 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3325 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3326 const struct ofp_queue_stats_request *qsr;
3327 struct queue_stats_cbdata cbdata;
3328 struct ofport *port;
3329 unsigned int port_no;
3332 qsr = ofputil_stats_body(oh);
3334 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3337 COVERAGE_INC(ofproto_queue_req);
3339 cbdata.ofconn = ofconn;
3340 cbdata.msg = start_ofp_stats_reply(oh, 128);
3342 port_no = ntohs(qsr->port_no);
3343 queue_id = ntohl(qsr->queue_id);
3344 if (port_no == OFPP_ALL) {
3345 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3346 handle_queue_stats_for_port(port, queue_id, &cbdata);
3348 } else if (port_no < ofproto->max_ports) {
3349 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3351 handle_queue_stats_for_port(port, queue_id, &cbdata);
3354 ofpbuf_delete(cbdata.msg);
3355 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3357 ofconn_send_reply(ofconn, cbdata.msg);
3362 /* Updates 'facet''s used time. Caller is responsible for calling
3363 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3365 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3368 if (used > facet->used) {
3370 if (used > facet->rule->used) {
3371 facet->rule->used = used;
3373 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3377 /* Folds the statistics from 'stats' into the counters in 'facet'.
3379 * Because of the meaning of a facet's counters, it only makes sense to do this
3380 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3381 * packet that was sent by hand or if it represents statistics that have been
3382 * cleared out of the datapath. */
3384 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3385 const struct dpif_flow_stats *stats)
3387 if (stats->n_packets || stats->used > facet->used) {
3388 facet_update_time(ofproto, facet, stats->used);
3389 facet->packet_count += stats->n_packets;
3390 facet->byte_count += stats->n_bytes;
3391 facet_push_stats(ofproto, facet);
3392 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3397 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3399 uint64_t rs_packets, rs_bytes;
3401 assert(facet->packet_count >= facet->rs_packet_count);
3402 assert(facet->byte_count >= facet->rs_byte_count);
3403 assert(facet->used >= facet->rs_used);
3405 rs_packets = facet->packet_count - facet->rs_packet_count;
3406 rs_bytes = facet->byte_count - facet->rs_byte_count;
3408 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3409 facet->rs_packet_count = facet->packet_count;
3410 facet->rs_byte_count = facet->byte_count;
3411 facet->rs_used = facet->used;
3413 flow_push_stats(ofproto, facet->rule, &facet->flow,
3414 rs_packets, rs_bytes, facet->used);
3418 struct ofproto_push {
3419 struct action_xlate_ctx ctx;
3426 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3428 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3431 rule->packet_count += push->packets;
3432 rule->byte_count += push->bytes;
3433 rule->used = MAX(push->used, rule->used);
3437 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3438 * 'rule''s actions. */
3440 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3441 struct flow *flow, uint64_t packets, uint64_t bytes,
3444 struct ofproto_push push;
3446 push.packets = packets;
3450 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3451 push.ctx.resubmit_hook = push_resubmit;
3452 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3455 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3456 * in which no matching flow already exists in the flow table.
3458 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3459 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3460 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3462 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3465 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3467 struct ofproto *p = ofconn_get_ofproto(ofconn);
3468 struct ofpbuf *packet;
3473 if (fm->flags & OFPFF_CHECK_OVERLAP
3474 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3475 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3479 if (fm->buffer_id != UINT32_MAX) {
3480 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3484 in_port = UINT16_MAX;
3487 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3488 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3489 fm->flags & OFPFF_SEND_FLOW_REM);
3490 rule_insert(p, rule);
3492 rule_execute(p, rule, in_port, packet);
3497 static struct rule *
3498 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3500 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3504 send_buffered_packet(struct ofconn *ofconn,
3505 struct rule *rule, uint32_t buffer_id)
3507 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3508 struct ofpbuf *packet;
3512 if (buffer_id == UINT32_MAX) {
3516 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3521 rule_execute(ofproto, rule, in_port, packet);
3526 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3528 struct modify_flows_cbdata {
3529 struct ofproto *ofproto;
3530 const struct flow_mod *fm;
3534 static int modify_flow(struct ofproto *, const struct flow_mod *,
3537 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3538 * encoded by ofp_mkerr() on failure.
3540 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3543 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3545 struct ofproto *p = ofconn_get_ofproto(ofconn);
3546 struct rule *match = NULL;
3547 struct cls_cursor cursor;
3550 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3551 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3552 if (!rule_is_hidden(rule)) {
3554 modify_flow(p, fm, rule);
3559 /* This credits the packet to whichever flow happened to match last.
3560 * That's weird. Maybe we should do a lookup for the flow that
3561 * actually matches the packet? Who knows. */
3562 send_buffered_packet(ofconn, match, fm->buffer_id);
3565 return add_flow(ofconn, fm);
3569 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3570 * code as encoded by ofp_mkerr() on failure.
3572 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3575 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3577 struct ofproto *p = ofconn_get_ofproto(ofconn);
3578 struct rule *rule = find_flow_strict(p, fm);
3579 if (rule && !rule_is_hidden(rule)) {
3580 modify_flow(p, fm, rule);
3581 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3583 return add_flow(ofconn, fm);
3587 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3588 * been identified as a flow in 'p''s flow table to be modified, by changing
3589 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3590 * ofp_action[] structures). */
3592 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3594 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3596 rule->flow_cookie = fm->cookie;
3598 /* If the actions are the same, do nothing. */
3599 if (fm->n_actions == rule->n_actions
3601 || !memcmp(fm->actions, rule->actions, actions_len))) {
3605 /* Replace actions. */
3606 free(rule->actions);
3607 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3608 rule->n_actions = fm->n_actions;
3610 p->need_revalidate = true;
3615 /* OFPFC_DELETE implementation. */
3617 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3619 /* Implements OFPFC_DELETE. */
3621 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3623 struct rule *rule, *next_rule;
3624 struct cls_cursor cursor;
3626 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3627 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3628 delete_flow(p, rule, htons(fm->out_port));
3632 /* Implements OFPFC_DELETE_STRICT. */
3634 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3636 struct rule *rule = find_flow_strict(p, fm);
3638 delete_flow(p, rule, htons(fm->out_port));
3642 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3643 * been identified as a flow to delete from 'p''s flow table, by deleting the
3644 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3647 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3648 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3649 * specified 'out_port'. */
3651 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3653 if (rule_is_hidden(rule)) {
3657 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3661 rule_send_removed(p, rule, OFPRR_DELETE);
3662 rule_remove(p, rule);
3666 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3668 struct ofproto *p = ofconn_get_ofproto(ofconn);
3672 error = reject_slave_controller(ofconn, "flow_mod");
3677 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3682 /* We do not support the emergency flow cache. It will hopefully get
3683 * dropped from OpenFlow in the near future. */
3684 if (fm.flags & OFPFF_EMERG) {
3685 /* There isn't a good fit for an error code, so just state that the
3686 * flow table is full. */
3687 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3690 error = validate_actions(fm.actions, fm.n_actions,
3691 &fm.cr.flow, p->max_ports);
3696 switch (fm.command) {
3698 return add_flow(ofconn, &fm);
3701 return modify_flows_loose(ofconn, &fm);
3703 case OFPFC_MODIFY_STRICT:
3704 return modify_flow_strict(ofconn, &fm);
3707 delete_flows_loose(p, &fm);
3710 case OFPFC_DELETE_STRICT:
3711 delete_flow_strict(p, &fm);
3715 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3720 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3722 const struct nxt_tun_id_cookie *msg
3723 = (const struct nxt_tun_id_cookie *) oh;
3724 enum nx_flow_format flow_format;
3726 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3727 ofconn_set_flow_format(ofconn, flow_format);
3733 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3735 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3736 struct nx_role_request *reply;
3740 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3741 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3742 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3745 role = ntohl(nrr->role);
3746 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3747 && role != NX_ROLE_SLAVE) {
3748 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3750 /* There's no good error code for this. */
3751 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3754 ofconn_set_role(ofconn, role);
3756 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3757 reply->role = htonl(role);
3758 ofconn_send_reply(ofconn, buf);
3764 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3766 const struct nxt_set_flow_format *msg
3767 = (const struct nxt_set_flow_format *) oh;
3770 format = ntohl(msg->format);
3771 if (format == NXFF_OPENFLOW10
3772 || format == NXFF_TUN_ID_FROM_COOKIE
3773 || format == NXFF_NXM) {
3774 ofconn_set_flow_format(ofconn, format);
3777 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3782 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3784 struct ofp_header *ob;
3787 /* Currently, everything executes synchronously, so we can just
3788 * immediately send the barrier reply. */
3789 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3790 ofconn_send_reply(ofconn, buf);
3795 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3797 const struct ofp_header *oh = msg->data;
3798 const struct ofputil_msg_type *type;
3801 error = ofputil_decode_msg_type(oh, &type);
3806 switch (ofputil_msg_type_code(type)) {
3807 /* OpenFlow requests. */
3808 case OFPUTIL_OFPT_ECHO_REQUEST:
3809 return handle_echo_request(ofconn, oh);
3811 case OFPUTIL_OFPT_FEATURES_REQUEST:
3812 return handle_features_request(ofconn, oh);
3814 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3815 return handle_get_config_request(ofconn, oh);
3817 case OFPUTIL_OFPT_SET_CONFIG:
3818 return handle_set_config(ofconn, msg->data);
3820 case OFPUTIL_OFPT_PACKET_OUT:
3821 return handle_packet_out(ofconn, oh);
3823 case OFPUTIL_OFPT_PORT_MOD:
3824 return handle_port_mod(ofconn, oh);
3826 case OFPUTIL_OFPT_FLOW_MOD:
3827 return handle_flow_mod(ofconn, oh);
3829 case OFPUTIL_OFPT_BARRIER_REQUEST:
3830 return handle_barrier_request(ofconn, oh);
3832 /* OpenFlow replies. */
3833 case OFPUTIL_OFPT_ECHO_REPLY:
3836 /* Nicira extension requests. */
3837 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3838 return handle_tun_id_from_cookie(ofconn, oh);
3840 case OFPUTIL_NXT_ROLE_REQUEST:
3841 return handle_role_request(ofconn, oh);
3843 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3844 return handle_nxt_set_flow_format(ofconn, oh);
3846 case OFPUTIL_NXT_FLOW_MOD:
3847 return handle_flow_mod(ofconn, oh);
3849 /* OpenFlow statistics requests. */
3850 case OFPUTIL_OFPST_DESC_REQUEST:
3851 return handle_desc_stats_request(ofconn, oh);
3853 case OFPUTIL_OFPST_FLOW_REQUEST:
3854 return handle_flow_stats_request(ofconn, oh);
3856 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3857 return handle_aggregate_stats_request(ofconn, oh);
3859 case OFPUTIL_OFPST_TABLE_REQUEST:
3860 return handle_table_stats_request(ofconn, oh);
3862 case OFPUTIL_OFPST_PORT_REQUEST:
3863 return handle_port_stats_request(ofconn, oh);
3865 case OFPUTIL_OFPST_QUEUE_REQUEST:
3866 return handle_queue_stats_request(ofconn, oh);
3868 /* Nicira extension statistics requests. */
3869 case OFPUTIL_NXST_FLOW_REQUEST:
3870 return handle_nxst_flow(ofconn, oh);
3872 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3873 return handle_nxst_aggregate(ofconn, oh);
3875 case OFPUTIL_INVALID:
3876 case OFPUTIL_OFPT_HELLO:
3877 case OFPUTIL_OFPT_ERROR:
3878 case OFPUTIL_OFPT_FEATURES_REPLY:
3879 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3880 case OFPUTIL_OFPT_PACKET_IN:
3881 case OFPUTIL_OFPT_FLOW_REMOVED:
3882 case OFPUTIL_OFPT_PORT_STATUS:
3883 case OFPUTIL_OFPT_BARRIER_REPLY:
3884 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3885 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3886 case OFPUTIL_OFPST_DESC_REPLY:
3887 case OFPUTIL_OFPST_FLOW_REPLY:
3888 case OFPUTIL_OFPST_QUEUE_REPLY:
3889 case OFPUTIL_OFPST_PORT_REPLY:
3890 case OFPUTIL_OFPST_TABLE_REPLY:
3891 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3892 case OFPUTIL_NXT_ROLE_REPLY:
3893 case OFPUTIL_NXT_FLOW_REMOVED:
3894 case OFPUTIL_NXST_FLOW_REPLY:
3895 case OFPUTIL_NXST_AGGREGATE_REPLY:
3897 if (VLOG_IS_WARN_ENABLED()) {
3898 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3899 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3902 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3903 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3905 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3911 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3913 int error = handle_openflow__(ofconn, ofp_msg);
3915 send_error_oh(ofconn, ofp_msg->data, error);
3917 COVERAGE_INC(ofproto_recv_openflow);
3921 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3923 struct facet *facet;
3926 /* Obtain in_port and tun_id, at least. */
3927 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3929 /* Set header pointers in 'flow'. */
3930 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3932 if (cfm_should_process_flow(&flow)) {
3933 ofproto_process_cfm(p, &flow, upcall->packet);
3934 ofpbuf_delete(upcall->packet);
3936 } else if (p->ofhooks->special_cb
3937 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3938 ofpbuf_delete(upcall->packet);
3942 /* Check with in-band control to see if this packet should be sent
3943 * to the local port regardless of the flow table. */
3944 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3945 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3948 facet = facet_lookup_valid(p, &flow);
3950 struct rule *rule = rule_lookup(p, &flow);
3952 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3953 struct ofport *port = get_port(p, flow.in_port);
3955 if (port->opp.config & htonl(OFPPC_NO_PACKET_IN)) {
3956 COVERAGE_INC(ofproto_no_packet_in);
3957 /* XXX install 'drop' flow entry */
3958 ofpbuf_delete(upcall->packet);
3962 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3966 COVERAGE_INC(ofproto_packet_in);
3967 send_packet_in(p, upcall, &flow, false);
3971 facet = facet_create(p, rule, &flow, upcall->packet);
3972 } else if (!facet->may_install) {
3973 /* The facet is not installable, that is, we need to process every
3974 * packet, so process the current packet's actions into 'facet'. */
3975 facet_make_actions(p, facet, upcall->packet);
3978 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3980 * Extra-special case for fail-open mode.
3982 * We are in fail-open mode and the packet matched the fail-open rule,
3983 * but we are connected to a controller too. We should send the packet
3984 * up to the controller in the hope that it will try to set up a flow
3985 * and thereby allow us to exit fail-open.
3987 * See the top-level comment in fail-open.c for more information.
3989 send_packet_in(p, upcall, &flow, true);
3992 facet_execute(p, facet, upcall->packet);
3993 facet_install(p, facet, false);
3997 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4001 switch (upcall->type) {
4002 case DPIF_UC_ACTION:
4003 COVERAGE_INC(ofproto_ctlr_action);
4004 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4005 send_packet_in(p, upcall, &flow, false);
4008 case DPIF_UC_SAMPLE:
4010 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4011 ofproto_sflow_received(p->sflow, upcall, &flow);
4013 ofpbuf_delete(upcall->packet);
4017 handle_miss_upcall(p, upcall);
4020 case DPIF_N_UC_TYPES:
4022 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4027 /* Flow expiration. */
4029 static int ofproto_dp_max_idle(const struct ofproto *);
4030 static void ofproto_update_stats(struct ofproto *);
4031 static void rule_expire(struct ofproto *, struct rule *);
4032 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4034 /* This function is called periodically by ofproto_run(). Its job is to
4035 * collect updates for the flows that have been installed into the datapath,
4036 * most importantly when they last were used, and then use that information to
4037 * expire flows that have not been used recently.
4039 * Returns the number of milliseconds after which it should be called again. */
4041 ofproto_expire(struct ofproto *ofproto)
4043 struct rule *rule, *next_rule;
4044 struct cls_cursor cursor;
4047 /* Update stats for each flow in the datapath. */
4048 ofproto_update_stats(ofproto);
4050 /* Expire facets that have been idle too long. */
4051 dp_max_idle = ofproto_dp_max_idle(ofproto);
4052 ofproto_expire_facets(ofproto, dp_max_idle);
4054 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4055 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4056 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4057 rule_expire(ofproto, rule);
4060 /* Let the hook know that we're at a stable point: all outstanding data
4061 * in existing flows has been accounted to the account_cb. Thus, the
4062 * hook can now reasonably do operations that depend on having accurate
4063 * flow volume accounting (currently, that's just bond rebalancing). */
4064 if (ofproto->ofhooks->account_checkpoint_cb) {
4065 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4068 return MIN(dp_max_idle, 1000);
4071 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4073 * This function also pushes statistics updates to rules which each facet
4074 * resubmits into. Generally these statistics will be accurate. However, if a
4075 * facet changes the rule it resubmits into at some time in between
4076 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4077 * old rule will be incorrectly attributed to the new rule. This could be
4078 * avoided by calling ofproto_update_stats() whenever rules are created or
4079 * deleted. However, the performance impact of making so many calls to the
4080 * datapath do not justify the benefit of having perfectly accurate statistics.
4083 ofproto_update_stats(struct ofproto *p)
4085 const struct dpif_flow_stats *stats;
4086 struct dpif_flow_dump dump;
4087 const struct nlattr *key;
4090 dpif_flow_dump_start(&dump, p->dpif);
4091 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4092 struct facet *facet;
4095 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4099 odp_flow_key_format(key, key_len, &s);
4100 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4106 facet = facet_find(p, &flow);
4108 if (facet && facet->installed) {
4110 if (stats->n_packets >= facet->dp_packet_count) {
4111 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4113 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4116 if (stats->n_bytes >= facet->dp_byte_count) {
4117 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4119 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4122 facet->dp_packet_count = stats->n_packets;
4123 facet->dp_byte_count = stats->n_bytes;
4125 facet_update_time(p, facet, stats->used);
4126 facet_account(p, facet, stats->n_bytes);
4127 facet_push_stats(p, facet);
4129 /* There's a flow in the datapath that we know nothing about.
4131 COVERAGE_INC(ofproto_unexpected_rule);
4132 dpif_flow_del(p->dpif, key, key_len, NULL);
4135 dpif_flow_dump_done(&dump);
4138 /* Calculates and returns the number of milliseconds of idle time after which
4139 * facets should expire from the datapath and we should fold their statistics
4140 * into their parent rules in userspace. */
4142 ofproto_dp_max_idle(const struct ofproto *ofproto)
4145 * Idle time histogram.
4147 * Most of the time a switch has a relatively small number of facets. When
4148 * this is the case we might as well keep statistics for all of them in
4149 * userspace and to cache them in the kernel datapath for performance as
4152 * As the number of facets increases, the memory required to maintain
4153 * statistics about them in userspace and in the kernel becomes
4154 * significant. However, with a large number of facets it is likely that
4155 * only a few of them are "heavy hitters" that consume a large amount of
4156 * bandwidth. At this point, only heavy hitters are worth caching in the
4157 * kernel and maintaining in userspaces; other facets we can discard.
4159 * The technique used to compute the idle time is to build a histogram with
4160 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4161 * that is installed in the kernel gets dropped in the appropriate bucket.
4162 * After the histogram has been built, we compute the cutoff so that only
4163 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4164 * cached. At least the most-recently-used bucket of facets is kept, so
4165 * actually an arbitrary number of facets can be kept in any given
4166 * expiration run (though the next run will delete most of those unless
4167 * they receive additional data).
4169 * This requires a second pass through the facets, in addition to the pass
4170 * made by ofproto_update_stats(), because the former function never looks
4171 * at uninstallable facets.
4173 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4174 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4175 int buckets[N_BUCKETS] = { 0 };
4176 struct facet *facet;
4181 total = hmap_count(&ofproto->facets);
4182 if (total <= 1000) {
4183 return N_BUCKETS * BUCKET_WIDTH;
4186 /* Build histogram. */
4188 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4189 long long int idle = now - facet->used;
4190 int bucket = (idle <= 0 ? 0
4191 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4192 : (unsigned int) idle / BUCKET_WIDTH);
4196 /* Find the first bucket whose flows should be expired. */
4197 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4198 if (buckets[bucket]) {
4201 subtotal += buckets[bucket++];
4202 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4207 if (VLOG_IS_DBG_ENABLED()) {
4211 ds_put_cstr(&s, "keep");
4212 for (i = 0; i < N_BUCKETS; i++) {
4214 ds_put_cstr(&s, ", drop");
4217 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4220 VLOG_INFO("%s: %s (msec:count)",
4221 dpif_name(ofproto->dpif), ds_cstr(&s));
4225 return bucket * BUCKET_WIDTH;
4229 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4231 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4232 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4233 struct ofexpired expired;
4235 if (facet->installed) {
4236 struct dpif_flow_stats stats;
4238 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4240 facet_update_stats(ofproto, facet, &stats);
4243 expired.flow = facet->flow;
4244 expired.packet_count = facet->packet_count;
4245 expired.byte_count = facet->byte_count;
4246 expired.used = facet->used;
4247 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4252 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4254 long long int cutoff = time_msec() - dp_max_idle;
4255 struct facet *facet, *next_facet;
4257 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4258 facet_active_timeout(ofproto, facet);
4259 if (facet->used < cutoff) {
4260 facet_remove(ofproto, facet);
4265 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4266 * then delete it entirely. */
4268 rule_expire(struct ofproto *ofproto, struct rule *rule)
4270 struct facet *facet, *next_facet;
4274 /* Has 'rule' expired? */
4276 if (rule->hard_timeout
4277 && now > rule->created + rule->hard_timeout * 1000) {
4278 reason = OFPRR_HARD_TIMEOUT;
4279 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4280 && now >rule->used + rule->idle_timeout * 1000) {
4281 reason = OFPRR_IDLE_TIMEOUT;
4286 COVERAGE_INC(ofproto_expired);
4288 /* Update stats. (This is a no-op if the rule expired due to an idle
4289 * timeout, because that only happens when the rule has no facets left.) */
4290 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4291 facet_remove(ofproto, facet);
4294 /* Get rid of the rule. */
4295 if (!rule_is_hidden(rule)) {
4296 rule_send_removed(ofproto, rule, reason);
4298 rule_remove(ofproto, rule);
4302 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4304 struct ofputil_flow_removed fr;
4306 if (!rule->send_flow_removed) {
4311 fr.cookie = rule->flow_cookie;
4313 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4314 fr.idle_timeout = rule->idle_timeout;
4315 fr.packet_count = rule->packet_count;
4316 fr.byte_count = rule->byte_count;
4318 connmgr_send_flow_removed(p->connmgr, &fr);
4321 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4322 * The returned statistics include statistics for all of 'rule''s facets. */
4324 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4327 struct facet *facet;
4329 /* Start from historical data for 'rule' itself that are no longer tracked
4330 * in facets. This counts, for example, facets that have expired. */
4331 p = rule->packet_count;
4332 b = rule->byte_count;
4334 /* Add any statistics that are tracked by facets. This includes
4335 * statistical data recently updated by ofproto_update_stats() as well as
4336 * stats for packets that were executed "by hand" via dpif_execute(). */
4337 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4338 p += facet->packet_count;
4339 b += facet->byte_count;
4346 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4347 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4348 * their individual configurations.
4350 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4351 * Otherwise, ownership is transferred to this function. */
4353 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4354 const struct flow *flow, bool clone)
4356 struct ofputil_packet_in pin;
4358 pin.packet = upcall->packet;
4359 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4360 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4361 pin.buffer_id = 0; /* not yet known */
4362 pin.send_len = upcall->userdata;
4363 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4364 clone ? NULL : upcall->packet);
4368 pick_datapath_id(const struct ofproto *ofproto)
4370 const struct ofport *port;
4372 port = get_port(ofproto, ODPP_LOCAL);
4374 uint8_t ea[ETH_ADDR_LEN];
4377 error = netdev_get_etheraddr(port->netdev, ea);
4379 return eth_addr_to_uint64(ea);
4381 VLOG_WARN("could not get MAC address for %s (%s)",
4382 netdev_get_name(port->netdev), strerror(error));
4384 return ofproto->fallback_dpid;
4388 pick_fallback_dpid(void)
4390 uint8_t ea[ETH_ADDR_LEN];
4391 eth_addr_nicira_random(ea);
4392 return eth_addr_to_uint64(ea);
4396 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4397 void *aux OVS_UNUSED)
4399 const struct shash_node *node;
4403 SHASH_FOR_EACH (node, &all_ofprotos) {
4404 ds_put_format(&results, "%s\n", node->name);
4406 unixctl_command_reply(conn, 200, ds_cstr(&results));
4407 ds_destroy(&results);
4410 struct ofproto_trace {
4411 struct action_xlate_ctx ctx;
4417 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4419 ds_put_char_multiple(result, '\t', level);
4421 ds_put_cstr(result, "No match\n");
4425 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4426 ntohll(rule->flow_cookie));
4427 cls_rule_format(&rule->cr, result);
4428 ds_put_char(result, '\n');
4430 ds_put_char_multiple(result, '\t', level);
4431 ds_put_cstr(result, "OpenFlow ");
4432 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4433 rule->n_actions * sizeof *rule->actions);
4434 ds_put_char(result, '\n');
4438 trace_format_flow(struct ds *result, int level, const char *title,
4439 struct ofproto_trace *trace)
4441 ds_put_char_multiple(result, '\t', level);
4442 ds_put_format(result, "%s: ", title);
4443 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4444 ds_put_cstr(result, "unchanged");
4446 flow_format(result, &trace->ctx.flow);
4447 trace->flow = trace->ctx.flow;
4449 ds_put_char(result, '\n');
4453 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4455 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4456 struct ds *result = trace->result;
4458 ds_put_char(result, '\n');
4459 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4460 trace_format_rule(result, ctx->recurse + 1, rule);
4464 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4465 void *aux OVS_UNUSED)
4467 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4468 char *args = xstrdup(args_);
4469 char *save_ptr = NULL;
4470 struct ofproto *ofproto;
4471 struct ofpbuf packet;
4479 ofpbuf_init(&packet, strlen(args) / 2);
4482 dpname = strtok_r(args, " ", &save_ptr);
4483 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4484 in_port_s = strtok_r(NULL, " ", &save_ptr);
4485 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4486 if (!dpname || !in_port_s || !packet_s) {
4487 unixctl_command_reply(conn, 501, "Bad command syntax");
4491 ofproto = shash_find_data(&all_ofprotos, dpname);
4493 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4498 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4499 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4501 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4502 packet_s += strspn(packet_s, " ");
4503 if (*packet_s != '\0') {
4504 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4507 if (packet.size < ETH_HEADER_LEN) {
4508 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4512 ds_put_cstr(&result, "Packet: ");
4513 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4514 ds_put_cstr(&result, s);
4517 flow_extract(&packet, tun_id, in_port, &flow);
4518 ds_put_cstr(&result, "Flow: ");
4519 flow_format(&result, &flow);
4520 ds_put_char(&result, '\n');
4522 rule = rule_lookup(ofproto, &flow);
4523 trace_format_rule(&result, 0, rule);
4525 struct ofproto_trace trace;
4526 struct ofpbuf *odp_actions;
4528 trace.result = &result;
4530 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4531 trace.ctx.resubmit_hook = trace_resubmit;
4532 odp_actions = xlate_actions(&trace.ctx,
4533 rule->actions, rule->n_actions);
4535 ds_put_char(&result, '\n');
4536 trace_format_flow(&result, 0, "Final flow", &trace);
4537 ds_put_cstr(&result, "Datapath actions: ");
4538 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4539 ofpbuf_delete(odp_actions);
4542 unixctl_command_reply(conn, 200, ds_cstr(&result));
4545 ds_destroy(&result);
4546 ofpbuf_uninit(&packet);
4551 ofproto_unixctl_init(void)
4553 static bool registered;
4559 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4560 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4564 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4565 struct ofpbuf *odp_actions, tag_type *tags,
4566 uint16_t *nf_output_iface, void *ofproto_)
4568 struct ofproto *ofproto = ofproto_;
4569 struct mac_entry *dst_mac;
4571 /* Drop frames for reserved multicast addresses. */
4572 if (eth_addr_is_reserved(flow->dl_dst)) {
4576 /* Learn source MAC (but don't try to learn from revalidation). */
4578 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4579 struct mac_entry *src_mac;
4581 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4582 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4583 /* The log messages here could actually be useful in debugging,
4584 * so keep the rate limit relatively high. */
4585 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4586 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4587 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4589 ofproto_revalidate(ofproto,
4590 mac_learning_changed(ofproto->ml, src_mac));
4591 src_mac->port.i = flow->in_port;
4595 /* Determine output port. */
4596 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4598 flood_packets(ofproto, flow->in_port, htonl(OFPPC_NO_FLOOD),
4599 nf_output_iface, odp_actions);
4601 int out_port = dst_mac->port.i;
4602 if (out_port != flow->in_port) {
4603 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4604 *nf_output_iface = out_port;
4613 static const struct ofhooks default_ofhooks = {
4614 default_normal_ofhook_cb,