2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
29 #include "classifier.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
38 #include "mac-learning.h"
39 #include "multipath.h"
45 #include "ofp-print.h"
47 #include "ofproto-sflow.h"
49 #include "openflow/nicira-ext.h"
50 #include "openflow/openflow.h"
51 #include "openvswitch/datapath-protocol.h"
55 #include "poll-loop.h"
59 #include "stream-ssl.h"
64 #include "unaligned.h"
69 VLOG_DEFINE_THIS_MODULE(ofproto);
71 COVERAGE_DEFINE(facet_changed_rule);
72 COVERAGE_DEFINE(facet_revalidate);
73 COVERAGE_DEFINE(odp_overflow);
74 COVERAGE_DEFINE(ofproto_agg_request);
75 COVERAGE_DEFINE(ofproto_costly_flags);
76 COVERAGE_DEFINE(ofproto_ctlr_action);
77 COVERAGE_DEFINE(ofproto_del_rule);
78 COVERAGE_DEFINE(ofproto_error);
79 COVERAGE_DEFINE(ofproto_expiration);
80 COVERAGE_DEFINE(ofproto_expired);
81 COVERAGE_DEFINE(ofproto_flows_req);
82 COVERAGE_DEFINE(ofproto_flush);
83 COVERAGE_DEFINE(ofproto_invalidated);
84 COVERAGE_DEFINE(ofproto_no_packet_in);
85 COVERAGE_DEFINE(ofproto_ofp2odp);
86 COVERAGE_DEFINE(ofproto_packet_in);
87 COVERAGE_DEFINE(ofproto_packet_out);
88 COVERAGE_DEFINE(ofproto_queue_req);
89 COVERAGE_DEFINE(ofproto_recv_openflow);
90 COVERAGE_DEFINE(ofproto_reinit_ports);
91 COVERAGE_DEFINE(ofproto_unexpected_rule);
92 COVERAGE_DEFINE(ofproto_uninstallable);
93 COVERAGE_DEFINE(ofproto_update_port);
95 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
96 * flow translation. */
97 #define MAX_RESUBMIT_RECURSION 16
102 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
103 struct netdev *netdev;
104 struct ofp_phy_port opp; /* In host byte order. */
106 struct cfm *cfm; /* Connectivity Fault Management, if any. */
109 static void ofport_free(struct ofport *);
110 static void ofport_run(struct ofproto *, struct ofport *);
111 static void ofport_wait(struct ofport *);
113 struct action_xlate_ctx {
114 /* action_xlate_ctx_init() initializes these members. */
117 struct ofproto *ofproto;
119 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
120 * this flow when actions change header fields. */
123 /* The packet corresponding to 'flow', or a null pointer if we are
124 * revalidating without a packet to refer to. */
125 const struct ofpbuf *packet;
127 /* If nonnull, called just before executing a resubmit action.
129 * This is normally null so the client has to set it manually after
130 * calling action_xlate_ctx_init(). */
131 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
133 /* If true, the speciality of 'flow' should be checked before executing
134 * its actions. If special_cb returns false on 'flow' rendered
135 * uninstallable and no actions will be executed. */
138 /* xlate_actions() initializes and uses these members. The client might want
139 * to look at them after it returns. */
141 struct ofpbuf *odp_actions; /* Datapath actions. */
142 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
143 bool may_set_up_flow; /* True ordinarily; false if the actions must
144 * be reassessed for every packet. */
145 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
147 /* xlate_actions() initializes and uses these members, but the client has no
148 * reason to look at them. */
150 int recurse; /* Recursion level, via xlate_table_action. */
151 int last_pop_priority; /* Offset in 'odp_actions' just past most
152 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
155 static void action_xlate_ctx_init(struct action_xlate_ctx *,
156 struct ofproto *, const struct flow *,
157 const struct ofpbuf *);
158 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
159 const union ofp_action *in, size_t n_in);
161 /* An OpenFlow flow. */
163 long long int used; /* Time last used; time created if not used. */
164 long long int created; /* Creation time. */
168 * - Do include packets and bytes from facets that have been deleted or
169 * whose own statistics have been folded into the rule.
171 * - Do include packets and bytes sent "by hand" that were accounted to
172 * the rule without any facet being involved (this is a rare corner
173 * case in rule_execute()).
175 * - Do not include packet or bytes that can be obtained from any facet's
176 * packet_count or byte_count member or that can be obtained from the
177 * datapath by, e.g., dpif_flow_get() for any facet.
179 uint64_t packet_count; /* Number of packets received. */
180 uint64_t byte_count; /* Number of bytes received. */
182 ovs_be64 flow_cookie; /* Controller-issued identifier. */
184 struct cls_rule cr; /* In owning ofproto's classifier. */
185 uint16_t idle_timeout; /* In seconds from time of last use. */
186 uint16_t hard_timeout; /* In seconds from time of creation. */
187 bool send_flow_removed; /* Send a flow removed message? */
188 int n_actions; /* Number of elements in actions[]. */
189 union ofp_action *actions; /* OpenFlow actions. */
190 struct list facets; /* List of "struct facet"s. */
193 static struct rule *rule_from_cls_rule(const struct cls_rule *);
194 static bool rule_is_hidden(const struct rule *);
196 static struct rule *rule_create(const struct cls_rule *,
197 const union ofp_action *, size_t n_actions,
198 uint16_t idle_timeout, uint16_t hard_timeout,
199 ovs_be64 flow_cookie, bool send_flow_removed);
200 static void rule_destroy(struct ofproto *, struct rule *);
201 static void rule_free(struct rule *);
203 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
204 static void rule_insert(struct ofproto *, struct rule *);
205 static void rule_remove(struct ofproto *, struct rule *);
207 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
208 static void rule_get_stats(const struct rule *, uint64_t *packets,
211 /* An exact-match instantiation of an OpenFlow flow. */
213 long long int used; /* Time last used; time created if not used. */
217 * - Do include packets and bytes sent "by hand", e.g. with
220 * - Do include packets and bytes that were obtained from the datapath
221 * when a flow was deleted (e.g. dpif_flow_del()) or when its
222 * statistics were reset (e.g. dpif_flow_put() with
223 * DPIF_FP_ZERO_STATS).
225 * - Do not include any packets or bytes that can currently be obtained
226 * from the datapath by, e.g., dpif_flow_get().
228 uint64_t packet_count; /* Number of packets received. */
229 uint64_t byte_count; /* Number of bytes received. */
231 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
232 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
234 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
235 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
236 long long int rs_used; /* Used time pushed to resubmit children. */
238 /* Number of bytes passed to account_cb. This may include bytes that can
239 * currently obtained from the datapath (thus, it can be greater than
241 uint64_t accounted_bytes;
243 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
244 struct list list_node; /* In owning rule's 'facets' list. */
245 struct rule *rule; /* Owning rule. */
246 struct flow flow; /* Exact-match flow. */
247 bool installed; /* Installed in datapath? */
248 bool may_install; /* True ordinarily; false if actions must
249 * be reassessed for every packet. */
250 size_t actions_len; /* Number of bytes in actions[]. */
251 struct nlattr *actions; /* Datapath actions. */
252 tag_type tags; /* Tags (set only by hooks). */
253 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
256 static struct facet *facet_create(struct ofproto *, struct rule *,
258 const struct ofpbuf *packet);
259 static void facet_remove(struct ofproto *, struct facet *);
260 static void facet_free(struct facet *);
262 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
263 static bool facet_revalidate(struct ofproto *, struct facet *);
265 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
266 static void facet_uninstall(struct ofproto *, struct facet *);
267 static void facet_flush_stats(struct ofproto *, struct facet *);
269 static void facet_make_actions(struct ofproto *, struct facet *,
270 const struct ofpbuf *packet);
271 static void facet_update_stats(struct ofproto *, struct facet *,
272 const struct dpif_flow_stats *);
273 static void facet_push_stats(struct ofproto *, struct facet *);
275 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
276 const struct flow *, bool clone);
280 uint64_t datapath_id; /* Datapath ID. */
281 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
282 char *mfr_desc; /* Manufacturer. */
283 char *hw_desc; /* Hardware. */
284 char *sw_desc; /* Software version. */
285 char *serial_desc; /* Serial number. */
286 char *dp_desc; /* Datapath description. */
290 struct netdev_monitor *netdev_monitor;
291 struct hmap ports; /* Contains "struct ofport"s. */
292 struct shash port_by_name;
296 struct netflow *netflow;
297 struct ofproto_sflow *sflow;
300 struct classifier cls;
301 struct timer next_expiration;
305 bool need_revalidate;
306 struct tag_set revalidate_set;
308 /* OpenFlow connections. */
309 struct connmgr *connmgr;
311 /* Hooks for ovs-vswitchd. */
312 const struct ofhooks *ofhooks;
315 /* Used by default ofhooks. */
316 struct mac_learning *ml;
319 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
320 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
322 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
324 static const struct ofhooks default_ofhooks;
326 static uint64_t pick_datapath_id(const struct ofproto *);
327 static uint64_t pick_fallback_dpid(void);
329 static int ofproto_expire(struct ofproto *);
330 static void flow_push_stats(struct ofproto *, const struct rule *,
331 struct flow *, uint64_t packets, uint64_t bytes,
334 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
336 static void handle_openflow(struct ofconn *, struct ofpbuf *);
338 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
339 static void update_port(struct ofproto *, const char *devname);
340 static int init_ports(struct ofproto *);
341 static void reinit_ports(struct ofproto *);
343 static void ofproto_unixctl_init(void);
346 ofproto_create(const char *datapath, const char *datapath_type,
347 const struct ofhooks *ofhooks, void *aux,
348 struct ofproto **ofprotop)
350 char local_name[IF_NAMESIZE];
357 ofproto_unixctl_init();
359 /* Connect to datapath and start listening for messages. */
360 error = dpif_open(datapath, datapath_type, &dpif);
362 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
365 error = dpif_recv_set_mask(dpif,
366 ((1u << DPIF_UC_MISS) |
367 (1u << DPIF_UC_ACTION) |
368 (1u << DPIF_UC_SAMPLE)));
370 VLOG_ERR("failed to listen on datapath %s: %s",
371 datapath, strerror(error));
375 dpif_flow_flush(dpif);
376 dpif_recv_purge(dpif);
378 error = dpif_port_get_name(dpif, ODPP_LOCAL,
379 local_name, sizeof local_name);
381 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
382 datapath, strerror(error));
386 /* Initialize settings. */
387 p = xzalloc(sizeof *p);
388 p->fallback_dpid = pick_fallback_dpid();
389 p->datapath_id = p->fallback_dpid;
390 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
391 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
392 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
393 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
394 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
396 /* Initialize datapath. */
398 p->netdev_monitor = netdev_monitor_create();
399 hmap_init(&p->ports);
400 shash_init(&p->port_by_name);
401 p->max_ports = dpif_get_max_ports(dpif);
403 /* Initialize submodules. */
407 /* Initialize flow table. */
408 classifier_init(&p->cls);
409 timer_set_duration(&p->next_expiration, 1000);
411 /* Initialize facet table. */
412 hmap_init(&p->facets);
413 p->need_revalidate = false;
414 tag_set_init(&p->revalidate_set);
416 /* Initialize hooks. */
418 p->ofhooks = ofhooks;
422 p->ofhooks = &default_ofhooks;
424 p->ml = mac_learning_create();
427 /* Pick final datapath ID. */
428 p->datapath_id = pick_datapath_id(p);
429 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
431 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
433 /* Initialize OpenFlow connections. */
434 p->connmgr = connmgr_create(p, datapath, local_name);
441 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
443 uint64_t old_dpid = p->datapath_id;
444 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
445 if (p->datapath_id != old_dpid) {
446 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
448 /* Force all active connections to reconnect, since there is no way to
449 * notify a controller that the datapath ID has changed. */
450 ofproto_reconnect_controllers(p);
455 ofproto_set_controllers(struct ofproto *p,
456 const struct ofproto_controller *controllers,
457 size_t n_controllers)
459 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
463 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
465 connmgr_set_fail_mode(p->connmgr, fail_mode);
468 /* Drops the connections between 'ofproto' and all of its controllers, forcing
469 * them to reconnect. */
471 ofproto_reconnect_controllers(struct ofproto *ofproto)
473 connmgr_reconnect(ofproto->connmgr);
476 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
477 * in-band control should guarantee access, in the same way that in-band
478 * control guarantees access to OpenFlow controllers. */
480 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
481 const struct sockaddr_in *extras, size_t n)
483 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
486 /* Sets the OpenFlow queue used by flows set up by in-band control on
487 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
488 * flows will use the default queue. */
490 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
492 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
496 ofproto_set_desc(struct ofproto *p,
497 const char *mfr_desc, const char *hw_desc,
498 const char *sw_desc, const char *serial_desc,
501 struct ofp_desc_stats *ods;
504 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
505 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
506 sizeof ods->mfr_desc);
509 p->mfr_desc = xstrdup(mfr_desc);
512 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
513 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
514 sizeof ods->hw_desc);
517 p->hw_desc = xstrdup(hw_desc);
520 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
521 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
522 sizeof ods->sw_desc);
525 p->sw_desc = xstrdup(sw_desc);
528 if (strlen(serial_desc) >= sizeof ods->serial_num) {
529 VLOG_WARN("truncating serial_desc, must be less than %zu "
531 sizeof ods->serial_num);
533 free(p->serial_desc);
534 p->serial_desc = xstrdup(serial_desc);
537 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
538 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
539 sizeof ods->dp_desc);
542 p->dp_desc = xstrdup(dp_desc);
547 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
549 return connmgr_set_snoops(ofproto->connmgr, snoops);
553 ofproto_set_netflow(struct ofproto *ofproto,
554 const struct netflow_options *nf_options)
556 if (nf_options && nf_options->collectors.n) {
557 if (!ofproto->netflow) {
558 ofproto->netflow = netflow_create();
560 return netflow_set_options(ofproto->netflow, nf_options);
562 netflow_destroy(ofproto->netflow);
563 ofproto->netflow = NULL;
569 ofproto_set_sflow(struct ofproto *ofproto,
570 const struct ofproto_sflow_options *oso)
572 struct ofproto_sflow *os = ofproto->sflow;
575 struct ofport *ofport;
577 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
578 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
579 ofproto_sflow_add_port(os, ofport->odp_port,
580 netdev_get_name(ofport->netdev));
583 ofproto_sflow_set_options(os, oso);
585 ofproto_sflow_destroy(os);
586 ofproto->sflow = NULL;
590 /* Connectivity Fault Management configuration. */
592 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
594 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
596 struct ofport *ofport = get_port(ofproto, port_no);
597 if (ofport && ofport->cfm){
598 cfm_destroy(ofport->cfm);
603 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
604 * basic configuration from the configuration members in 'cfm', and the set of
605 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
606 * Ignores the statistics members of 'cfm'.
608 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
610 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
611 const struct cfm *cfm,
612 const uint16_t *remote_mps, size_t n_remote_mps)
614 struct ofport *ofport;
616 ofport = get_port(ofproto, port_no);
618 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
619 dpif_name(ofproto->dpif), port_no);
624 ofport->cfm = cfm_create();
627 ofport->cfm->mpid = cfm->mpid;
628 ofport->cfm->interval = cfm->interval;
629 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
631 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
633 if (!cfm_configure(ofport->cfm)) {
634 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
635 dpif_name(ofproto->dpif), port_no,
636 netdev_get_name(ofport->netdev));
637 cfm_destroy(ofport->cfm);
642 /* Returns the connectivity fault management object associated with 'port_no'
643 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
644 * 'port_no' or if that port does not have CFM configured. The caller must not
645 * modify or destroy the returned object. */
647 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
649 struct ofport *ofport = get_port(ofproto, port_no);
650 return ofport ? ofport->cfm : NULL;
654 ofproto_get_datapath_id(const struct ofproto *ofproto)
656 return ofproto->datapath_id;
660 ofproto_has_primary_controller(const struct ofproto *ofproto)
662 return connmgr_has_controllers(ofproto->connmgr);
665 enum ofproto_fail_mode
666 ofproto_get_fail_mode(const struct ofproto *p)
668 return connmgr_get_fail_mode(p->connmgr);
672 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
674 connmgr_get_snoops(ofproto->connmgr, snoops);
678 ofproto_destroy(struct ofproto *p)
680 struct ofport *ofport, *next_ofport;
686 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
688 ofproto_flush_flows(p);
689 connmgr_destroy(p->connmgr);
690 classifier_destroy(&p->cls);
691 hmap_destroy(&p->facets);
694 netdev_monitor_destroy(p->netdev_monitor);
695 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
696 hmap_remove(&p->ports, &ofport->hmap_node);
699 shash_destroy(&p->port_by_name);
701 netflow_destroy(p->netflow);
702 ofproto_sflow_destroy(p->sflow);
704 mac_learning_destroy(p->ml);
709 free(p->serial_desc);
712 hmap_destroy(&p->ports);
718 ofproto_run(struct ofproto *p)
720 int error = ofproto_run1(p);
722 error = ofproto_run2(p, false);
728 process_port_change(struct ofproto *ofproto, int error, char *devname)
730 if (error == ENOBUFS) {
731 reinit_ports(ofproto);
733 update_port(ofproto, devname);
739 ofproto_run1(struct ofproto *p)
741 struct ofport *ofport;
746 if (shash_is_empty(&p->port_by_name)) {
750 for (i = 0; i < 50; i++) {
751 struct dpif_upcall packet;
753 error = dpif_recv(p->dpif, &packet);
755 if (error == ENODEV) {
756 /* Someone destroyed the datapath behind our back. The caller
757 * better destroy us and give up, because we're just going to
758 * spin from here on out. */
759 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
760 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
767 handle_upcall(p, &packet);
770 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
771 process_port_change(p, error, devname);
773 while ((error = netdev_monitor_poll(p->netdev_monitor,
774 &devname)) != EAGAIN) {
775 process_port_change(p, error, devname);
778 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
779 ofport_run(p, ofport);
782 connmgr_run(p->connmgr, handle_openflow);
784 if (timer_expired(&p->next_expiration)) {
785 int delay = ofproto_expire(p);
786 timer_set_duration(&p->next_expiration, delay);
787 COVERAGE_INC(ofproto_expiration);
791 netflow_run(p->netflow);
794 ofproto_sflow_run(p->sflow);
801 ofproto_run2(struct ofproto *p, bool revalidate_all)
803 /* Figure out what we need to revalidate now, if anything. */
804 struct tag_set revalidate_set = p->revalidate_set;
805 if (p->need_revalidate) {
806 revalidate_all = true;
809 /* Clear the revalidation flags. */
810 tag_set_init(&p->revalidate_set);
811 p->need_revalidate = false;
813 /* Now revalidate if there's anything to do. */
814 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
815 struct facet *facet, *next;
817 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
819 || tag_set_intersects(&revalidate_set, facet->tags)) {
820 facet_revalidate(p, facet);
829 ofproto_wait(struct ofproto *p)
831 struct ofport *ofport;
833 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
836 dpif_recv_wait(p->dpif);
837 dpif_port_poll_wait(p->dpif);
838 netdev_monitor_poll_wait(p->netdev_monitor);
840 ofproto_sflow_wait(p->sflow);
842 if (!tag_set_is_empty(&p->revalidate_set)) {
843 poll_immediate_wake();
845 if (p->need_revalidate) {
846 /* Shouldn't happen, but if it does just go around again. */
847 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
848 poll_immediate_wake();
850 timer_wait(&p->next_expiration);
852 connmgr_wait(p->connmgr);
856 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
858 tag_set_add(&ofproto->revalidate_set, tag);
862 ofproto_get_revalidate_set(struct ofproto *ofproto)
864 return &ofproto->revalidate_set;
868 ofproto_is_alive(const struct ofproto *p)
870 return connmgr_has_controllers(p->connmgr);
874 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
877 connmgr_get_controller_info(ofproto->connmgr, info);
881 ofproto_free_ofproto_controller_info(struct shash *info)
883 struct shash_node *node;
885 SHASH_FOR_EACH (node, info) {
886 struct ofproto_controller_info *cinfo = node->data;
887 while (cinfo->pairs.n) {
888 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
895 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
897 * This is almost the same as calling dpif_port_del() directly on the
898 * datapath, but it also makes 'ofproto' close its open netdev for the port
899 * (if any). This makes it possible to create a new netdev of a different
900 * type under the same name, which otherwise the netdev library would refuse
901 * to do because of the conflict. (The netdev would eventually get closed on
902 * the next trip through ofproto_run(), but this interface is more direct.)
904 * Returns 0 if successful, otherwise a positive errno. */
906 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
908 struct ofport *ofport = get_port(ofproto, odp_port);
909 const char *name = ofport ? ofport->opp.name : "<unknown>";
912 error = dpif_port_del(ofproto->dpif, odp_port);
914 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
915 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
917 /* 'name' is ofport->opp.name and update_port() is going to destroy
918 * 'ofport'. Just in case update_port() refers to 'name' after it
919 * destroys 'ofport', make a copy of it around the update_port()
921 char *devname = xstrdup(name);
922 update_port(ofproto, devname);
928 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
929 * true if 'odp_port' exists and should be included, false otherwise. */
931 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
933 struct ofport *ofport = get_port(ofproto, odp_port);
934 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
937 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
938 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
939 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
941 * Returns 0 if successful, otherwise a positive errno value. */
943 ofproto_send_packet(struct ofproto *ofproto,
944 uint32_t port_no, uint16_t vlan_tci,
945 const struct ofpbuf *packet)
947 struct ofpbuf odp_actions;
950 ofpbuf_init(&odp_actions, 32);
952 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
953 ntohs(vlan_tci & ~VLAN_CFI));
955 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
956 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
958 ofpbuf_uninit(&odp_actions);
961 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
962 dpif_name(ofproto->dpif), port_no, strerror(error));
967 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
968 * performs the 'n_actions' actions in 'actions'. The new flow will not
971 * If cls_rule->priority is in the range of priorities supported by OpenFlow
972 * (0...65535, inclusive) then the flow will be visible to OpenFlow
973 * controllers; otherwise, it will be hidden.
975 * The caller retains ownership of 'cls_rule' and 'actions'. */
977 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
978 const union ofp_action *actions, size_t n_actions)
981 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
982 rule_insert(p, rule);
986 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
990 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
993 rule_remove(ofproto, rule);
998 ofproto_flush_flows(struct ofproto *ofproto)
1000 struct facet *facet, *next_facet;
1001 struct rule *rule, *next_rule;
1002 struct cls_cursor cursor;
1004 COVERAGE_INC(ofproto_flush);
1006 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1007 /* Mark the facet as not installed so that facet_remove() doesn't
1008 * bother trying to uninstall it. There is no point in uninstalling it
1009 * individually since we are about to blow away all the facets with
1010 * dpif_flow_flush(). */
1011 facet->installed = false;
1012 facet->dp_packet_count = 0;
1013 facet->dp_byte_count = 0;
1014 facet_remove(ofproto, facet);
1017 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1018 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1019 rule_remove(ofproto, rule);
1022 dpif_flow_flush(ofproto->dpif);
1023 connmgr_flushed(ofproto->connmgr);
1027 reinit_ports(struct ofproto *p)
1029 struct dpif_port_dump dump;
1030 struct sset devnames;
1031 struct ofport *ofport;
1032 struct dpif_port dpif_port;
1033 const char *devname;
1035 COVERAGE_INC(ofproto_reinit_ports);
1037 sset_init(&devnames);
1038 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1039 sset_add(&devnames, ofport->opp.name);
1041 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1042 sset_add(&devnames, dpif_port.name);
1045 SSET_FOR_EACH (devname, &devnames) {
1046 update_port(p, devname);
1048 sset_destroy(&devnames);
1051 static struct ofport *
1052 make_ofport(const struct dpif_port *dpif_port)
1054 struct netdev_options netdev_options;
1055 enum netdev_flags flags;
1056 struct ofport *ofport;
1057 struct netdev *netdev;
1060 memset(&netdev_options, 0, sizeof netdev_options);
1061 netdev_options.name = dpif_port->name;
1062 netdev_options.type = dpif_port->type;
1063 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1065 error = netdev_open(&netdev_options, &netdev);
1067 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1068 "cannot be opened (%s)",
1069 dpif_port->name, dpif_port->port_no,
1070 dpif_port->name, strerror(error));
1074 ofport = xzalloc(sizeof *ofport);
1075 ofport->netdev = netdev;
1076 ofport->odp_port = dpif_port->port_no;
1077 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1078 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1079 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1081 netdev_get_flags(netdev, &flags);
1082 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1084 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1086 netdev_get_features(netdev,
1087 &ofport->opp.curr, &ofport->opp.advertised,
1088 &ofport->opp.supported, &ofport->opp.peer);
1093 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1095 if (get_port(p, dpif_port->port_no)) {
1096 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1097 dpif_port->port_no);
1099 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1100 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1109 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1111 const struct ofp_phy_port *a = &a_->opp;
1112 const struct ofp_phy_port *b = &b_->opp;
1114 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1115 return (a->port_no == b->port_no
1116 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1117 && !strcmp(a->name, b->name)
1118 && a->state == b->state
1119 && a->config == b->config
1120 && a->curr == b->curr
1121 && a->advertised == b->advertised
1122 && a->supported == b->supported
1123 && a->peer == b->peer);
1127 ofport_install(struct ofproto *p, struct ofport *ofport)
1129 const char *netdev_name = ofport->opp.name;
1131 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1132 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1133 shash_add(&p->port_by_name, netdev_name, ofport);
1135 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1140 ofport_remove(struct ofproto *p, struct ofport *ofport)
1142 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1143 hmap_remove(&p->ports, &ofport->hmap_node);
1144 shash_delete(&p->port_by_name,
1145 shash_find(&p->port_by_name, ofport->opp.name));
1147 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1152 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1155 cfm_run(ofport->cfm);
1157 if (cfm_should_send_ccm(ofport->cfm)) {
1158 struct ofpbuf packet;
1161 ofpbuf_init(&packet, 0);
1162 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1163 ETH_TYPE_CFM, sizeof *ccm);
1164 cfm_compose_ccm(ofport->cfm, ccm);
1165 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1166 ofpbuf_uninit(&packet);
1172 ofport_wait(struct ofport *ofport)
1175 cfm_wait(ofport->cfm);
1180 ofport_free(struct ofport *ofport)
1183 cfm_destroy(ofport->cfm);
1184 netdev_close(ofport->netdev);
1189 static struct ofport *
1190 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1192 struct ofport *port;
1194 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1195 hash_int(odp_port, 0), &ofproto->ports) {
1196 if (port->odp_port == odp_port) {
1204 update_port(struct ofproto *p, const char *devname)
1206 struct dpif_port dpif_port;
1207 struct ofport *old_ofport;
1208 struct ofport *new_ofport;
1211 COVERAGE_INC(ofproto_update_port);
1213 /* Query the datapath for port information. */
1214 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1216 /* Find the old ofport. */
1217 old_ofport = shash_find_data(&p->port_by_name, devname);
1220 /* There's no port named 'devname' but there might be a port with
1221 * the same port number. This could happen if a port is deleted
1222 * and then a new one added in its place very quickly, or if a port
1223 * is renamed. In the former case we want to send an OFPPR_DELETE
1224 * and an OFPPR_ADD, and in the latter case we want to send a
1225 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1226 * the old port's ifindex against the new port, or perhaps less
1227 * reliably but more portably by comparing the old port's MAC
1228 * against the new port's MAC. However, this code isn't that smart
1229 * and always sends an OFPPR_MODIFY (XXX). */
1230 old_ofport = get_port(p, dpif_port.port_no);
1232 } else if (error != ENOENT && error != ENODEV) {
1233 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1234 "%s", strerror(error));
1238 /* Create a new ofport. */
1239 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1241 /* Eliminate a few pathological cases. */
1242 if (!old_ofport && !new_ofport) {
1244 } else if (old_ofport && new_ofport) {
1245 /* Most of the 'config' bits are OpenFlow soft state, but
1246 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1247 * OpenFlow bits from old_ofport. (make_ofport() only sets
1248 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1249 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1251 if (ofport_equal(old_ofport, new_ofport)) {
1252 /* False alarm--no change. */
1253 ofport_free(new_ofport);
1258 /* Now deal with the normal cases. */
1260 ofport_remove(p, old_ofport);
1263 ofport_install(p, new_ofport);
1265 connmgr_send_port_status(p->connmgr,
1266 new_ofport ? &new_ofport->opp : &old_ofport->opp,
1267 (!old_ofport ? OFPPR_ADD
1268 : !new_ofport ? OFPPR_DELETE
1270 ofport_free(old_ofport);
1273 dpif_port_destroy(&dpif_port);
1277 init_ports(struct ofproto *p)
1279 struct dpif_port_dump dump;
1280 struct dpif_port dpif_port;
1282 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1283 if (!ofport_conflicts(p, &dpif_port)) {
1284 struct ofport *ofport = make_ofport(&dpif_port);
1286 ofport_install(p, ofport);
1294 /* Returns true if 'rule' should be hidden from the controller.
1296 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1297 * (e.g. by in-band control) and are intentionally hidden from the
1300 rule_is_hidden(const struct rule *rule)
1302 return rule->cr.priority > UINT16_MAX;
1305 /* Creates and returns a new rule initialized as specified.
1307 * The caller is responsible for inserting the rule into the classifier (with
1308 * rule_insert()). */
1309 static struct rule *
1310 rule_create(const struct cls_rule *cls_rule,
1311 const union ofp_action *actions, size_t n_actions,
1312 uint16_t idle_timeout, uint16_t hard_timeout,
1313 ovs_be64 flow_cookie, bool send_flow_removed)
1315 struct rule *rule = xzalloc(sizeof *rule);
1316 rule->cr = *cls_rule;
1317 rule->idle_timeout = idle_timeout;
1318 rule->hard_timeout = hard_timeout;
1319 rule->flow_cookie = flow_cookie;
1320 rule->used = rule->created = time_msec();
1321 rule->send_flow_removed = send_flow_removed;
1322 list_init(&rule->facets);
1323 if (n_actions > 0) {
1324 rule->n_actions = n_actions;
1325 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1331 static struct rule *
1332 rule_from_cls_rule(const struct cls_rule *cls_rule)
1334 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1338 rule_free(struct rule *rule)
1340 free(rule->actions);
1344 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1345 * destroying any that no longer has a rule (which is probably all of them).
1347 * The caller must have already removed 'rule' from the classifier. */
1349 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1351 struct facet *facet, *next_facet;
1352 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1353 facet_revalidate(ofproto, facet);
1358 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1359 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1362 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1364 const union ofp_action *oa;
1365 struct actions_iterator i;
1367 if (out_port == htons(OFPP_NONE)) {
1370 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1371 oa = actions_next(&i)) {
1372 if (action_outputs_to_port(oa, out_port)) {
1379 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1380 * 'packet', which arrived on 'in_port'.
1382 * Takes ownership of 'packet'. */
1384 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1385 const struct nlattr *odp_actions, size_t actions_len,
1386 struct ofpbuf *packet)
1388 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1389 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1390 /* As an optimization, avoid a round-trip from userspace to kernel to
1391 * userspace. This also avoids possibly filling up kernel packet
1392 * buffers along the way. */
1393 struct dpif_upcall upcall;
1395 upcall.type = DPIF_UC_ACTION;
1396 upcall.packet = packet;
1399 upcall.userdata = nl_attr_get_u64(odp_actions);
1400 upcall.sample_pool = 0;
1401 upcall.actions = NULL;
1402 upcall.actions_len = 0;
1404 send_packet_in(ofproto, &upcall, flow, false);
1410 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1411 ofpbuf_delete(packet);
1416 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1417 * statistics appropriately. 'packet' must have at least sizeof(struct
1418 * ofp_packet_in) bytes of headroom.
1420 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1421 * applying flow_extract() to 'packet' would yield the same flow as
1424 * 'facet' must have accurately composed ODP actions; that is, it must not be
1425 * in need of revalidation.
1427 * Takes ownership of 'packet'. */
1429 facet_execute(struct ofproto *ofproto, struct facet *facet,
1430 struct ofpbuf *packet)
1432 struct dpif_flow_stats stats;
1434 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1436 flow_extract_stats(&facet->flow, packet, &stats);
1437 stats.used = time_msec();
1438 if (execute_odp_actions(ofproto, &facet->flow,
1439 facet->actions, facet->actions_len, packet)) {
1440 facet_update_stats(ofproto, facet, &stats);
1444 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1445 * statistics (or the statistics for one of its facets) appropriately.
1446 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1448 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1449 * with statistics for 'packet' either way.
1451 * Takes ownership of 'packet'. */
1453 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1454 struct ofpbuf *packet)
1456 struct action_xlate_ctx ctx;
1457 struct ofpbuf *odp_actions;
1458 struct facet *facet;
1462 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1464 flow_extract(packet, 0, in_port, &flow);
1466 /* First look for a related facet. If we find one, account it to that. */
1467 facet = facet_lookup_valid(ofproto, &flow);
1468 if (facet && facet->rule == rule) {
1469 facet_execute(ofproto, facet, packet);
1473 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1474 * create a new facet for it and use that. */
1475 if (rule_lookup(ofproto, &flow) == rule) {
1476 facet = facet_create(ofproto, rule, &flow, packet);
1477 facet_execute(ofproto, facet, packet);
1478 facet_install(ofproto, facet, true);
1482 /* We can't account anything to a facet. If we were to try, then that
1483 * facet would have a non-matching rule, busting our invariants. */
1484 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1485 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1486 size = packet->size;
1487 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1488 odp_actions->size, packet)) {
1489 rule->used = time_msec();
1490 rule->packet_count++;
1491 rule->byte_count += size;
1492 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1494 ofpbuf_delete(odp_actions);
1497 /* Inserts 'rule' into 'p''s flow table. */
1499 rule_insert(struct ofproto *p, struct rule *rule)
1501 struct rule *displaced_rule;
1503 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1504 if (displaced_rule) {
1505 rule_destroy(p, displaced_rule);
1507 p->need_revalidate = true;
1510 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1511 * 'flow' and an example 'packet' within that flow.
1513 * The caller must already have determined that no facet with an identical
1514 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1515 * 'ofproto''s classifier table. */
1516 static struct facet *
1517 facet_create(struct ofproto *ofproto, struct rule *rule,
1518 const struct flow *flow, const struct ofpbuf *packet)
1520 struct facet *facet;
1522 facet = xzalloc(sizeof *facet);
1523 facet->used = time_msec();
1524 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1525 list_push_back(&rule->facets, &facet->list_node);
1527 facet->flow = *flow;
1528 netflow_flow_init(&facet->nf_flow);
1529 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1531 facet_make_actions(ofproto, facet, packet);
1537 facet_free(struct facet *facet)
1539 free(facet->actions);
1543 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1545 * - Removes 'rule' from the classifier.
1547 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1548 * destroys them), via rule_destroy().
1551 rule_remove(struct ofproto *ofproto, struct rule *rule)
1553 COVERAGE_INC(ofproto_del_rule);
1554 ofproto->need_revalidate = true;
1555 classifier_remove(&ofproto->cls, &rule->cr);
1556 rule_destroy(ofproto, rule);
1559 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1561 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1562 * rule's statistics, via facet_uninstall().
1564 * - Removes 'facet' from its rule and from ofproto->facets.
1567 facet_remove(struct ofproto *ofproto, struct facet *facet)
1569 facet_uninstall(ofproto, facet);
1570 facet_flush_stats(ofproto, facet);
1571 hmap_remove(&ofproto->facets, &facet->hmap_node);
1572 list_remove(&facet->list_node);
1576 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1578 facet_make_actions(struct ofproto *p, struct facet *facet,
1579 const struct ofpbuf *packet)
1581 const struct rule *rule = facet->rule;
1582 struct ofpbuf *odp_actions;
1583 struct action_xlate_ctx ctx;
1585 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1586 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1587 facet->tags = ctx.tags;
1588 facet->may_install = ctx.may_set_up_flow;
1589 facet->nf_flow.output_iface = ctx.nf_output_iface;
1591 if (facet->actions_len != odp_actions->size
1592 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1593 free(facet->actions);
1594 facet->actions_len = odp_actions->size;
1595 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1598 ofpbuf_delete(odp_actions);
1602 facet_put__(struct ofproto *ofproto, struct facet *facet,
1603 const struct nlattr *actions, size_t actions_len,
1604 struct dpif_flow_stats *stats)
1606 struct odputil_keybuf keybuf;
1607 enum dpif_flow_put_flags flags;
1610 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1612 flags |= DPIF_FP_ZERO_STATS;
1613 facet->dp_packet_count = 0;
1614 facet->dp_byte_count = 0;
1617 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1618 odp_flow_key_from_flow(&key, &facet->flow);
1620 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1621 actions, actions_len, stats);
1624 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1625 * 'zero_stats' is true, clears any existing statistics from the datapath for
1628 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1630 struct dpif_flow_stats stats;
1632 if (facet->may_install
1633 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1634 zero_stats ? &stats : NULL)) {
1635 facet->installed = true;
1639 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1640 * to the accounting hook function in the ofhooks structure. */
1642 facet_account(struct ofproto *ofproto,
1643 struct facet *facet, uint64_t extra_bytes)
1645 uint64_t total_bytes = facet->byte_count + extra_bytes;
1647 if (ofproto->ofhooks->account_flow_cb
1648 && total_bytes > facet->accounted_bytes)
1650 ofproto->ofhooks->account_flow_cb(
1651 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1652 total_bytes - facet->accounted_bytes, ofproto->aux);
1653 facet->accounted_bytes = total_bytes;
1657 /* If 'rule' is installed in the datapath, uninstalls it. */
1659 facet_uninstall(struct ofproto *p, struct facet *facet)
1661 if (facet->installed) {
1662 struct odputil_keybuf keybuf;
1663 struct dpif_flow_stats stats;
1666 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1667 odp_flow_key_from_flow(&key, &facet->flow);
1669 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1670 facet_update_stats(p, facet, &stats);
1672 facet->installed = false;
1673 facet->dp_packet_count = 0;
1674 facet->dp_byte_count = 0;
1676 assert(facet->dp_packet_count == 0);
1677 assert(facet->dp_byte_count == 0);
1681 /* Returns true if the only action for 'facet' is to send to the controller.
1682 * (We don't report NetFlow expiration messages for such facets because they
1683 * are just part of the control logic for the network, not real traffic). */
1685 facet_is_controller_flow(struct facet *facet)
1688 && facet->rule->n_actions == 1
1689 && action_outputs_to_port(&facet->rule->actions[0],
1690 htons(OFPP_CONTROLLER)));
1693 /* Folds all of 'facet''s statistics into its rule. Also updates the
1694 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1695 * 'facet''s statistics in the datapath should have been zeroed and folded into
1696 * its packet and byte counts before this function is called. */
1698 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1700 assert(!facet->dp_byte_count);
1701 assert(!facet->dp_packet_count);
1703 facet_push_stats(ofproto, facet);
1704 facet_account(ofproto, facet, 0);
1706 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1707 struct ofexpired expired;
1708 expired.flow = facet->flow;
1709 expired.packet_count = facet->packet_count;
1710 expired.byte_count = facet->byte_count;
1711 expired.used = facet->used;
1712 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1715 facet->rule->packet_count += facet->packet_count;
1716 facet->rule->byte_count += facet->byte_count;
1718 /* Reset counters to prevent double counting if 'facet' ever gets
1720 facet->packet_count = 0;
1721 facet->byte_count = 0;
1722 facet->rs_packet_count = 0;
1723 facet->rs_byte_count = 0;
1724 facet->accounted_bytes = 0;
1726 netflow_flow_clear(&facet->nf_flow);
1729 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1730 * Returns it if found, otherwise a null pointer.
1732 * The returned facet might need revalidation; use facet_lookup_valid()
1733 * instead if that is important. */
1734 static struct facet *
1735 facet_find(struct ofproto *ofproto, const struct flow *flow)
1737 struct facet *facet;
1739 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1741 if (flow_equal(flow, &facet->flow)) {
1749 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1750 * Returns it if found, otherwise a null pointer.
1752 * The returned facet is guaranteed to be valid. */
1753 static struct facet *
1754 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1756 struct facet *facet = facet_find(ofproto, flow);
1758 /* The facet we found might not be valid, since we could be in need of
1759 * revalidation. If it is not valid, don't return it. */
1761 && ofproto->need_revalidate
1762 && !facet_revalidate(ofproto, facet)) {
1763 COVERAGE_INC(ofproto_invalidated);
1770 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1772 * - If the rule found is different from 'facet''s current rule, moves
1773 * 'facet' to the new rule and recompiles its actions.
1775 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1776 * where it is and recompiles its actions anyway.
1778 * - If there is none, destroys 'facet'.
1780 * Returns true if 'facet' still exists, false if it has been destroyed. */
1782 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1784 struct action_xlate_ctx ctx;
1785 struct ofpbuf *odp_actions;
1786 struct rule *new_rule;
1787 bool actions_changed;
1789 COVERAGE_INC(facet_revalidate);
1791 /* Determine the new rule. */
1792 new_rule = rule_lookup(ofproto, &facet->flow);
1794 /* No new rule, so delete the facet. */
1795 facet_remove(ofproto, facet);
1799 /* Calculate new ODP actions.
1801 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1802 * emit a NetFlow expiration and, if so, we need to have the old state
1803 * around to properly compose it. */
1804 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1805 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1806 actions_changed = (facet->actions_len != odp_actions->size
1807 || memcmp(facet->actions, odp_actions->data,
1808 facet->actions_len));
1810 /* If the ODP actions changed or the installability changed, then we need
1811 * to talk to the datapath. */
1812 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
1813 if (ctx.may_set_up_flow) {
1814 struct dpif_flow_stats stats;
1816 facet_put__(ofproto, facet,
1817 odp_actions->data, odp_actions->size, &stats);
1818 facet_update_stats(ofproto, facet, &stats);
1820 facet_uninstall(ofproto, facet);
1823 /* The datapath flow is gone or has zeroed stats, so push stats out of
1824 * 'facet' into 'rule'. */
1825 facet_flush_stats(ofproto, facet);
1828 /* Update 'facet' now that we've taken care of all the old state. */
1829 facet->tags = ctx.tags;
1830 facet->nf_flow.output_iface = ctx.nf_output_iface;
1831 facet->may_install = ctx.may_set_up_flow;
1832 if (actions_changed) {
1833 free(facet->actions);
1834 facet->actions_len = odp_actions->size;
1835 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1837 if (facet->rule != new_rule) {
1838 COVERAGE_INC(facet_changed_rule);
1839 list_remove(&facet->list_node);
1840 list_push_back(&new_rule->facets, &facet->list_node);
1841 facet->rule = new_rule;
1842 facet->used = new_rule->created;
1843 facet->rs_used = facet->used;
1846 ofpbuf_delete(odp_actions);
1852 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1855 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
1857 COVERAGE_INC(ofproto_error);
1858 ofconn_send_reply(ofconn, buf);
1863 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1865 ofconn_send_reply(ofconn, make_echo_reply(oh));
1870 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1872 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1873 struct ofp_switch_features *osf;
1875 struct ofport *port;
1877 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1878 osf->datapath_id = htonll(ofproto->datapath_id);
1879 osf->n_buffers = htonl(pktbuf_capacity());
1881 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1882 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1883 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1884 (1u << OFPAT_SET_VLAN_VID) |
1885 (1u << OFPAT_SET_VLAN_PCP) |
1886 (1u << OFPAT_STRIP_VLAN) |
1887 (1u << OFPAT_SET_DL_SRC) |
1888 (1u << OFPAT_SET_DL_DST) |
1889 (1u << OFPAT_SET_NW_SRC) |
1890 (1u << OFPAT_SET_NW_DST) |
1891 (1u << OFPAT_SET_NW_TOS) |
1892 (1u << OFPAT_SET_TP_SRC) |
1893 (1u << OFPAT_SET_TP_DST) |
1894 (1u << OFPAT_ENQUEUE));
1896 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1897 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1900 ofconn_send_reply(ofconn, buf);
1905 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1907 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1909 struct ofp_switch_config *osc;
1913 /* Figure out flags. */
1914 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
1915 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1918 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1919 osc->flags = htons(flags);
1920 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1921 ofconn_send_reply(ofconn, buf);
1927 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1929 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1930 uint16_t flags = ntohs(osc->flags);
1932 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1933 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1934 switch (flags & OFPC_FRAG_MASK) {
1935 case OFPC_FRAG_NORMAL:
1936 dpif_set_drop_frags(ofproto->dpif, false);
1938 case OFPC_FRAG_DROP:
1939 dpif_set_drop_frags(ofproto->dpif, true);
1942 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1948 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1953 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1954 struct action_xlate_ctx *ctx);
1957 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1959 const struct ofport *ofport = get_port(ctx->ofproto, port);
1962 if (ofport->opp.config & OFPPC_NO_FWD) {
1963 /* Forwarding disabled on port. */
1968 * We don't have an ofport record for this port, but it doesn't hurt to
1969 * allow forwarding to it anyhow. Maybe such a port will appear later
1970 * and we're pre-populating the flow table.
1974 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
1975 ctx->nf_output_iface = port;
1978 static struct rule *
1979 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
1981 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1985 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1987 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1988 uint16_t old_in_port;
1991 /* Look up a flow with 'in_port' as the input port. Then restore the
1992 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1993 * have surprising behavior). */
1994 old_in_port = ctx->flow.in_port;
1995 ctx->flow.in_port = in_port;
1996 rule = rule_lookup(ctx->ofproto, &ctx->flow);
1997 ctx->flow.in_port = old_in_port;
1999 if (ctx->resubmit_hook) {
2000 ctx->resubmit_hook(ctx, rule);
2005 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2009 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2011 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2012 MAX_RESUBMIT_RECURSION);
2017 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2018 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2020 struct ofport *ofport;
2022 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2023 uint16_t odp_port = ofport->odp_port;
2024 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2025 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2028 *nf_output_iface = NF_OUT_FLOOD;
2032 xlate_output_action__(struct action_xlate_ctx *ctx,
2033 uint16_t port, uint16_t max_len)
2036 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2038 ctx->nf_output_iface = NF_OUT_DROP;
2042 add_output_action(ctx, ctx->flow.in_port);
2045 xlate_table_action(ctx, ctx->flow.in_port);
2048 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2049 ctx->odp_actions, &ctx->tags,
2050 &ctx->nf_output_iface,
2051 ctx->ofproto->aux)) {
2052 COVERAGE_INC(ofproto_uninstallable);
2053 ctx->may_set_up_flow = false;
2057 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2058 &ctx->nf_output_iface, ctx->odp_actions);
2061 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2062 &ctx->nf_output_iface, ctx->odp_actions);
2064 case OFPP_CONTROLLER:
2065 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2068 add_output_action(ctx, ODPP_LOCAL);
2071 odp_port = ofp_port_to_odp_port(port);
2072 if (odp_port != ctx->flow.in_port) {
2073 add_output_action(ctx, odp_port);
2078 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2079 ctx->nf_output_iface = NF_OUT_FLOOD;
2080 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2081 ctx->nf_output_iface = prev_nf_output_iface;
2082 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2083 ctx->nf_output_iface != NF_OUT_FLOOD) {
2084 ctx->nf_output_iface = NF_OUT_MULTI;
2089 xlate_output_action(struct action_xlate_ctx *ctx,
2090 const struct ofp_action_output *oao)
2092 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2095 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2096 * optimization, because we're going to add another action that sets the
2097 * priority immediately after, or because there are no actions following the
2100 remove_pop_action(struct action_xlate_ctx *ctx)
2102 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2103 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2104 ctx->last_pop_priority = -1;
2109 add_pop_action(struct action_xlate_ctx *ctx)
2111 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2112 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2113 ctx->last_pop_priority = ctx->odp_actions->size;
2118 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2119 const struct ofp_action_enqueue *oae)
2121 uint16_t ofp_port, odp_port;
2125 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2128 /* Fall back to ordinary output action. */
2129 xlate_output_action__(ctx, ntohs(oae->port), 0);
2133 /* Figure out ODP output port. */
2134 ofp_port = ntohs(oae->port);
2135 if (ofp_port != OFPP_IN_PORT) {
2136 odp_port = ofp_port_to_odp_port(ofp_port);
2138 odp_port = ctx->flow.in_port;
2141 /* Add ODP actions. */
2142 remove_pop_action(ctx);
2143 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2144 add_output_action(ctx, odp_port);
2145 add_pop_action(ctx);
2147 /* Update NetFlow output port. */
2148 if (ctx->nf_output_iface == NF_OUT_DROP) {
2149 ctx->nf_output_iface = odp_port;
2150 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2151 ctx->nf_output_iface = NF_OUT_MULTI;
2156 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2157 const struct nx_action_set_queue *nasq)
2162 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2165 /* Couldn't translate queue to a priority, so ignore. A warning
2166 * has already been logged. */
2170 remove_pop_action(ctx);
2171 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2175 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2177 ovs_be16 tci = ctx->flow.vlan_tci;
2178 if (!(tci & htons(VLAN_CFI))) {
2179 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2181 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2182 tci & ~htons(VLAN_CFI));
2186 struct xlate_reg_state {
2192 save_reg_state(const struct action_xlate_ctx *ctx,
2193 struct xlate_reg_state *state)
2195 state->vlan_tci = ctx->flow.vlan_tci;
2196 state->tun_id = ctx->flow.tun_id;
2200 update_reg_state(struct action_xlate_ctx *ctx,
2201 const struct xlate_reg_state *state)
2203 if (ctx->flow.vlan_tci != state->vlan_tci) {
2204 xlate_set_dl_tci(ctx);
2206 if (ctx->flow.tun_id != state->tun_id) {
2207 nl_msg_put_be64(ctx->odp_actions,
2208 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2213 xlate_nicira_action(struct action_xlate_ctx *ctx,
2214 const struct nx_action_header *nah)
2216 const struct nx_action_resubmit *nar;
2217 const struct nx_action_set_tunnel *nast;
2218 const struct nx_action_set_queue *nasq;
2219 const struct nx_action_multipath *nam;
2220 enum nx_action_subtype subtype = ntohs(nah->subtype);
2221 struct xlate_reg_state state;
2224 assert(nah->vendor == htonl(NX_VENDOR_ID));
2226 case NXAST_RESUBMIT:
2227 nar = (const struct nx_action_resubmit *) nah;
2228 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2231 case NXAST_SET_TUNNEL:
2232 nast = (const struct nx_action_set_tunnel *) nah;
2233 tun_id = htonll(ntohl(nast->tun_id));
2234 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2235 ctx->flow.tun_id = tun_id;
2238 case NXAST_DROP_SPOOFED_ARP:
2239 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2240 nl_msg_put_flag(ctx->odp_actions,
2241 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2245 case NXAST_SET_QUEUE:
2246 nasq = (const struct nx_action_set_queue *) nah;
2247 xlate_set_queue_action(ctx, nasq);
2250 case NXAST_POP_QUEUE:
2251 add_pop_action(ctx);
2254 case NXAST_REG_MOVE:
2255 save_reg_state(ctx, &state);
2256 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2258 update_reg_state(ctx, &state);
2261 case NXAST_REG_LOAD:
2262 save_reg_state(ctx, &state);
2263 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2265 update_reg_state(ctx, &state);
2269 /* Nothing to do. */
2272 case NXAST_SET_TUNNEL64:
2273 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2274 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2275 ctx->flow.tun_id = tun_id;
2278 case NXAST_MULTIPATH:
2279 nam = (const struct nx_action_multipath *) nah;
2280 multipath_execute(nam, &ctx->flow);
2283 /* If you add a new action here that modifies flow data, don't forget to
2284 * update the flow key in ctx->flow at the same time. */
2286 case NXAST_SNAT__OBSOLETE:
2288 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2294 do_xlate_actions(const union ofp_action *in, size_t n_in,
2295 struct action_xlate_ctx *ctx)
2297 struct actions_iterator iter;
2298 const union ofp_action *ia;
2299 const struct ofport *port;
2301 port = get_port(ctx->ofproto, ctx->flow.in_port);
2302 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2303 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2304 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2305 /* Drop this flow. */
2309 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2310 enum ofp_action_type type = ntohs(ia->type);
2311 const struct ofp_action_dl_addr *oada;
2315 xlate_output_action(ctx, &ia->output);
2318 case OFPAT_SET_VLAN_VID:
2319 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2320 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2321 xlate_set_dl_tci(ctx);
2324 case OFPAT_SET_VLAN_PCP:
2325 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2326 ctx->flow.vlan_tci |= htons(
2327 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2328 xlate_set_dl_tci(ctx);
2331 case OFPAT_STRIP_VLAN:
2332 ctx->flow.vlan_tci = htons(0);
2333 xlate_set_dl_tci(ctx);
2336 case OFPAT_SET_DL_SRC:
2337 oada = ((struct ofp_action_dl_addr *) ia);
2338 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2339 oada->dl_addr, ETH_ADDR_LEN);
2340 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2343 case OFPAT_SET_DL_DST:
2344 oada = ((struct ofp_action_dl_addr *) ia);
2345 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2346 oada->dl_addr, ETH_ADDR_LEN);
2347 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2350 case OFPAT_SET_NW_SRC:
2351 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2352 ia->nw_addr.nw_addr);
2353 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2356 case OFPAT_SET_NW_DST:
2357 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2358 ia->nw_addr.nw_addr);
2359 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2362 case OFPAT_SET_NW_TOS:
2363 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2365 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2368 case OFPAT_SET_TP_SRC:
2369 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2370 ia->tp_port.tp_port);
2371 ctx->flow.tp_src = ia->tp_port.tp_port;
2374 case OFPAT_SET_TP_DST:
2375 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2376 ia->tp_port.tp_port);
2377 ctx->flow.tp_dst = ia->tp_port.tp_port;
2381 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2385 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2389 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2396 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2397 struct ofproto *ofproto, const struct flow *flow,
2398 const struct ofpbuf *packet)
2400 ctx->ofproto = ofproto;
2402 ctx->packet = packet;
2403 ctx->resubmit_hook = NULL;
2404 ctx->check_special = true;
2408 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2409 const struct ofpbuf *packet)
2411 struct ofport *ofport;
2413 ofport = get_port(ofproto, flow->in_port);
2414 if (ofport && ofport->cfm) {
2415 cfm_process_heartbeat(ofport->cfm, packet);
2419 static struct ofpbuf *
2420 xlate_actions(struct action_xlate_ctx *ctx,
2421 const union ofp_action *in, size_t n_in)
2423 COVERAGE_INC(ofproto_ofp2odp);
2425 ctx->odp_actions = ofpbuf_new(512);
2427 ctx->may_set_up_flow = true;
2428 ctx->nf_output_iface = NF_OUT_DROP;
2430 ctx->last_pop_priority = -1;
2432 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2434 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2436 ctx->may_set_up_flow = false;
2437 } else if (ctx->check_special
2438 && ctx->ofproto->ofhooks->special_cb
2439 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2440 ctx->ofproto->aux)) {
2441 ctx->may_set_up_flow = false;
2443 do_xlate_actions(in, n_in, ctx);
2446 remove_pop_action(ctx);
2448 /* Check with in-band control to see if we're allowed to set up this
2450 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2451 ctx->odp_actions->data,
2452 ctx->odp_actions->size)) {
2453 ctx->may_set_up_flow = false;
2456 return ctx->odp_actions;
2459 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2460 * error message code (composed with ofp_mkerr()) for the caller to propagate
2461 * upward. Otherwise, returns 0.
2463 * The log message mentions 'msg_type'. */
2465 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2467 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2468 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2469 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2470 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2473 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2480 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2482 struct ofproto *p = ofconn_get_ofproto(ofconn);
2483 struct ofp_packet_out *opo;
2484 struct ofpbuf payload, *buffer;
2485 union ofp_action *ofp_actions;
2486 struct action_xlate_ctx ctx;
2487 struct ofpbuf *odp_actions;
2488 struct ofpbuf request;
2490 size_t n_ofp_actions;
2494 COVERAGE_INC(ofproto_packet_out);
2496 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2501 /* Get ofp_packet_out. */
2502 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2503 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2506 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2507 &ofp_actions, &n_ofp_actions);
2513 if (opo->buffer_id != htonl(UINT32_MAX)) {
2514 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2516 if (error || !buffer) {
2525 /* Extract flow, check actions. */
2526 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2528 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2534 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2535 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2536 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2537 ofpbuf_delete(odp_actions);
2540 ofpbuf_delete(buffer);
2545 update_port_config(struct ofproto *p, struct ofport *port,
2546 uint32_t config, uint32_t mask)
2548 mask &= config ^ port->opp.config;
2549 if (mask & OFPPC_PORT_DOWN) {
2550 if (config & OFPPC_PORT_DOWN) {
2551 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2553 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2556 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2557 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2558 if (mask & REVALIDATE_BITS) {
2559 COVERAGE_INC(ofproto_costly_flags);
2560 port->opp.config ^= mask & REVALIDATE_BITS;
2561 p->need_revalidate = true;
2563 #undef REVALIDATE_BITS
2564 if (mask & OFPPC_NO_PACKET_IN) {
2565 port->opp.config ^= OFPPC_NO_PACKET_IN;
2570 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2572 struct ofproto *p = ofconn_get_ofproto(ofconn);
2573 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2574 struct ofport *port;
2577 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2582 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2584 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2585 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2586 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2588 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2589 if (opm->advertise) {
2590 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2596 static struct ofpbuf *
2597 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2599 struct ofp_stats_reply *osr;
2602 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2603 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2605 osr->flags = htons(0);
2609 static struct ofpbuf *
2610 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2612 const struct ofp_stats_request *osr
2613 = (const struct ofp_stats_request *) request;
2614 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2618 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2619 struct ofpbuf **msgp)
2621 struct ofpbuf *msg = *msgp;
2622 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2623 if (nbytes + msg->size > UINT16_MAX) {
2624 struct ofp_stats_reply *reply = msg->data;
2625 reply->flags = htons(OFPSF_REPLY_MORE);
2626 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2627 ofconn_send_reply(ofconn, msg);
2629 return ofpbuf_put_uninit(*msgp, nbytes);
2632 static struct ofpbuf *
2633 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2635 struct nicira_stats_msg *nsm;
2638 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2639 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2640 nsm->type = htons(OFPST_VENDOR);
2641 nsm->flags = htons(0);
2642 nsm->vendor = htonl(NX_VENDOR_ID);
2643 nsm->subtype = subtype;
2647 static struct ofpbuf *
2648 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2650 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2654 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2655 struct ofpbuf **msgp)
2657 struct ofpbuf *msg = *msgp;
2658 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2659 if (nbytes + msg->size > UINT16_MAX) {
2660 struct nicira_stats_msg *reply = msg->data;
2661 reply->flags = htons(OFPSF_REPLY_MORE);
2662 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2663 ofconn_send_reply(ofconn, msg);
2665 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2669 handle_desc_stats_request(struct ofconn *ofconn,
2670 const struct ofp_header *request)
2672 struct ofproto *p = ofconn_get_ofproto(ofconn);
2673 struct ofp_desc_stats *ods;
2676 msg = start_ofp_stats_reply(request, sizeof *ods);
2677 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2678 memset(ods, 0, sizeof *ods);
2679 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2680 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2681 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2682 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2683 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2684 ofconn_send_reply(ofconn, msg);
2690 handle_table_stats_request(struct ofconn *ofconn,
2691 const struct ofp_header *request)
2693 struct ofproto *p = ofconn_get_ofproto(ofconn);
2694 struct ofp_table_stats *ots;
2697 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2699 /* Classifier table. */
2700 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2701 memset(ots, 0, sizeof *ots);
2702 strcpy(ots->name, "classifier");
2703 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2704 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2705 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2706 ots->active_count = htonl(classifier_count(&p->cls));
2707 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2708 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2710 ofconn_send_reply(ofconn, msg);
2715 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2716 struct ofpbuf **msgp)
2718 struct netdev_stats stats;
2719 struct ofp_port_stats *ops;
2721 /* Intentionally ignore return value, since errors will set
2722 * 'stats' to all-1s, which is correct for OpenFlow, and
2723 * netdev_get_stats() will log errors. */
2724 netdev_get_stats(port->netdev, &stats);
2726 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2727 ops->port_no = htons(port->opp.port_no);
2728 memset(ops->pad, 0, sizeof ops->pad);
2729 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2730 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2731 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2732 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2733 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2734 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2735 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2736 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2737 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2738 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2739 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2740 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2744 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2746 struct ofproto *p = ofconn_get_ofproto(ofconn);
2747 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2748 struct ofp_port_stats *ops;
2750 struct ofport *port;
2752 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2753 if (psr->port_no != htons(OFPP_NONE)) {
2754 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2756 append_port_stat(port, ofconn, &msg);
2759 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2760 append_port_stat(port, ofconn, &msg);
2764 ofconn_send_reply(ofconn, msg);
2769 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2771 long long int msecs = time_msec() - start;
2772 *sec = msecs / 1000;
2773 *nsec = (msecs % 1000) * (1000 * 1000);
2777 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2781 calc_flow_duration__(start, &sec, &nsec);
2782 *sec_be = htonl(sec);
2783 *nsec_be = htonl(nsec);
2787 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2788 ovs_be16 out_port, struct ofpbuf **replyp)
2790 struct ofp_flow_stats *ofs;
2791 uint64_t packet_count, byte_count;
2793 size_t act_len, len;
2795 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2799 act_len = sizeof *rule->actions * rule->n_actions;
2800 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2802 rule_get_stats(rule, &packet_count, &byte_count);
2804 ofs = append_ofp_stats_reply(len, ofconn, replyp);
2805 ofs->length = htons(len);
2808 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
2809 &ofs->match, rule->flow_cookie, &cookie);
2810 put_32aligned_be64(&ofs->cookie, cookie);
2811 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
2812 ofs->priority = htons(rule->cr.priority);
2813 ofs->idle_timeout = htons(rule->idle_timeout);
2814 ofs->hard_timeout = htons(rule->hard_timeout);
2815 memset(ofs->pad2, 0, sizeof ofs->pad2);
2816 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
2817 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
2818 if (rule->n_actions > 0) {
2819 memcpy(ofs->actions, rule->actions, act_len);
2824 is_valid_table(uint8_t table_id)
2826 if (table_id == 0 || table_id == 0xff) {
2829 /* It would probably be better to reply with an error but there doesn't
2830 * seem to be any appropriate value, so that might just be
2832 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
2839 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2841 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
2842 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2843 struct ofpbuf *reply;
2845 COVERAGE_INC(ofproto_flows_req);
2846 reply = start_ofp_stats_reply(oh, 1024);
2847 if (is_valid_table(fsr->table_id)) {
2848 struct cls_cursor cursor;
2849 struct cls_rule target;
2852 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
2854 cls_cursor_init(&cursor, &ofproto->cls, &target);
2855 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2856 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
2859 ofconn_send_reply(ofconn, reply);
2865 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
2866 ovs_be16 out_port, struct ofpbuf **replyp)
2868 struct nx_flow_stats *nfs;
2869 uint64_t packet_count, byte_count;
2870 size_t act_len, start_len;
2871 struct ofpbuf *reply;
2873 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2877 rule_get_stats(rule, &packet_count, &byte_count);
2879 act_len = sizeof *rule->actions * rule->n_actions;
2881 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
2882 start_len = (*replyp)->size;
2885 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
2888 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
2889 nfs->cookie = rule->flow_cookie;
2890 nfs->priority = htons(rule->cr.priority);
2891 nfs->idle_timeout = htons(rule->idle_timeout);
2892 nfs->hard_timeout = htons(rule->hard_timeout);
2893 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
2894 memset(nfs->pad2, 0, sizeof nfs->pad2);
2895 nfs->packet_count = htonll(packet_count);
2896 nfs->byte_count = htonll(byte_count);
2897 if (rule->n_actions > 0) {
2898 ofpbuf_put(reply, rule->actions, act_len);
2900 nfs->length = htons(reply->size - start_len);
2904 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
2906 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2907 struct nx_flow_stats_request *nfsr;
2908 struct cls_rule target;
2909 struct ofpbuf *reply;
2913 ofpbuf_use_const(&b, oh, ntohs(oh->length));
2915 /* Dissect the message. */
2916 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
2917 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
2922 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2925 COVERAGE_INC(ofproto_flows_req);
2926 reply = start_nxstats_reply(&nfsr->nsm, 1024);
2927 if (is_valid_table(nfsr->table_id)) {
2928 struct cls_cursor cursor;
2931 cls_cursor_init(&cursor, &ofproto->cls, &target);
2932 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2933 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
2936 ofconn_send_reply(ofconn, reply);
2942 flow_stats_ds(struct rule *rule, struct ds *results)
2944 uint64_t packet_count, byte_count;
2945 size_t act_len = sizeof *rule->actions * rule->n_actions;
2947 rule_get_stats(rule, &packet_count, &byte_count);
2949 ds_put_format(results, "duration=%llds, ",
2950 (time_msec() - rule->created) / 1000);
2951 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
2952 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2953 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2954 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2955 cls_rule_format(&rule->cr, results);
2956 ds_put_char(results, ',');
2958 ofp_print_actions(results, &rule->actions->header, act_len);
2960 ds_put_cstr(results, "drop");
2962 ds_put_cstr(results, "\n");
2965 /* Adds a pretty-printed description of all flows to 'results', including
2966 * hidden flows (e.g., set up by in-band control). */
2968 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2970 struct cls_cursor cursor;
2973 cls_cursor_init(&cursor, &p->cls, NULL);
2974 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2975 flow_stats_ds(rule, results);
2980 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
2981 ovs_be16 out_port, uint8_t table_id,
2982 struct ofp_aggregate_stats_reply *oasr)
2984 uint64_t total_packets = 0;
2985 uint64_t total_bytes = 0;
2988 COVERAGE_INC(ofproto_agg_request);
2990 if (is_valid_table(table_id)) {
2991 struct cls_cursor cursor;
2994 cls_cursor_init(&cursor, &ofproto->cls, target);
2995 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2996 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
2997 uint64_t packet_count;
2998 uint64_t byte_count;
3000 rule_get_stats(rule, &packet_count, &byte_count);
3002 total_packets += packet_count;
3003 total_bytes += byte_count;
3009 oasr->flow_count = htonl(n_flows);
3010 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3011 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3012 memset(oasr->pad, 0, sizeof oasr->pad);
3016 handle_aggregate_stats_request(struct ofconn *ofconn,
3017 const struct ofp_header *oh)
3019 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3020 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3021 struct ofp_aggregate_stats_reply *reply;
3022 struct cls_rule target;
3025 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3028 msg = start_ofp_stats_reply(oh, sizeof *reply);
3029 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3030 query_aggregate_stats(ofproto, &target, request->out_port,
3031 request->table_id, reply);
3032 ofconn_send_reply(ofconn, msg);
3037 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3039 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3040 struct nx_aggregate_stats_request *request;
3041 struct ofp_aggregate_stats_reply *reply;
3042 struct cls_rule target;
3047 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3049 /* Dissect the message. */
3050 request = ofpbuf_pull(&b, sizeof *request);
3051 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3056 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3060 COVERAGE_INC(ofproto_flows_req);
3061 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3062 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3063 query_aggregate_stats(ofproto, &target, request->out_port,
3064 request->table_id, reply);
3065 ofconn_send_reply(ofconn, buf);
3070 struct queue_stats_cbdata {
3071 struct ofconn *ofconn;
3072 struct ofport *ofport;
3077 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3078 const struct netdev_queue_stats *stats)
3080 struct ofp_queue_stats *reply;
3082 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3083 reply->port_no = htons(cbdata->ofport->opp.port_no);
3084 memset(reply->pad, 0, sizeof reply->pad);
3085 reply->queue_id = htonl(queue_id);
3086 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3087 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3088 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3092 handle_queue_stats_dump_cb(uint32_t queue_id,
3093 struct netdev_queue_stats *stats,
3096 struct queue_stats_cbdata *cbdata = cbdata_;
3098 put_queue_stats(cbdata, queue_id, stats);
3102 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3103 struct queue_stats_cbdata *cbdata)
3105 cbdata->ofport = port;
3106 if (queue_id == OFPQ_ALL) {
3107 netdev_dump_queue_stats(port->netdev,
3108 handle_queue_stats_dump_cb, cbdata);
3110 struct netdev_queue_stats stats;
3112 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3113 put_queue_stats(cbdata, queue_id, &stats);
3119 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3121 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3122 const struct ofp_queue_stats_request *qsr;
3123 struct queue_stats_cbdata cbdata;
3124 struct ofport *port;
3125 unsigned int port_no;
3128 qsr = ofputil_stats_body(oh);
3130 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3133 COVERAGE_INC(ofproto_queue_req);
3135 cbdata.ofconn = ofconn;
3136 cbdata.msg = start_ofp_stats_reply(oh, 128);
3138 port_no = ntohs(qsr->port_no);
3139 queue_id = ntohl(qsr->queue_id);
3140 if (port_no == OFPP_ALL) {
3141 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3142 handle_queue_stats_for_port(port, queue_id, &cbdata);
3144 } else if (port_no < ofproto->max_ports) {
3145 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3147 handle_queue_stats_for_port(port, queue_id, &cbdata);
3150 ofpbuf_delete(cbdata.msg);
3151 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3153 ofconn_send_reply(ofconn, cbdata.msg);
3158 /* Updates 'facet''s used time. Caller is responsible for calling
3159 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3161 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3164 if (used > facet->used) {
3166 if (used > facet->rule->used) {
3167 facet->rule->used = used;
3169 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3173 /* Folds the statistics from 'stats' into the counters in 'facet'.
3175 * Because of the meaning of a facet's counters, it only makes sense to do this
3176 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3177 * packet that was sent by hand or if it represents statistics that have been
3178 * cleared out of the datapath. */
3180 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3181 const struct dpif_flow_stats *stats)
3183 if (stats->n_packets || stats->used > facet->used) {
3184 facet_update_time(ofproto, facet, stats->used);
3185 facet->packet_count += stats->n_packets;
3186 facet->byte_count += stats->n_bytes;
3187 facet_push_stats(ofproto, facet);
3188 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3193 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3195 uint64_t rs_packets, rs_bytes;
3197 assert(facet->packet_count >= facet->rs_packet_count);
3198 assert(facet->byte_count >= facet->rs_byte_count);
3199 assert(facet->used >= facet->rs_used);
3201 rs_packets = facet->packet_count - facet->rs_packet_count;
3202 rs_bytes = facet->byte_count - facet->rs_byte_count;
3204 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3205 facet->rs_packet_count = facet->packet_count;
3206 facet->rs_byte_count = facet->byte_count;
3207 facet->rs_used = facet->used;
3209 flow_push_stats(ofproto, facet->rule, &facet->flow,
3210 rs_packets, rs_bytes, facet->used);
3214 struct ofproto_push {
3215 struct action_xlate_ctx ctx;
3222 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3224 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3227 rule->packet_count += push->packets;
3228 rule->byte_count += push->bytes;
3229 rule->used = MAX(push->used, rule->used);
3233 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3234 * 'rule''s actions. */
3236 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3237 struct flow *flow, uint64_t packets, uint64_t bytes,
3240 struct ofproto_push push;
3242 push.packets = packets;
3246 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3247 push.ctx.resubmit_hook = push_resubmit;
3248 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3251 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3252 * in which no matching flow already exists in the flow table.
3254 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3255 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3256 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3258 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3261 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3263 struct ofproto *p = ofconn_get_ofproto(ofconn);
3264 struct ofpbuf *packet;
3269 if (fm->flags & OFPFF_CHECK_OVERLAP
3270 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3271 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3275 if (fm->buffer_id != UINT32_MAX) {
3276 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3280 in_port = UINT16_MAX;
3283 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3284 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3285 fm->flags & OFPFF_SEND_FLOW_REM);
3286 rule_insert(p, rule);
3288 rule_execute(p, rule, in_port, packet);
3293 static struct rule *
3294 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3296 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3300 send_buffered_packet(struct ofconn *ofconn,
3301 struct rule *rule, uint32_t buffer_id)
3303 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3304 struct ofpbuf *packet;
3308 if (buffer_id == UINT32_MAX) {
3312 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3317 rule_execute(ofproto, rule, in_port, packet);
3322 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3324 struct modify_flows_cbdata {
3325 struct ofproto *ofproto;
3326 const struct flow_mod *fm;
3330 static int modify_flow(struct ofproto *, const struct flow_mod *,
3333 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3334 * encoded by ofp_mkerr() on failure.
3336 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3339 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3341 struct ofproto *p = ofconn_get_ofproto(ofconn);
3342 struct rule *match = NULL;
3343 struct cls_cursor cursor;
3346 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3347 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3348 if (!rule_is_hidden(rule)) {
3350 modify_flow(p, fm, rule);
3355 /* This credits the packet to whichever flow happened to match last.
3356 * That's weird. Maybe we should do a lookup for the flow that
3357 * actually matches the packet? Who knows. */
3358 send_buffered_packet(ofconn, match, fm->buffer_id);
3361 return add_flow(ofconn, fm);
3365 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3366 * code as encoded by ofp_mkerr() on failure.
3368 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3371 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3373 struct ofproto *p = ofconn_get_ofproto(ofconn);
3374 struct rule *rule = find_flow_strict(p, fm);
3375 if (rule && !rule_is_hidden(rule)) {
3376 modify_flow(p, fm, rule);
3377 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3379 return add_flow(ofconn, fm);
3383 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3384 * been identified as a flow in 'p''s flow table to be modified, by changing
3385 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3386 * ofp_action[] structures). */
3388 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3390 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3392 rule->flow_cookie = fm->cookie;
3394 /* If the actions are the same, do nothing. */
3395 if (fm->n_actions == rule->n_actions
3397 || !memcmp(fm->actions, rule->actions, actions_len))) {
3401 /* Replace actions. */
3402 free(rule->actions);
3403 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3404 rule->n_actions = fm->n_actions;
3406 p->need_revalidate = true;
3411 /* OFPFC_DELETE implementation. */
3413 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3415 /* Implements OFPFC_DELETE. */
3417 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3419 struct rule *rule, *next_rule;
3420 struct cls_cursor cursor;
3422 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3423 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3424 delete_flow(p, rule, htons(fm->out_port));
3428 /* Implements OFPFC_DELETE_STRICT. */
3430 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3432 struct rule *rule = find_flow_strict(p, fm);
3434 delete_flow(p, rule, htons(fm->out_port));
3438 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3439 * been identified as a flow to delete from 'p''s flow table, by deleting the
3440 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3443 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3444 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3445 * specified 'out_port'. */
3447 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3449 if (rule_is_hidden(rule)) {
3453 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3457 rule_send_removed(p, rule, OFPRR_DELETE);
3458 rule_remove(p, rule);
3462 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3464 struct ofproto *p = ofconn_get_ofproto(ofconn);
3468 error = reject_slave_controller(ofconn, "flow_mod");
3473 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3478 /* We do not support the emergency flow cache. It will hopefully get
3479 * dropped from OpenFlow in the near future. */
3480 if (fm.flags & OFPFF_EMERG) {
3481 /* There isn't a good fit for an error code, so just state that the
3482 * flow table is full. */
3483 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3486 error = validate_actions(fm.actions, fm.n_actions,
3487 &fm.cr.flow, p->max_ports);
3492 switch (fm.command) {
3494 return add_flow(ofconn, &fm);
3497 return modify_flows_loose(ofconn, &fm);
3499 case OFPFC_MODIFY_STRICT:
3500 return modify_flow_strict(ofconn, &fm);
3503 delete_flows_loose(p, &fm);
3506 case OFPFC_DELETE_STRICT:
3507 delete_flow_strict(p, &fm);
3511 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3516 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3518 const struct nxt_tun_id_cookie *msg
3519 = (const struct nxt_tun_id_cookie *) oh;
3520 enum nx_flow_format flow_format;
3522 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3523 ofconn_set_flow_format(ofconn, flow_format);
3529 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3531 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3532 struct nx_role_request *reply;
3536 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3537 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3538 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3541 role = ntohl(nrr->role);
3542 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3543 && role != NX_ROLE_SLAVE) {
3544 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3546 /* There's no good error code for this. */
3547 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3550 ofconn_set_role(ofconn, role);
3552 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3553 reply->role = htonl(role);
3554 ofconn_send_reply(ofconn, buf);
3560 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3562 const struct nxt_set_flow_format *msg
3563 = (const struct nxt_set_flow_format *) oh;
3566 format = ntohl(msg->format);
3567 if (format == NXFF_OPENFLOW10
3568 || format == NXFF_TUN_ID_FROM_COOKIE
3569 || format == NXFF_NXM) {
3570 ofconn_set_flow_format(ofconn, format);
3573 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3578 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3580 struct ofp_header *ob;
3583 /* Currently, everything executes synchronously, so we can just
3584 * immediately send the barrier reply. */
3585 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3586 ofconn_send_reply(ofconn, buf);
3591 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3593 const struct ofp_header *oh = msg->data;
3594 const struct ofputil_msg_type *type;
3597 error = ofputil_decode_msg_type(oh, &type);
3602 switch (ofputil_msg_type_code(type)) {
3603 /* OpenFlow requests. */
3604 case OFPUTIL_OFPT_ECHO_REQUEST:
3605 return handle_echo_request(ofconn, oh);
3607 case OFPUTIL_OFPT_FEATURES_REQUEST:
3608 return handle_features_request(ofconn, oh);
3610 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3611 return handle_get_config_request(ofconn, oh);
3613 case OFPUTIL_OFPT_SET_CONFIG:
3614 return handle_set_config(ofconn, msg->data);
3616 case OFPUTIL_OFPT_PACKET_OUT:
3617 return handle_packet_out(ofconn, oh);
3619 case OFPUTIL_OFPT_PORT_MOD:
3620 return handle_port_mod(ofconn, oh);
3622 case OFPUTIL_OFPT_FLOW_MOD:
3623 return handle_flow_mod(ofconn, oh);
3625 case OFPUTIL_OFPT_BARRIER_REQUEST:
3626 return handle_barrier_request(ofconn, oh);
3628 /* OpenFlow replies. */
3629 case OFPUTIL_OFPT_ECHO_REPLY:
3632 /* Nicira extension requests. */
3633 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3634 return handle_tun_id_from_cookie(ofconn, oh);
3636 case OFPUTIL_NXT_ROLE_REQUEST:
3637 return handle_role_request(ofconn, oh);
3639 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3640 return handle_nxt_set_flow_format(ofconn, oh);
3642 case OFPUTIL_NXT_FLOW_MOD:
3643 return handle_flow_mod(ofconn, oh);
3645 /* OpenFlow statistics requests. */
3646 case OFPUTIL_OFPST_DESC_REQUEST:
3647 return handle_desc_stats_request(ofconn, oh);
3649 case OFPUTIL_OFPST_FLOW_REQUEST:
3650 return handle_flow_stats_request(ofconn, oh);
3652 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3653 return handle_aggregate_stats_request(ofconn, oh);
3655 case OFPUTIL_OFPST_TABLE_REQUEST:
3656 return handle_table_stats_request(ofconn, oh);
3658 case OFPUTIL_OFPST_PORT_REQUEST:
3659 return handle_port_stats_request(ofconn, oh);
3661 case OFPUTIL_OFPST_QUEUE_REQUEST:
3662 return handle_queue_stats_request(ofconn, oh);
3664 /* Nicira extension statistics requests. */
3665 case OFPUTIL_NXST_FLOW_REQUEST:
3666 return handle_nxst_flow(ofconn, oh);
3668 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3669 return handle_nxst_aggregate(ofconn, oh);
3671 case OFPUTIL_INVALID:
3672 case OFPUTIL_OFPT_HELLO:
3673 case OFPUTIL_OFPT_ERROR:
3674 case OFPUTIL_OFPT_FEATURES_REPLY:
3675 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3676 case OFPUTIL_OFPT_PACKET_IN:
3677 case OFPUTIL_OFPT_FLOW_REMOVED:
3678 case OFPUTIL_OFPT_PORT_STATUS:
3679 case OFPUTIL_OFPT_BARRIER_REPLY:
3680 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3681 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3682 case OFPUTIL_OFPST_DESC_REPLY:
3683 case OFPUTIL_OFPST_FLOW_REPLY:
3684 case OFPUTIL_OFPST_QUEUE_REPLY:
3685 case OFPUTIL_OFPST_PORT_REPLY:
3686 case OFPUTIL_OFPST_TABLE_REPLY:
3687 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3688 case OFPUTIL_NXT_ROLE_REPLY:
3689 case OFPUTIL_NXT_FLOW_REMOVED:
3690 case OFPUTIL_NXST_FLOW_REPLY:
3691 case OFPUTIL_NXST_AGGREGATE_REPLY:
3693 if (VLOG_IS_WARN_ENABLED()) {
3694 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3695 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3698 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3699 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3701 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3707 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3709 int error = handle_openflow__(ofconn, ofp_msg);
3711 send_error_oh(ofconn, ofp_msg->data, error);
3713 COVERAGE_INC(ofproto_recv_openflow);
3717 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3719 struct facet *facet;
3722 /* Obtain in_port and tun_id, at least. */
3723 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3725 /* Set header pointers in 'flow'. */
3726 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3728 if (cfm_should_process_flow(&flow)) {
3729 ofproto_process_cfm(p, &flow, upcall->packet);
3730 ofpbuf_delete(upcall->packet);
3732 } else if (p->ofhooks->special_cb
3733 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3734 ofpbuf_delete(upcall->packet);
3738 /* Check with in-band control to see if this packet should be sent
3739 * to the local port regardless of the flow table. */
3740 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3741 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3744 facet = facet_lookup_valid(p, &flow);
3746 struct rule *rule = rule_lookup(p, &flow);
3748 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3749 struct ofport *port = get_port(p, flow.in_port);
3751 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3752 COVERAGE_INC(ofproto_no_packet_in);
3753 /* XXX install 'drop' flow entry */
3754 ofpbuf_delete(upcall->packet);
3758 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3762 COVERAGE_INC(ofproto_packet_in);
3763 send_packet_in(p, upcall, &flow, false);
3767 facet = facet_create(p, rule, &flow, upcall->packet);
3768 } else if (!facet->may_install) {
3769 /* The facet is not installable, that is, we need to process every
3770 * packet, so process the current packet's actions into 'facet'. */
3771 facet_make_actions(p, facet, upcall->packet);
3774 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3776 * Extra-special case for fail-open mode.
3778 * We are in fail-open mode and the packet matched the fail-open rule,
3779 * but we are connected to a controller too. We should send the packet
3780 * up to the controller in the hope that it will try to set up a flow
3781 * and thereby allow us to exit fail-open.
3783 * See the top-level comment in fail-open.c for more information.
3785 send_packet_in(p, upcall, &flow, true);
3788 facet_execute(p, facet, upcall->packet);
3789 facet_install(p, facet, false);
3793 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3797 switch (upcall->type) {
3798 case DPIF_UC_ACTION:
3799 COVERAGE_INC(ofproto_ctlr_action);
3800 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3801 send_packet_in(p, upcall, &flow, false);
3804 case DPIF_UC_SAMPLE:
3806 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3807 ofproto_sflow_received(p->sflow, upcall, &flow);
3809 ofpbuf_delete(upcall->packet);
3813 handle_miss_upcall(p, upcall);
3816 case DPIF_N_UC_TYPES:
3818 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3823 /* Flow expiration. */
3825 static int ofproto_dp_max_idle(const struct ofproto *);
3826 static void ofproto_update_stats(struct ofproto *);
3827 static void rule_expire(struct ofproto *, struct rule *);
3828 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
3830 /* This function is called periodically by ofproto_run(). Its job is to
3831 * collect updates for the flows that have been installed into the datapath,
3832 * most importantly when they last were used, and then use that information to
3833 * expire flows that have not been used recently.
3835 * Returns the number of milliseconds after which it should be called again. */
3837 ofproto_expire(struct ofproto *ofproto)
3839 struct rule *rule, *next_rule;
3840 struct cls_cursor cursor;
3843 /* Update stats for each flow in the datapath. */
3844 ofproto_update_stats(ofproto);
3846 /* Expire facets that have been idle too long. */
3847 dp_max_idle = ofproto_dp_max_idle(ofproto);
3848 ofproto_expire_facets(ofproto, dp_max_idle);
3850 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3851 cls_cursor_init(&cursor, &ofproto->cls, NULL);
3852 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3853 rule_expire(ofproto, rule);
3856 /* Let the hook know that we're at a stable point: all outstanding data
3857 * in existing flows has been accounted to the account_cb. Thus, the
3858 * hook can now reasonably do operations that depend on having accurate
3859 * flow volume accounting (currently, that's just bond rebalancing). */
3860 if (ofproto->ofhooks->account_checkpoint_cb) {
3861 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
3864 return MIN(dp_max_idle, 1000);
3867 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3869 * This function also pushes statistics updates to rules which each facet
3870 * resubmits into. Generally these statistics will be accurate. However, if a
3871 * facet changes the rule it resubmits into at some time in between
3872 * ofproto_update_stats() runs, it is possible that statistics accrued to the
3873 * old rule will be incorrectly attributed to the new rule. This could be
3874 * avoided by calling ofproto_update_stats() whenever rules are created or
3875 * deleted. However, the performance impact of making so many calls to the
3876 * datapath do not justify the benefit of having perfectly accurate statistics.
3879 ofproto_update_stats(struct ofproto *p)
3881 const struct dpif_flow_stats *stats;
3882 struct dpif_flow_dump dump;
3883 const struct nlattr *key;
3886 dpif_flow_dump_start(&dump, p->dpif);
3887 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3888 struct facet *facet;
3891 if (odp_flow_key_to_flow(key, key_len, &flow)) {
3895 odp_flow_key_format(key, key_len, &s);
3896 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
3902 facet = facet_find(p, &flow);
3904 if (facet && facet->installed) {
3906 if (stats->n_packets >= facet->dp_packet_count) {
3907 facet->packet_count += stats->n_packets - facet->dp_packet_count;
3909 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3912 if (stats->n_bytes >= facet->dp_byte_count) {
3913 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
3915 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3918 facet->dp_packet_count = stats->n_packets;
3919 facet->dp_byte_count = stats->n_bytes;
3921 facet_update_time(p, facet, stats->used);
3922 facet_account(p, facet, stats->n_bytes);
3923 facet_push_stats(p, facet);
3925 /* There's a flow in the datapath that we know nothing about.
3927 COVERAGE_INC(ofproto_unexpected_rule);
3928 dpif_flow_del(p->dpif, key, key_len, NULL);
3931 dpif_flow_dump_done(&dump);
3934 /* Calculates and returns the number of milliseconds of idle time after which
3935 * facets should expire from the datapath and we should fold their statistics
3936 * into their parent rules in userspace. */
3938 ofproto_dp_max_idle(const struct ofproto *ofproto)
3941 * Idle time histogram.
3943 * Most of the time a switch has a relatively small number of facets. When
3944 * this is the case we might as well keep statistics for all of them in
3945 * userspace and to cache them in the kernel datapath for performance as
3948 * As the number of facets increases, the memory required to maintain
3949 * statistics about them in userspace and in the kernel becomes
3950 * significant. However, with a large number of facets it is likely that
3951 * only a few of them are "heavy hitters" that consume a large amount of
3952 * bandwidth. At this point, only heavy hitters are worth caching in the
3953 * kernel and maintaining in userspaces; other facets we can discard.
3955 * The technique used to compute the idle time is to build a histogram with
3956 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
3957 * that is installed in the kernel gets dropped in the appropriate bucket.
3958 * After the histogram has been built, we compute the cutoff so that only
3959 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
3960 * cached. At least the most-recently-used bucket of facets is kept, so
3961 * actually an arbitrary number of facets can be kept in any given
3962 * expiration run (though the next run will delete most of those unless
3963 * they receive additional data).
3965 * This requires a second pass through the facets, in addition to the pass
3966 * made by ofproto_update_stats(), because the former function never looks
3967 * at uninstallable facets.
3969 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3970 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3971 int buckets[N_BUCKETS] = { 0 };
3972 struct facet *facet;
3977 total = hmap_count(&ofproto->facets);
3978 if (total <= 1000) {
3979 return N_BUCKETS * BUCKET_WIDTH;
3982 /* Build histogram. */
3984 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
3985 long long int idle = now - facet->used;
3986 int bucket = (idle <= 0 ? 0
3987 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3988 : (unsigned int) idle / BUCKET_WIDTH);
3992 /* Find the first bucket whose flows should be expired. */
3993 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
3994 if (buckets[bucket]) {
3997 subtotal += buckets[bucket++];
3998 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4003 if (VLOG_IS_DBG_ENABLED()) {
4007 ds_put_cstr(&s, "keep");
4008 for (i = 0; i < N_BUCKETS; i++) {
4010 ds_put_cstr(&s, ", drop");
4013 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4016 VLOG_INFO("%s: %s (msec:count)",
4017 dpif_name(ofproto->dpif), ds_cstr(&s));
4021 return bucket * BUCKET_WIDTH;
4025 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4027 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4028 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4029 struct ofexpired expired;
4031 if (facet->installed) {
4032 struct dpif_flow_stats stats;
4034 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4036 facet_update_stats(ofproto, facet, &stats);
4039 expired.flow = facet->flow;
4040 expired.packet_count = facet->packet_count;
4041 expired.byte_count = facet->byte_count;
4042 expired.used = facet->used;
4043 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4048 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4050 long long int cutoff = time_msec() - dp_max_idle;
4051 struct facet *facet, *next_facet;
4053 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4054 facet_active_timeout(ofproto, facet);
4055 if (facet->used < cutoff) {
4056 facet_remove(ofproto, facet);
4061 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4062 * then delete it entirely. */
4064 rule_expire(struct ofproto *ofproto, struct rule *rule)
4066 struct facet *facet, *next_facet;
4070 /* Has 'rule' expired? */
4072 if (rule->hard_timeout
4073 && now > rule->created + rule->hard_timeout * 1000) {
4074 reason = OFPRR_HARD_TIMEOUT;
4075 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4076 && now >rule->used + rule->idle_timeout * 1000) {
4077 reason = OFPRR_IDLE_TIMEOUT;
4082 COVERAGE_INC(ofproto_expired);
4084 /* Update stats. (This is a no-op if the rule expired due to an idle
4085 * timeout, because that only happens when the rule has no facets left.) */
4086 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4087 facet_remove(ofproto, facet);
4090 /* Get rid of the rule. */
4091 if (!rule_is_hidden(rule)) {
4092 rule_send_removed(ofproto, rule, reason);
4094 rule_remove(ofproto, rule);
4098 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4100 struct ofputil_flow_removed fr;
4102 if (!rule->send_flow_removed) {
4107 fr.cookie = rule->flow_cookie;
4109 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4110 fr.idle_timeout = rule->idle_timeout;
4111 fr.packet_count = rule->packet_count;
4112 fr.byte_count = rule->byte_count;
4114 connmgr_send_flow_removed(p->connmgr, &fr);
4117 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4118 * The returned statistics include statistics for all of 'rule''s facets. */
4120 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4123 struct facet *facet;
4125 /* Start from historical data for 'rule' itself that are no longer tracked
4126 * in facets. This counts, for example, facets that have expired. */
4127 p = rule->packet_count;
4128 b = rule->byte_count;
4130 /* Add any statistics that are tracked by facets. This includes
4131 * statistical data recently updated by ofproto_update_stats() as well as
4132 * stats for packets that were executed "by hand" via dpif_execute(). */
4133 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4134 p += facet->packet_count;
4135 b += facet->byte_count;
4142 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4143 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4144 * their individual configurations.
4146 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4147 * Otherwise, ownership is transferred to this function. */
4149 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4150 const struct flow *flow, bool clone)
4152 struct ofputil_packet_in pin;
4154 pin.packet = upcall->packet;
4155 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4156 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4157 pin.buffer_id = 0; /* not yet known */
4158 pin.send_len = upcall->userdata;
4159 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4160 clone ? NULL : upcall->packet);
4164 pick_datapath_id(const struct ofproto *ofproto)
4166 const struct ofport *port;
4168 port = get_port(ofproto, ODPP_LOCAL);
4170 uint8_t ea[ETH_ADDR_LEN];
4173 error = netdev_get_etheraddr(port->netdev, ea);
4175 return eth_addr_to_uint64(ea);
4177 VLOG_WARN("could not get MAC address for %s (%s)",
4178 netdev_get_name(port->netdev), strerror(error));
4180 return ofproto->fallback_dpid;
4184 pick_fallback_dpid(void)
4186 uint8_t ea[ETH_ADDR_LEN];
4187 eth_addr_nicira_random(ea);
4188 return eth_addr_to_uint64(ea);
4192 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4193 void *aux OVS_UNUSED)
4195 const struct shash_node *node;
4199 SHASH_FOR_EACH (node, &all_ofprotos) {
4200 ds_put_format(&results, "%s\n", node->name);
4202 unixctl_command_reply(conn, 200, ds_cstr(&results));
4203 ds_destroy(&results);
4206 struct ofproto_trace {
4207 struct action_xlate_ctx ctx;
4213 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4215 ds_put_char_multiple(result, '\t', level);
4217 ds_put_cstr(result, "No match\n");
4221 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4222 ntohll(rule->flow_cookie));
4223 cls_rule_format(&rule->cr, result);
4224 ds_put_char(result, '\n');
4226 ds_put_char_multiple(result, '\t', level);
4227 ds_put_cstr(result, "OpenFlow ");
4228 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4229 rule->n_actions * sizeof *rule->actions);
4230 ds_put_char(result, '\n');
4234 trace_format_flow(struct ds *result, int level, const char *title,
4235 struct ofproto_trace *trace)
4237 ds_put_char_multiple(result, '\t', level);
4238 ds_put_format(result, "%s: ", title);
4239 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4240 ds_put_cstr(result, "unchanged");
4242 flow_format(result, &trace->ctx.flow);
4243 trace->flow = trace->ctx.flow;
4245 ds_put_char(result, '\n');
4249 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4251 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4252 struct ds *result = trace->result;
4254 ds_put_char(result, '\n');
4255 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4256 trace_format_rule(result, ctx->recurse + 1, rule);
4260 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4261 void *aux OVS_UNUSED)
4263 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4264 char *args = xstrdup(args_);
4265 char *save_ptr = NULL;
4266 struct ofproto *ofproto;
4267 struct ofpbuf packet;
4275 ofpbuf_init(&packet, strlen(args) / 2);
4278 dpname = strtok_r(args, " ", &save_ptr);
4279 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4280 in_port_s = strtok_r(NULL, " ", &save_ptr);
4281 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4282 if (!dpname || !in_port_s || !packet_s) {
4283 unixctl_command_reply(conn, 501, "Bad command syntax");
4287 ofproto = shash_find_data(&all_ofprotos, dpname);
4289 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4294 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4295 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4297 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4298 packet_s += strspn(packet_s, " ");
4299 if (*packet_s != '\0') {
4300 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4303 if (packet.size < ETH_HEADER_LEN) {
4304 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4308 ds_put_cstr(&result, "Packet: ");
4309 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4310 ds_put_cstr(&result, s);
4313 flow_extract(&packet, tun_id, in_port, &flow);
4314 ds_put_cstr(&result, "Flow: ");
4315 flow_format(&result, &flow);
4316 ds_put_char(&result, '\n');
4318 rule = rule_lookup(ofproto, &flow);
4319 trace_format_rule(&result, 0, rule);
4321 struct ofproto_trace trace;
4322 struct ofpbuf *odp_actions;
4324 trace.result = &result;
4326 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4327 trace.ctx.resubmit_hook = trace_resubmit;
4328 odp_actions = xlate_actions(&trace.ctx,
4329 rule->actions, rule->n_actions);
4331 ds_put_char(&result, '\n');
4332 trace_format_flow(&result, 0, "Final flow", &trace);
4333 ds_put_cstr(&result, "Datapath actions: ");
4334 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4335 ofpbuf_delete(odp_actions);
4338 unixctl_command_reply(conn, 200, ds_cstr(&result));
4341 ds_destroy(&result);
4342 ofpbuf_uninit(&packet);
4347 ofproto_unixctl_init(void)
4349 static bool registered;
4355 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4356 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4360 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4361 struct ofpbuf *odp_actions, tag_type *tags,
4362 uint16_t *nf_output_iface, void *ofproto_)
4364 struct ofproto *ofproto = ofproto_;
4365 struct mac_entry *dst_mac;
4367 /* Drop frames for reserved multicast addresses. */
4368 if (eth_addr_is_reserved(flow->dl_dst)) {
4372 /* Learn source MAC (but don't try to learn from revalidation). */
4374 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4375 struct mac_entry *src_mac;
4377 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4378 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4379 /* The log messages here could actually be useful in debugging,
4380 * so keep the rate limit relatively high. */
4381 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4382 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4383 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4385 ofproto_revalidate(ofproto,
4386 mac_learning_changed(ofproto->ml, src_mac));
4387 src_mac->port.i = flow->in_port;
4391 /* Determine output port. */
4392 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4394 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4395 nf_output_iface, odp_actions);
4397 int out_port = dst_mac->port.i;
4398 if (out_port != flow->in_port) {
4399 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4400 *nf_output_iface = out_port;
4409 static const struct ofhooks default_ofhooks = {
4410 default_normal_ofhook_cb,