2 * Copyright (c) 2009, 2010 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 "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
34 #include "mac-learning.h"
38 #include "ofp-print.h"
40 #include "ofproto-sflow.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
44 #include "openvswitch/datapath-protocol.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
54 #include "stream-ssl.h"
62 #define THIS_MODULE VLM_ofproto
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER = 1
73 struct netdev *netdev;
74 struct ofp_phy_port opp; /* In host byte order. */
77 static void ofport_free(struct ofport *);
78 static void hton_ofp_phy_port(struct ofp_phy_port *);
80 static int xlate_actions(const union ofp_action *in, size_t n_in,
81 const flow_t *flow, struct ofproto *ofproto,
82 const struct ofpbuf *packet,
83 struct odp_actions *out, tag_type *tags,
84 bool *may_set_up_flow, uint16_t *nf_output_iface);
89 uint64_t flow_cookie; /* Controller-issued identifier.
90 (Kept in network-byte order.) */
91 uint16_t idle_timeout; /* In seconds from time of last use. */
92 uint16_t hard_timeout; /* In seconds from time of creation. */
93 bool send_flow_removed; /* Send a flow removed message? */
94 long long int used; /* Last-used time (0 if never used). */
95 long long int created; /* Creation time. */
96 uint64_t packet_count; /* Number of packets received. */
97 uint64_t byte_count; /* Number of bytes received. */
98 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
99 tag_type tags; /* Tags (set only by hooks). */
100 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
102 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
103 * exact-match rule (having cr.wc.wildcards of 0) generated from the
104 * wildcard rule 'super'. In this case, 'list' is an element of the
107 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
108 * a list of subrules. A super-rule with no wildcards (where
109 * cr.wc.wildcards is 0) will never have any subrules. */
115 * 'n_actions' is the number of elements in the 'actions' array. A single
116 * action may take up more more than one element's worth of space.
118 * A subrule has no actions (it uses the super-rule's actions). */
120 union ofp_action *actions;
124 * A super-rule with wildcard fields never has ODP actions (since the
125 * datapath only supports exact-match flows). */
126 bool installed; /* Installed in datapath? */
127 bool may_install; /* True ordinarily; false if actions must
128 * be reassessed for every packet. */
130 union odp_action *odp_actions;
134 rule_is_hidden(const struct rule *rule)
136 /* Subrules are merely an implementation detail, so hide them from the
138 if (rule->super != NULL) {
142 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
143 * (e.g. by in-band control) and are intentionally hidden from the
145 if (rule->cr.priority > UINT16_MAX) {
152 static struct rule *rule_create(struct ofproto *, struct rule *super,
153 const union ofp_action *, size_t n_actions,
154 uint16_t idle_timeout, uint16_t hard_timeout,
155 uint64_t flow_cookie, bool send_flow_removed);
156 static void rule_free(struct rule *);
157 static void rule_destroy(struct ofproto *, struct rule *);
158 static struct rule *rule_from_cls_rule(const struct cls_rule *);
159 static void rule_insert(struct ofproto *, struct rule *,
160 struct ofpbuf *packet, uint16_t in_port);
161 static void rule_remove(struct ofproto *, struct rule *);
162 static bool rule_make_actions(struct ofproto *, struct rule *,
163 const struct ofpbuf *packet);
164 static void rule_install(struct ofproto *, struct rule *,
165 struct rule *displaced_rule);
166 static void rule_uninstall(struct ofproto *, struct rule *);
167 static void rule_post_uninstall(struct ofproto *, struct rule *);
168 static void send_flow_removed(struct ofproto *p, struct rule *rule,
169 long long int now, uint8_t reason);
171 /* ofproto supports two kinds of OpenFlow connections:
173 * - "Controller connections": Connections to ordinary OpenFlow controllers.
174 * ofproto maintains persistent connections to these controllers and by
175 * default sends them asynchronous messages such as packet-ins.
177 * - "Transient connections", e.g. from ovs-ofctl. When these connections
178 * drop, it is the other side's responsibility to reconnect them if
179 * necessary. ofproto does not send them asynchronous messages by default.
182 OFCONN_CONTROLLER, /* An OpenFlow controller. */
183 OFCONN_TRANSIENT /* A transient connection. */
186 /* An OpenFlow connection. */
188 struct ofproto *ofproto; /* The ofproto that owns this connection. */
189 struct list node; /* In struct ofproto's "all_conns" list. */
190 struct rconn *rconn; /* OpenFlow connection. */
191 enum ofconn_type type; /* Type. */
193 /* OFPT_PACKET_IN related data. */
194 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
195 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
196 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
197 int miss_send_len; /* Bytes to send of buffered packets. */
199 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
200 * requests, and the maximum number before we stop reading OpenFlow
202 #define OFCONN_REPLY_MAX 100
203 struct rconn_packet_counter *reply_counter;
205 /* type == OFCONN_CONTROLLER only. */
206 enum nx_role role; /* Role. */
207 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
208 struct discovery *discovery; /* Controller discovery object, if enabled. */
209 struct status_category *ss; /* Switch status category. */
210 enum ofproto_band band; /* In-band or out-of-band? */
213 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
214 * "schedulers" array. Their values are 0 and 1, and their meanings and values
215 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
216 * case anything ever changes, check their values here. */
217 #define N_SCHEDULERS 2
218 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
219 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
220 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
221 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
223 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
225 static void ofconn_destroy(struct ofconn *);
226 static void ofconn_run(struct ofconn *, struct ofproto *);
227 static void ofconn_wait(struct ofconn *);
228 static bool ofconn_receives_async_msgs(const struct ofconn *);
229 static char *ofconn_make_name(const struct ofproto *, const char *target);
231 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
232 struct rconn_packet_counter *counter);
234 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
235 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
239 uint64_t datapath_id; /* Datapath ID. */
240 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
241 char *mfr_desc; /* Manufacturer. */
242 char *hw_desc; /* Hardware. */
243 char *sw_desc; /* Software version. */
244 char *serial_desc; /* Serial number. */
245 char *dp_desc; /* Datapath description. */
249 struct netdev_monitor *netdev_monitor;
250 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
252 struct shash port_by_name;
256 struct switch_status *switch_status;
257 struct fail_open *fail_open;
258 struct netflow *netflow;
259 struct ofproto_sflow *sflow;
261 /* In-band control. */
262 struct in_band *in_band;
263 long long int next_in_band_update;
264 struct sockaddr_in *extra_in_band_remotes;
265 size_t n_extra_remotes;
268 struct classifier cls;
269 bool need_revalidate;
270 long long int next_expiration;
271 struct tag_set revalidate_set;
272 bool tun_id_from_cookie;
274 /* OpenFlow connections. */
275 struct hmap controllers; /* Controller "struct ofconn"s. */
276 struct list all_conns; /* Contains "struct ofconn"s. */
277 struct pvconn **listeners;
279 struct pvconn **snoops;
282 /* Hooks for ovs-vswitchd. */
283 const struct ofhooks *ofhooks;
286 /* Used by default ofhooks. */
287 struct mac_learning *ml;
290 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
292 static const struct ofhooks default_ofhooks;
294 static uint64_t pick_datapath_id(const struct ofproto *);
295 static uint64_t pick_fallback_dpid(void);
297 static void update_used(struct ofproto *);
298 static void update_stats(struct ofproto *, struct rule *,
299 const struct odp_flow_stats *);
300 static void expire_rule(struct cls_rule *, void *ofproto);
301 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
302 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
303 static void revalidate_cb(struct cls_rule *rule_, void *p_);
305 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
307 static void handle_openflow(struct ofconn *, struct ofproto *,
310 static void refresh_port_groups(struct ofproto *);
312 static void update_port(struct ofproto *, const char *devname);
313 static int init_ports(struct ofproto *);
314 static void reinit_ports(struct ofproto *);
317 ofproto_create(const char *datapath, const char *datapath_type,
318 const struct ofhooks *ofhooks, void *aux,
319 struct ofproto **ofprotop)
321 struct odp_stats stats;
328 /* Connect to datapath and start listening for messages. */
329 error = dpif_open(datapath, datapath_type, &dpif);
331 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
334 error = dpif_get_dp_stats(dpif, &stats);
336 VLOG_ERR("failed to obtain stats for datapath %s: %s",
337 datapath, strerror(error));
341 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
343 VLOG_ERR("failed to listen on datapath %s: %s",
344 datapath, strerror(error));
348 dpif_flow_flush(dpif);
349 dpif_recv_purge(dpif);
351 /* Initialize settings. */
352 p = xzalloc(sizeof *p);
353 p->fallback_dpid = pick_fallback_dpid();
354 p->datapath_id = p->fallback_dpid;
355 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
356 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
357 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
358 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
359 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
361 /* Initialize datapath. */
363 p->netdev_monitor = netdev_monitor_create();
364 port_array_init(&p->ports);
365 shash_init(&p->port_by_name);
366 p->max_ports = stats.max_ports;
368 /* Initialize submodules. */
369 p->switch_status = switch_status_create(p);
375 /* Initialize flow table. */
376 classifier_init(&p->cls);
377 p->need_revalidate = false;
378 p->next_expiration = time_msec() + 1000;
379 tag_set_init(&p->revalidate_set);
381 /* Initialize OpenFlow connections. */
382 list_init(&p->all_conns);
383 hmap_init(&p->controllers);
389 /* Initialize hooks. */
391 p->ofhooks = ofhooks;
395 p->ofhooks = &default_ofhooks;
397 p->ml = mac_learning_create();
400 /* Pick final datapath ID. */
401 p->datapath_id = pick_datapath_id(p);
402 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
409 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
411 uint64_t old_dpid = p->datapath_id;
412 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
413 if (p->datapath_id != old_dpid) {
414 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
416 /* Force all active connections to reconnect, since there is no way to
417 * notify a controller that the datapath ID has changed. */
418 ofproto_reconnect_controllers(p);
423 is_discovery_controller(const struct ofproto_controller *c)
425 return !strcmp(c->target, "discover");
429 is_in_band_controller(const struct ofproto_controller *c)
431 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
434 /* Creates a new controller in 'ofproto'. Some of the settings are initially
435 * drawn from 'c', but update_controller() needs to be called later to finish
436 * the new ofconn's configuration. */
438 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
440 struct discovery *discovery;
441 struct ofconn *ofconn;
443 if (is_discovery_controller(c)) {
444 int error = discovery_create(c->accept_re, c->update_resolv_conf,
445 ofproto->dpif, ofproto->switch_status,
454 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
455 ofconn->pktbuf = pktbuf_create();
456 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
458 ofconn->discovery = discovery;
460 char *name = ofconn_make_name(ofproto, c->target);
461 rconn_connect(ofconn->rconn, c->target, name);
464 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
465 hash_string(c->target, 0));
468 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
469 * target or turn discovery on or off (these are done by creating new ofconns
470 * and deleting old ones), but it can update the rest of an ofconn's
473 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
475 struct ofproto *ofproto = ofconn->ofproto;
479 ofconn->band = (is_in_band_controller(c)
480 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
482 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
484 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
485 rconn_set_probe_interval(ofconn->rconn, probe_interval);
487 if (ofconn->discovery) {
488 discovery_set_update_resolv_conf(ofconn->discovery,
489 c->update_resolv_conf);
490 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
493 for (i = 0; i < N_SCHEDULERS; i++) {
494 struct pinsched **s = &ofconn->schedulers[i];
496 if (c->rate_limit > 0) {
498 *s = pinsched_create(c->rate_limit, c->burst_limit,
499 ofproto->switch_status);
501 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
504 pinsched_destroy(*s);
511 ofconn_get_target(const struct ofconn *ofconn)
513 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
516 static struct ofconn *
517 find_controller_by_target(struct ofproto *ofproto, const char *target)
519 struct ofconn *ofconn;
521 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
522 hash_string(target, 0), &ofproto->controllers) {
523 if (!strcmp(ofconn_get_target(ofconn), target)) {
531 update_in_band_remotes(struct ofproto *ofproto)
533 const struct ofconn *ofconn;
534 struct sockaddr_in *addrs;
535 size_t max_addrs, n_addrs;
539 /* Allocate enough memory for as many remotes as we could possibly have. */
540 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
541 addrs = xmalloc(max_addrs * sizeof *addrs);
544 /* Add all the remotes. */
546 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
547 struct sockaddr_in *sin = &addrs[n_addrs];
549 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
553 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
554 if (sin->sin_addr.s_addr) {
555 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
558 if (ofconn->discovery) {
562 for (i = 0; i < ofproto->n_extra_remotes; i++) {
563 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
566 /* Create or update or destroy in-band.
568 * Ordinarily we only enable in-band if there's at least one remote
569 * address, but discovery needs the in-band rules for DHCP to be installed
570 * even before we know any remote addresses. */
571 if (n_addrs || discovery) {
572 if (!ofproto->in_band) {
573 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
576 if (ofproto->in_band) {
577 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
579 ofproto->next_in_band_update = time_msec() + 1000;
581 in_band_destroy(ofproto->in_band);
582 ofproto->in_band = NULL;
590 ofproto_set_controllers(struct ofproto *p,
591 const struct ofproto_controller *controllers,
592 size_t n_controllers)
594 struct shash new_controllers;
595 enum ofproto_fail_mode fail_mode;
596 struct ofconn *ofconn, *next;
600 shash_init(&new_controllers);
601 for (i = 0; i < n_controllers; i++) {
602 const struct ofproto_controller *c = &controllers[i];
604 shash_add_once(&new_controllers, c->target, &controllers[i]);
605 if (!find_controller_by_target(p, c->target)) {
606 add_controller(p, c);
610 fail_mode = OFPROTO_FAIL_STANDALONE;
612 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
614 struct ofproto_controller *c;
616 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
618 ofconn_destroy(ofconn);
620 update_controller(ofconn, c);
624 if (c->fail == OFPROTO_FAIL_SECURE) {
625 fail_mode = OFPROTO_FAIL_SECURE;
629 shash_destroy(&new_controllers);
631 update_in_band_remotes(p);
633 if (!hmap_is_empty(&p->controllers)
634 && fail_mode == OFPROTO_FAIL_STANDALONE) {
635 struct rconn **rconns;
639 p->fail_open = fail_open_create(p, p->switch_status);
643 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
644 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
645 rconns[n++] = ofconn->rconn;
648 fail_open_set_controllers(p->fail_open, rconns, n);
649 /* p->fail_open takes ownership of 'rconns'. */
651 fail_open_destroy(p->fail_open);
655 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
656 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
657 struct ofconn, hmap_node);
658 ofconn->ss = switch_status_register(p->switch_status, "remote",
659 rconn_status_cb, ofconn->rconn);
663 /* Drops the connections between 'ofproto' and all of its controllers, forcing
664 * them to reconnect. */
666 ofproto_reconnect_controllers(struct ofproto *ofproto)
668 struct ofconn *ofconn;
670 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
671 rconn_reconnect(ofconn->rconn);
676 any_extras_changed(const struct ofproto *ofproto,
677 const struct sockaddr_in *extras, size_t n)
681 if (n != ofproto->n_extra_remotes) {
685 for (i = 0; i < n; i++) {
686 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
687 const struct sockaddr_in *new = &extras[i];
689 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
690 old->sin_port != new->sin_port) {
698 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
699 * in-band control should guarantee access, in the same way that in-band
700 * control guarantees access to OpenFlow controllers. */
702 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
703 const struct sockaddr_in *extras, size_t n)
705 if (!any_extras_changed(ofproto, extras, n)) {
709 free(ofproto->extra_in_band_remotes);
710 ofproto->n_extra_remotes = n;
711 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
713 update_in_band_remotes(ofproto);
717 ofproto_set_desc(struct ofproto *p,
718 const char *mfr_desc, const char *hw_desc,
719 const char *sw_desc, const char *serial_desc,
722 struct ofp_desc_stats *ods;
725 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
726 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
727 sizeof ods->mfr_desc);
730 p->mfr_desc = xstrdup(mfr_desc);
733 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
734 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
735 sizeof ods->hw_desc);
738 p->hw_desc = xstrdup(hw_desc);
741 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
742 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
743 sizeof ods->sw_desc);
746 p->sw_desc = xstrdup(sw_desc);
749 if (strlen(serial_desc) >= sizeof ods->serial_num) {
750 VLOG_WARN("truncating serial_desc, must be less than %zu "
752 sizeof ods->serial_num);
754 free(p->serial_desc);
755 p->serial_desc = xstrdup(serial_desc);
758 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
759 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
760 sizeof ods->dp_desc);
763 p->dp_desc = xstrdup(dp_desc);
768 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
769 const struct svec *svec)
771 struct pvconn **pvconns = *pvconnsp;
772 size_t n_pvconns = *n_pvconnsp;
776 for (i = 0; i < n_pvconns; i++) {
777 pvconn_close(pvconns[i]);
781 pvconns = xmalloc(svec->n * sizeof *pvconns);
783 for (i = 0; i < svec->n; i++) {
784 const char *name = svec->names[i];
785 struct pvconn *pvconn;
788 error = pvconn_open(name, &pvconn);
790 pvconns[n_pvconns++] = pvconn;
792 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
800 *n_pvconnsp = n_pvconns;
806 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
808 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
812 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
814 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
818 ofproto_set_netflow(struct ofproto *ofproto,
819 const struct netflow_options *nf_options)
821 if (nf_options && nf_options->collectors.n) {
822 if (!ofproto->netflow) {
823 ofproto->netflow = netflow_create();
825 return netflow_set_options(ofproto->netflow, nf_options);
827 netflow_destroy(ofproto->netflow);
828 ofproto->netflow = NULL;
834 ofproto_set_sflow(struct ofproto *ofproto,
835 const struct ofproto_sflow_options *oso)
837 struct ofproto_sflow *os = ofproto->sflow;
840 struct ofport *ofport;
841 unsigned int odp_port;
843 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
844 refresh_port_groups(ofproto);
845 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
846 ofproto_sflow_add_port(os, odp_port,
847 netdev_get_name(ofport->netdev));
850 ofproto_sflow_set_options(os, oso);
852 ofproto_sflow_destroy(os);
853 ofproto->sflow = NULL;
858 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
862 VLOG_WARN("STP is not yet implemented");
870 ofproto_get_datapath_id(const struct ofproto *ofproto)
872 return ofproto->datapath_id;
876 ofproto_has_controller(const struct ofproto *ofproto)
878 return !hmap_is_empty(&ofproto->controllers);
882 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
886 for (i = 0; i < ofproto->n_listeners; i++) {
887 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
892 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
896 for (i = 0; i < ofproto->n_snoops; i++) {
897 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
902 ofproto_destroy(struct ofproto *p)
904 struct ofconn *ofconn, *next_ofconn;
905 struct ofport *ofport;
906 unsigned int port_no;
913 /* Destroy fail-open and in-band early, since they touch the classifier. */
914 fail_open_destroy(p->fail_open);
917 in_band_destroy(p->in_band);
919 free(p->extra_in_band_remotes);
921 ofproto_flush_flows(p);
922 classifier_destroy(&p->cls);
924 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
926 ofconn_destroy(ofconn);
928 hmap_destroy(&p->controllers);
931 netdev_monitor_destroy(p->netdev_monitor);
932 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
935 shash_destroy(&p->port_by_name);
937 switch_status_destroy(p->switch_status);
938 netflow_destroy(p->netflow);
939 ofproto_sflow_destroy(p->sflow);
941 for (i = 0; i < p->n_listeners; i++) {
942 pvconn_close(p->listeners[i]);
946 for (i = 0; i < p->n_snoops; i++) {
947 pvconn_close(p->snoops[i]);
951 mac_learning_destroy(p->ml);
956 free(p->serial_desc);
959 port_array_destroy(&p->ports);
965 ofproto_run(struct ofproto *p)
967 int error = ofproto_run1(p);
969 error = ofproto_run2(p, false);
975 process_port_change(struct ofproto *ofproto, int error, char *devname)
977 if (error == ENOBUFS) {
978 reinit_ports(ofproto);
980 update_port(ofproto, devname);
985 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
986 * means that 'ofconn' is more interesting for monitoring than a lower return
989 snoop_preference(const struct ofconn *ofconn)
991 switch (ofconn->role) {
999 /* Shouldn't happen. */
1004 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1005 * Connects this vconn to a controller. */
1007 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1009 struct ofconn *ofconn, *best;
1011 /* Pick a controller for monitoring. */
1013 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1014 if (ofconn->type == OFCONN_CONTROLLER
1015 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1021 rconn_add_monitor(best->rconn, vconn);
1023 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1029 ofproto_run1(struct ofproto *p)
1031 struct ofconn *ofconn, *next_ofconn;
1036 if (shash_is_empty(&p->port_by_name)) {
1040 for (i = 0; i < 50; i++) {
1044 error = dpif_recv(p->dpif, &buf);
1046 if (error == ENODEV) {
1047 /* Someone destroyed the datapath behind our back. The caller
1048 * better destroy us and give up, because we're just going to
1049 * spin from here on out. */
1050 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1051 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1052 dpif_name(p->dpif));
1058 handle_odp_msg(p, buf);
1061 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1062 process_port_change(p, error, devname);
1064 while ((error = netdev_monitor_poll(p->netdev_monitor,
1065 &devname)) != EAGAIN) {
1066 process_port_change(p, error, devname);
1070 if (time_msec() >= p->next_in_band_update) {
1071 update_in_band_remotes(p);
1073 in_band_run(p->in_band);
1076 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1078 ofconn_run(ofconn, p);
1081 /* Fail-open maintenance. Do this after processing the ofconns since
1082 * fail-open checks the status of the controller rconn. */
1084 fail_open_run(p->fail_open);
1087 for (i = 0; i < p->n_listeners; i++) {
1088 struct vconn *vconn;
1091 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1093 struct rconn *rconn;
1096 rconn = rconn_create(60, 0);
1097 name = ofconn_make_name(p, vconn_get_name(vconn));
1098 rconn_connect_unreliably(rconn, vconn, name);
1101 ofconn_create(p, rconn, OFCONN_TRANSIENT);
1102 } else if (retval != EAGAIN) {
1103 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1107 for (i = 0; i < p->n_snoops; i++) {
1108 struct vconn *vconn;
1111 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1113 add_snooper(p, vconn);
1114 } else if (retval != EAGAIN) {
1115 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1119 if (time_msec() >= p->next_expiration) {
1120 COVERAGE_INC(ofproto_expiration);
1121 p->next_expiration = time_msec() + 1000;
1124 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1126 /* Let the hook know that we're at a stable point: all outstanding data
1127 * in existing flows has been accounted to the account_cb. Thus, the
1128 * hook can now reasonably do operations that depend on having accurate
1129 * flow volume accounting (currently, that's just bond rebalancing). */
1130 if (p->ofhooks->account_checkpoint_cb) {
1131 p->ofhooks->account_checkpoint_cb(p->aux);
1136 netflow_run(p->netflow);
1139 ofproto_sflow_run(p->sflow);
1145 struct revalidate_cbdata {
1146 struct ofproto *ofproto;
1147 bool revalidate_all; /* Revalidate all exact-match rules? */
1148 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1149 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1153 ofproto_run2(struct ofproto *p, bool revalidate_all)
1155 if (p->need_revalidate || revalidate_all
1156 || !tag_set_is_empty(&p->revalidate_set)) {
1157 struct revalidate_cbdata cbdata;
1159 cbdata.revalidate_all = revalidate_all;
1160 cbdata.revalidate_subrules = p->need_revalidate;
1161 cbdata.revalidate_set = p->revalidate_set;
1162 tag_set_init(&p->revalidate_set);
1163 COVERAGE_INC(ofproto_revalidate);
1164 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1165 p->need_revalidate = false;
1172 ofproto_wait(struct ofproto *p)
1174 struct ofconn *ofconn;
1177 dpif_recv_wait(p->dpif);
1178 dpif_port_poll_wait(p->dpif);
1179 netdev_monitor_poll_wait(p->netdev_monitor);
1180 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1181 ofconn_wait(ofconn);
1184 poll_timer_wait_until(p->next_in_band_update);
1185 in_band_wait(p->in_band);
1188 fail_open_wait(p->fail_open);
1191 ofproto_sflow_wait(p->sflow);
1193 if (!tag_set_is_empty(&p->revalidate_set)) {
1194 poll_immediate_wake();
1196 if (p->need_revalidate) {
1197 /* Shouldn't happen, but if it does just go around again. */
1198 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1199 poll_immediate_wake();
1200 } else if (p->next_expiration != LLONG_MAX) {
1201 poll_timer_wait_until(p->next_expiration);
1203 for (i = 0; i < p->n_listeners; i++) {
1204 pvconn_wait(p->listeners[i]);
1206 for (i = 0; i < p->n_snoops; i++) {
1207 pvconn_wait(p->snoops[i]);
1212 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1214 tag_set_add(&ofproto->revalidate_set, tag);
1218 ofproto_get_revalidate_set(struct ofproto *ofproto)
1220 return &ofproto->revalidate_set;
1224 ofproto_is_alive(const struct ofproto *p)
1226 return !hmap_is_empty(&p->controllers);
1230 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1231 const union ofp_action *actions, size_t n_actions,
1232 const struct ofpbuf *packet)
1234 struct odp_actions odp_actions;
1237 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1243 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1245 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1246 odp_actions.n_actions, packet);
1251 ofproto_add_flow(struct ofproto *p,
1252 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1253 const union ofp_action *actions, size_t n_actions,
1257 rule = rule_create(p, NULL, actions, n_actions,
1258 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1260 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1261 rule_insert(p, rule, NULL, 0);
1265 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1266 uint32_t wildcards, unsigned int priority)
1270 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1274 rule_remove(ofproto, rule);
1279 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1281 struct rule *rule = rule_from_cls_rule(rule_);
1282 struct ofproto *ofproto = ofproto_;
1284 /* Mark the flow as not installed, even though it might really be
1285 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1286 * There is no point in uninstalling it individually since we are about to
1287 * blow away all the flows with dpif_flow_flush(). */
1288 rule->installed = false;
1290 rule_remove(ofproto, rule);
1294 ofproto_flush_flows(struct ofproto *ofproto)
1296 COVERAGE_INC(ofproto_flush);
1297 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1298 dpif_flow_flush(ofproto->dpif);
1299 if (ofproto->in_band) {
1300 in_band_flushed(ofproto->in_band);
1302 if (ofproto->fail_open) {
1303 fail_open_flushed(ofproto->fail_open);
1308 reinit_ports(struct ofproto *p)
1310 struct svec devnames;
1311 struct ofport *ofport;
1312 unsigned int port_no;
1313 struct odp_port *odp_ports;
1317 svec_init(&devnames);
1318 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1319 svec_add (&devnames, (char *) ofport->opp.name);
1321 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1322 for (i = 0; i < n_odp_ports; i++) {
1323 svec_add (&devnames, odp_ports[i].devname);
1327 svec_sort_unique(&devnames);
1328 for (i = 0; i < devnames.n; i++) {
1329 update_port(p, devnames.names[i]);
1331 svec_destroy(&devnames);
1335 refresh_port_group(struct ofproto *p, unsigned int group)
1339 struct ofport *port;
1340 unsigned int port_no;
1342 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1344 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1346 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1347 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1348 ports[n_ports++] = port_no;
1351 dpif_port_group_set(p->dpif, group, ports, n_ports);
1358 refresh_port_groups(struct ofproto *p)
1360 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1361 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1363 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1367 static struct ofport *
1368 make_ofport(const struct odp_port *odp_port)
1370 struct netdev_options netdev_options;
1371 enum netdev_flags flags;
1372 struct ofport *ofport;
1373 struct netdev *netdev;
1377 memset(&netdev_options, 0, sizeof netdev_options);
1378 netdev_options.name = odp_port->devname;
1379 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1381 error = netdev_open(&netdev_options, &netdev);
1383 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1384 "cannot be opened (%s)",
1385 odp_port->devname, odp_port->port,
1386 odp_port->devname, strerror(error));
1390 ofport = xmalloc(sizeof *ofport);
1391 ofport->netdev = netdev;
1392 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1393 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1394 memcpy(ofport->opp.name, odp_port->devname,
1395 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1396 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1398 netdev_get_flags(netdev, &flags);
1399 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1401 netdev_get_carrier(netdev, &carrier);
1402 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1404 netdev_get_features(netdev,
1405 &ofport->opp.curr, &ofport->opp.advertised,
1406 &ofport->opp.supported, &ofport->opp.peer);
1411 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1413 if (port_array_get(&p->ports, odp_port->port)) {
1414 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1417 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1418 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1427 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1429 const struct ofp_phy_port *a = &a_->opp;
1430 const struct ofp_phy_port *b = &b_->opp;
1432 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1433 return (a->port_no == b->port_no
1434 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1435 && !strcmp((char *) a->name, (char *) b->name)
1436 && a->state == b->state
1437 && a->config == b->config
1438 && a->curr == b->curr
1439 && a->advertised == b->advertised
1440 && a->supported == b->supported
1441 && a->peer == b->peer);
1445 send_port_status(struct ofproto *p, const struct ofport *ofport,
1448 /* XXX Should limit the number of queued port status change messages. */
1449 struct ofconn *ofconn;
1450 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1451 struct ofp_port_status *ops;
1454 if (!ofconn_receives_async_msgs(ofconn)) {
1458 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1459 ops->reason = reason;
1460 ops->desc = ofport->opp;
1461 hton_ofp_phy_port(&ops->desc);
1462 queue_tx(b, ofconn, NULL);
1464 if (p->ofhooks->port_changed_cb) {
1465 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1470 ofport_install(struct ofproto *p, struct ofport *ofport)
1472 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1473 const char *netdev_name = (const char *) ofport->opp.name;
1475 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1476 port_array_set(&p->ports, odp_port, ofport);
1477 shash_add(&p->port_by_name, netdev_name, ofport);
1479 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1484 ofport_remove(struct ofproto *p, struct ofport *ofport)
1486 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1488 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1489 port_array_delete(&p->ports, odp_port);
1490 shash_delete(&p->port_by_name,
1491 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1493 ofproto_sflow_del_port(p->sflow, odp_port);
1498 ofport_free(struct ofport *ofport)
1501 netdev_close(ofport->netdev);
1507 update_port(struct ofproto *p, const char *devname)
1509 struct odp_port odp_port;
1510 struct ofport *old_ofport;
1511 struct ofport *new_ofport;
1514 COVERAGE_INC(ofproto_update_port);
1516 /* Query the datapath for port information. */
1517 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1519 /* Find the old ofport. */
1520 old_ofport = shash_find_data(&p->port_by_name, devname);
1523 /* There's no port named 'devname' but there might be a port with
1524 * the same port number. This could happen if a port is deleted
1525 * and then a new one added in its place very quickly, or if a port
1526 * is renamed. In the former case we want to send an OFPPR_DELETE
1527 * and an OFPPR_ADD, and in the latter case we want to send a
1528 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1529 * the old port's ifindex against the new port, or perhaps less
1530 * reliably but more portably by comparing the old port's MAC
1531 * against the new port's MAC. However, this code isn't that smart
1532 * and always sends an OFPPR_MODIFY (XXX). */
1533 old_ofport = port_array_get(&p->ports, odp_port.port);
1535 } else if (error != ENOENT && error != ENODEV) {
1536 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1537 "%s", strerror(error));
1541 /* Create a new ofport. */
1542 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1544 /* Eliminate a few pathological cases. */
1545 if (!old_ofport && !new_ofport) {
1547 } else if (old_ofport && new_ofport) {
1548 /* Most of the 'config' bits are OpenFlow soft state, but
1549 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1550 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1551 * leaves the other bits 0.) */
1552 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1554 if (ofport_equal(old_ofport, new_ofport)) {
1555 /* False alarm--no change. */
1556 ofport_free(new_ofport);
1561 /* Now deal with the normal cases. */
1563 ofport_remove(p, old_ofport);
1566 ofport_install(p, new_ofport);
1568 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1569 (!old_ofport ? OFPPR_ADD
1570 : !new_ofport ? OFPPR_DELETE
1572 ofport_free(old_ofport);
1574 /* Update port groups. */
1575 refresh_port_groups(p);
1579 init_ports(struct ofproto *p)
1581 struct odp_port *ports;
1586 error = dpif_port_list(p->dpif, &ports, &n_ports);
1591 for (i = 0; i < n_ports; i++) {
1592 const struct odp_port *odp_port = &ports[i];
1593 if (!ofport_conflicts(p, odp_port)) {
1594 struct ofport *ofport = make_ofport(odp_port);
1596 ofport_install(p, ofport);
1601 refresh_port_groups(p);
1605 static struct ofconn *
1606 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1608 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1609 ofconn->ofproto = p;
1610 list_push_back(&p->all_conns, &ofconn->node);
1611 ofconn->rconn = rconn;
1612 ofconn->type = type;
1613 ofconn->role = NX_ROLE_OTHER;
1614 ofconn->packet_in_counter = rconn_packet_counter_create ();
1615 ofconn->pktbuf = NULL;
1616 ofconn->miss_send_len = 0;
1617 ofconn->reply_counter = rconn_packet_counter_create ();
1622 ofconn_destroy(struct ofconn *ofconn)
1624 if (ofconn->type == OFCONN_CONTROLLER) {
1625 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1627 discovery_destroy(ofconn->discovery);
1629 list_remove(&ofconn->node);
1630 switch_status_unregister(ofconn->ss);
1631 rconn_destroy(ofconn->rconn);
1632 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1633 rconn_packet_counter_destroy(ofconn->reply_counter);
1634 pktbuf_destroy(ofconn->pktbuf);
1639 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1644 if (ofconn->discovery) {
1645 char *controller_name;
1646 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1647 discovery_question_connectivity(ofconn->discovery);
1649 if (discovery_run(ofconn->discovery, &controller_name)) {
1650 if (controller_name) {
1651 char *ofconn_name = ofconn_make_name(p, controller_name);
1652 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1655 rconn_disconnect(ofconn->rconn);
1660 for (i = 0; i < N_SCHEDULERS; i++) {
1661 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1664 rconn_run(ofconn->rconn);
1666 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1667 /* Limit the number of iterations to prevent other tasks from
1669 for (iteration = 0; iteration < 50; iteration++) {
1670 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1675 fail_open_maybe_recover(p->fail_open);
1677 handle_openflow(ofconn, p, of_msg);
1678 ofpbuf_delete(of_msg);
1682 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1683 ofconn_destroy(ofconn);
1688 ofconn_wait(struct ofconn *ofconn)
1692 if (ofconn->discovery) {
1693 discovery_wait(ofconn->discovery);
1695 for (i = 0; i < N_SCHEDULERS; i++) {
1696 pinsched_wait(ofconn->schedulers[i]);
1698 rconn_run_wait(ofconn->rconn);
1699 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1700 rconn_recv_wait(ofconn->rconn);
1702 COVERAGE_INC(ofproto_ofconn_stuck);
1706 /* Returns true if 'ofconn' should receive asynchronous messages. */
1708 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1710 if (ofconn->type == OFCONN_CONTROLLER) {
1711 /* Ordinary controllers always get asynchronous messages unless they
1712 * have configured themselves as "slaves". */
1713 return ofconn->role != NX_ROLE_SLAVE;
1715 /* Transient connections don't get asynchronous messages unless they
1716 * have explicitly asked for them by setting a nonzero miss send
1718 return ofconn->miss_send_len > 0;
1722 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1723 * and 'target', suitable for use in log messages for identifying the
1726 * The name is dynamically allocated. The caller should free it (with free())
1727 * when it is no longer needed. */
1729 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1731 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1734 /* Caller is responsible for initializing the 'cr' member of the returned
1736 static struct rule *
1737 rule_create(struct ofproto *ofproto, struct rule *super,
1738 const union ofp_action *actions, size_t n_actions,
1739 uint16_t idle_timeout, uint16_t hard_timeout,
1740 uint64_t flow_cookie, bool send_flow_removed)
1742 struct rule *rule = xzalloc(sizeof *rule);
1743 rule->idle_timeout = idle_timeout;
1744 rule->hard_timeout = hard_timeout;
1745 rule->flow_cookie = flow_cookie;
1746 rule->used = rule->created = time_msec();
1747 rule->send_flow_removed = send_flow_removed;
1748 rule->super = super;
1750 list_push_back(&super->list, &rule->list);
1752 list_init(&rule->list);
1754 rule->n_actions = n_actions;
1755 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1756 netflow_flow_clear(&rule->nf_flow);
1757 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1762 static struct rule *
1763 rule_from_cls_rule(const struct cls_rule *cls_rule)
1765 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1769 rule_free(struct rule *rule)
1771 free(rule->actions);
1772 free(rule->odp_actions);
1776 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1777 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1778 * through all of its subrules and revalidates them, destroying any that no
1779 * longer has a super-rule (which is probably all of them).
1781 * Before calling this function, the caller must make have removed 'rule' from
1782 * the classifier. If 'rule' is an exact-match rule, the caller is also
1783 * responsible for ensuring that it has been uninstalled from the datapath. */
1785 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1788 struct rule *subrule, *next;
1789 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1790 revalidate_rule(ofproto, subrule);
1793 list_remove(&rule->list);
1799 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1801 const union ofp_action *oa;
1802 struct actions_iterator i;
1804 if (out_port == htons(OFPP_NONE)) {
1807 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1808 oa = actions_next(&i)) {
1809 if (action_outputs_to_port(oa, out_port)) {
1816 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1817 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1819 * The flow that 'packet' actually contains does not need to actually match
1820 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1821 * the packet and byte counters for 'rule' will be credited for the packet sent
1822 * out whether or not the packet actually matches 'rule'.
1824 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1825 * the caller must already have accurately composed ODP actions for it given
1826 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1827 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1828 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1829 * actions and apply them to 'packet'. */
1831 rule_execute(struct ofproto *ofproto, struct rule *rule,
1832 struct ofpbuf *packet, const flow_t *flow)
1834 const union odp_action *actions;
1836 struct odp_actions a;
1838 /* Grab or compose the ODP actions.
1840 * The special case for an exact-match 'rule' where 'flow' is not the
1841 * rule's flow is important to avoid, e.g., sending a packet out its input
1842 * port simply because the ODP actions were composed for the wrong
1844 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1845 struct rule *super = rule->super ? rule->super : rule;
1846 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1847 packet, &a, NULL, 0, NULL)) {
1850 actions = a.actions;
1851 n_actions = a.n_actions;
1853 actions = rule->odp_actions;
1854 n_actions = rule->n_odp_actions;
1857 /* Execute the ODP actions. */
1858 if (!dpif_execute(ofproto->dpif, flow->in_port,
1859 actions, n_actions, packet)) {
1860 struct odp_flow_stats stats;
1861 flow_extract_stats(flow, packet, &stats);
1862 update_stats(ofproto, rule, &stats);
1863 rule->used = time_msec();
1864 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1869 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1872 struct rule *displaced_rule;
1874 /* Insert the rule in the classifier. */
1875 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1876 if (!rule->cr.wc.wildcards) {
1877 rule_make_actions(p, rule, packet);
1880 /* Send the packet and credit it to the rule. */
1883 flow_extract(packet, 0, in_port, &flow);
1884 rule_execute(p, rule, packet, &flow);
1887 /* Install the rule in the datapath only after sending the packet, to
1888 * avoid packet reordering. */
1889 if (rule->cr.wc.wildcards) {
1890 COVERAGE_INC(ofproto_add_wc_flow);
1891 p->need_revalidate = true;
1893 rule_install(p, rule, displaced_rule);
1896 /* Free the rule that was displaced, if any. */
1897 if (displaced_rule) {
1898 rule_destroy(p, displaced_rule);
1902 static struct rule *
1903 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1906 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1907 rule->idle_timeout, rule->hard_timeout,
1909 COVERAGE_INC(ofproto_subrule_create);
1910 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1911 : rule->cr.priority), &subrule->cr);
1912 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1918 rule_remove(struct ofproto *ofproto, struct rule *rule)
1920 if (rule->cr.wc.wildcards) {
1921 COVERAGE_INC(ofproto_del_wc_flow);
1922 ofproto->need_revalidate = true;
1924 rule_uninstall(ofproto, rule);
1926 classifier_remove(&ofproto->cls, &rule->cr);
1927 rule_destroy(ofproto, rule);
1930 /* Returns true if the actions changed, false otherwise. */
1932 rule_make_actions(struct ofproto *p, struct rule *rule,
1933 const struct ofpbuf *packet)
1935 const struct rule *super;
1936 struct odp_actions a;
1939 assert(!rule->cr.wc.wildcards);
1941 super = rule->super ? rule->super : rule;
1943 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1944 packet, &a, &rule->tags, &rule->may_install,
1945 &rule->nf_flow.output_iface);
1947 actions_len = a.n_actions * sizeof *a.actions;
1948 if (rule->n_odp_actions != a.n_actions
1949 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1950 COVERAGE_INC(ofproto_odp_unchanged);
1951 free(rule->odp_actions);
1952 rule->n_odp_actions = a.n_actions;
1953 rule->odp_actions = xmemdup(a.actions, actions_len);
1961 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1962 struct odp_flow_put *put)
1964 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1965 put->flow.key = rule->cr.flow;
1966 put->flow.actions = rule->odp_actions;
1967 put->flow.n_actions = rule->n_odp_actions;
1968 put->flow.flags = 0;
1970 return dpif_flow_put(ofproto->dpif, put);
1974 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1976 assert(!rule->cr.wc.wildcards);
1978 if (rule->may_install) {
1979 struct odp_flow_put put;
1980 if (!do_put_flow(p, rule,
1981 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1983 rule->installed = true;
1984 if (displaced_rule) {
1985 update_stats(p, displaced_rule, &put.flow.stats);
1986 rule_post_uninstall(p, displaced_rule);
1989 } else if (displaced_rule) {
1990 rule_uninstall(p, displaced_rule);
1995 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1997 if (rule->installed) {
1998 struct odp_flow_put put;
1999 COVERAGE_INC(ofproto_dp_missed);
2000 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2002 rule_install(ofproto, rule, NULL);
2007 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2009 bool actions_changed;
2010 uint16_t new_out_iface, old_out_iface;
2012 old_out_iface = rule->nf_flow.output_iface;
2013 actions_changed = rule_make_actions(ofproto, rule, NULL);
2015 if (rule->may_install) {
2016 if (rule->installed) {
2017 if (actions_changed) {
2018 struct odp_flow_put put;
2019 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2020 | ODPPF_ZERO_STATS, &put);
2021 update_stats(ofproto, rule, &put.flow.stats);
2023 /* Temporarily set the old output iface so that NetFlow
2024 * messages have the correct output interface for the old
2026 new_out_iface = rule->nf_flow.output_iface;
2027 rule->nf_flow.output_iface = old_out_iface;
2028 rule_post_uninstall(ofproto, rule);
2029 rule->nf_flow.output_iface = new_out_iface;
2032 rule_install(ofproto, rule, NULL);
2035 rule_uninstall(ofproto, rule);
2040 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2042 uint64_t total_bytes = rule->byte_count + extra_bytes;
2044 if (ofproto->ofhooks->account_flow_cb
2045 && total_bytes > rule->accounted_bytes)
2047 ofproto->ofhooks->account_flow_cb(
2048 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2049 total_bytes - rule->accounted_bytes, ofproto->aux);
2050 rule->accounted_bytes = total_bytes;
2055 rule_uninstall(struct ofproto *p, struct rule *rule)
2057 assert(!rule->cr.wc.wildcards);
2058 if (rule->installed) {
2059 struct odp_flow odp_flow;
2061 odp_flow.key = rule->cr.flow;
2062 odp_flow.actions = NULL;
2063 odp_flow.n_actions = 0;
2065 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2066 update_stats(p, rule, &odp_flow.stats);
2068 rule->installed = false;
2070 rule_post_uninstall(p, rule);
2075 is_controller_rule(struct rule *rule)
2077 /* If the only action is send to the controller then don't report
2078 * NetFlow expiration messages since it is just part of the control
2079 * logic for the network and not real traffic. */
2083 && rule->super->n_actions == 1
2084 && action_outputs_to_port(&rule->super->actions[0],
2085 htons(OFPP_CONTROLLER)));
2089 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2091 struct rule *super = rule->super;
2093 rule_account(ofproto, rule, 0);
2095 if (ofproto->netflow && !is_controller_rule(rule)) {
2096 struct ofexpired expired;
2097 expired.flow = rule->cr.flow;
2098 expired.packet_count = rule->packet_count;
2099 expired.byte_count = rule->byte_count;
2100 expired.used = rule->used;
2101 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2104 super->packet_count += rule->packet_count;
2105 super->byte_count += rule->byte_count;
2107 /* Reset counters to prevent double counting if the rule ever gets
2109 rule->packet_count = 0;
2110 rule->byte_count = 0;
2111 rule->accounted_bytes = 0;
2113 netflow_flow_clear(&rule->nf_flow);
2118 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2119 struct rconn_packet_counter *counter)
2121 update_openflow_length(msg);
2122 if (rconn_send(ofconn->rconn, msg, counter)) {
2128 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2129 int error, const void *data, size_t len)
2132 struct ofp_error_msg *oem;
2134 if (!(error >> 16)) {
2135 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2140 COVERAGE_INC(ofproto_error);
2141 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2142 oh ? oh->xid : 0, &buf);
2143 oem->type = htons((unsigned int) error >> 16);
2144 oem->code = htons(error & 0xffff);
2145 memcpy(oem->data, data, len);
2146 queue_tx(buf, ofconn, ofconn->reply_counter);
2150 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2153 size_t oh_length = ntohs(oh->length);
2154 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2158 hton_ofp_phy_port(struct ofp_phy_port *opp)
2160 opp->port_no = htons(opp->port_no);
2161 opp->config = htonl(opp->config);
2162 opp->state = htonl(opp->state);
2163 opp->curr = htonl(opp->curr);
2164 opp->advertised = htonl(opp->advertised);
2165 opp->supported = htonl(opp->supported);
2166 opp->peer = htonl(opp->peer);
2170 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2172 struct ofp_header *rq = oh;
2173 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2178 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2179 struct ofp_header *oh)
2181 struct ofp_switch_features *osf;
2183 unsigned int port_no;
2184 struct ofport *port;
2186 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2187 osf->datapath_id = htonll(p->datapath_id);
2188 osf->n_buffers = htonl(pktbuf_capacity());
2190 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2191 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2192 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2193 (1u << OFPAT_SET_VLAN_VID) |
2194 (1u << OFPAT_SET_VLAN_PCP) |
2195 (1u << OFPAT_STRIP_VLAN) |
2196 (1u << OFPAT_SET_DL_SRC) |
2197 (1u << OFPAT_SET_DL_DST) |
2198 (1u << OFPAT_SET_NW_SRC) |
2199 (1u << OFPAT_SET_NW_DST) |
2200 (1u << OFPAT_SET_NW_TOS) |
2201 (1u << OFPAT_SET_TP_SRC) |
2202 (1u << OFPAT_SET_TP_DST) |
2203 (1u << OFPAT_ENQUEUE));
2205 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2206 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2209 queue_tx(buf, ofconn, ofconn->reply_counter);
2214 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2215 struct ofp_header *oh)
2218 struct ofp_switch_config *osc;
2222 /* Figure out flags. */
2223 dpif_get_drop_frags(p->dpif, &drop_frags);
2224 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2227 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2228 osc->flags = htons(flags);
2229 osc->miss_send_len = htons(ofconn->miss_send_len);
2230 queue_tx(buf, ofconn, ofconn->reply_counter);
2236 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2237 struct ofp_switch_config *osc)
2242 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2246 flags = ntohs(osc->flags);
2248 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2249 switch (flags & OFPC_FRAG_MASK) {
2250 case OFPC_FRAG_NORMAL:
2251 dpif_set_drop_frags(p->dpif, false);
2253 case OFPC_FRAG_DROP:
2254 dpif_set_drop_frags(p->dpif, true);
2257 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2263 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2269 add_output_group_action(struct odp_actions *actions, uint16_t group,
2270 uint16_t *nf_output_iface)
2272 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2274 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2275 *nf_output_iface = NF_OUT_FLOOD;
2280 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2282 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2283 a->controller.arg = max_len;
2286 struct action_xlate_ctx {
2288 flow_t flow; /* Flow to which these actions correspond. */
2289 int recurse; /* Recursion level, via xlate_table_action. */
2290 struct ofproto *ofproto;
2291 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2292 * null pointer if we are revalidating
2293 * without a packet to refer to. */
2296 struct odp_actions *out; /* Datapath actions. */
2297 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2298 bool may_set_up_flow; /* True ordinarily; false if the actions must
2299 * be reassessed for every packet. */
2300 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2303 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2304 struct action_xlate_ctx *ctx);
2307 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2309 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2312 if (ofport->opp.config & OFPPC_NO_FWD) {
2313 /* Forwarding disabled on port. */
2318 * We don't have an ofport record for this port, but it doesn't hurt to
2319 * allow forwarding to it anyhow. Maybe such a port will appear later
2320 * and we're pre-populating the flow table.
2324 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2325 ctx->nf_output_iface = port;
2328 static struct rule *
2329 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2332 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2334 /* The rule we found might not be valid, since we could be in need of
2335 * revalidation. If it is not valid, don't return it. */
2338 && ofproto->need_revalidate
2339 && !revalidate_rule(ofproto, rule)) {
2340 COVERAGE_INC(ofproto_invalidated);
2348 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2350 if (!ctx->recurse) {
2351 uint16_t old_in_port;
2354 /* Look up a flow with 'in_port' as the input port. Then restore the
2355 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2356 * have surprising behavior). */
2357 old_in_port = ctx->flow.in_port;
2358 ctx->flow.in_port = in_port;
2359 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2360 ctx->flow.in_port = old_in_port;
2368 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2375 xlate_output_action__(struct action_xlate_ctx *ctx,
2376 uint16_t port, uint16_t max_len)
2379 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2381 ctx->nf_output_iface = NF_OUT_DROP;
2385 add_output_action(ctx, ctx->flow.in_port);
2388 xlate_table_action(ctx, ctx->flow.in_port);
2391 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2392 ctx->out, ctx->tags,
2393 &ctx->nf_output_iface,
2394 ctx->ofproto->aux)) {
2395 COVERAGE_INC(ofproto_uninstallable);
2396 ctx->may_set_up_flow = false;
2400 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2401 &ctx->nf_output_iface);
2404 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2406 case OFPP_CONTROLLER:
2407 add_controller_action(ctx->out, max_len);
2410 add_output_action(ctx, ODPP_LOCAL);
2413 odp_port = ofp_port_to_odp_port(port);
2414 if (odp_port != ctx->flow.in_port) {
2415 add_output_action(ctx, odp_port);
2420 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2421 ctx->nf_output_iface = NF_OUT_FLOOD;
2422 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2423 ctx->nf_output_iface = prev_nf_output_iface;
2424 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2425 ctx->nf_output_iface != NF_OUT_FLOOD) {
2426 ctx->nf_output_iface = NF_OUT_MULTI;
2431 xlate_output_action(struct action_xlate_ctx *ctx,
2432 const struct ofp_action_output *oao)
2434 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2437 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2438 * optimization, because we're going to add another action that sets the
2439 * priority immediately after, or because there are no actions following the
2442 remove_pop_action(struct action_xlate_ctx *ctx)
2444 size_t n = ctx->out->n_actions;
2445 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2446 ctx->out->n_actions--;
2451 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2452 const struct ofp_action_enqueue *oae)
2454 uint16_t ofp_port, odp_port;
2458 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2461 /* Fall back to ordinary output action. */
2462 xlate_output_action__(ctx, ntohs(oae->port), 0);
2466 /* Figure out ODP output port. */
2467 ofp_port = ntohs(oae->port);
2468 if (ofp_port != OFPP_IN_PORT) {
2469 odp_port = ofp_port_to_odp_port(ofp_port);
2471 odp_port = ctx->flow.in_port;
2474 /* Add ODP actions. */
2475 remove_pop_action(ctx);
2476 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2478 add_output_action(ctx, odp_port);
2479 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2481 /* Update NetFlow output port. */
2482 if (ctx->nf_output_iface == NF_OUT_DROP) {
2483 ctx->nf_output_iface = odp_port;
2484 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2485 ctx->nf_output_iface = NF_OUT_MULTI;
2490 xlate_nicira_action(struct action_xlate_ctx *ctx,
2491 const struct nx_action_header *nah)
2493 const struct nx_action_resubmit *nar;
2494 const struct nx_action_set_tunnel *nast;
2495 union odp_action *oa;
2496 int subtype = ntohs(nah->subtype);
2498 assert(nah->vendor == htonl(NX_VENDOR_ID));
2500 case NXAST_RESUBMIT:
2501 nar = (const struct nx_action_resubmit *) nah;
2502 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2505 case NXAST_SET_TUNNEL:
2506 nast = (const struct nx_action_set_tunnel *) nah;
2507 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2508 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2511 /* If you add a new action here that modifies flow data, don't forget to
2512 * update the flow key in ctx->flow at the same time. */
2515 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2521 do_xlate_actions(const union ofp_action *in, size_t n_in,
2522 struct action_xlate_ctx *ctx)
2524 struct actions_iterator iter;
2525 const union ofp_action *ia;
2526 const struct ofport *port;
2528 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2529 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2530 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2531 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2532 /* Drop this flow. */
2536 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2537 uint16_t type = ntohs(ia->type);
2538 union odp_action *oa;
2542 xlate_output_action(ctx, &ia->output);
2545 case OFPAT_SET_VLAN_VID:
2546 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2547 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2550 case OFPAT_SET_VLAN_PCP:
2551 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2552 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2555 case OFPAT_STRIP_VLAN:
2556 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2557 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2558 ctx->flow.dl_vlan_pcp = 0;
2561 case OFPAT_SET_DL_SRC:
2562 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2563 memcpy(oa->dl_addr.dl_addr,
2564 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2565 memcpy(ctx->flow.dl_src,
2566 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2569 case OFPAT_SET_DL_DST:
2570 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2571 memcpy(oa->dl_addr.dl_addr,
2572 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2573 memcpy(ctx->flow.dl_dst,
2574 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2577 case OFPAT_SET_NW_SRC:
2578 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2579 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2582 case OFPAT_SET_NW_DST:
2583 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2584 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2587 case OFPAT_SET_NW_TOS:
2588 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2589 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2592 case OFPAT_SET_TP_SRC:
2593 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2594 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2597 case OFPAT_SET_TP_DST:
2598 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2599 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2603 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2607 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2611 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2618 xlate_actions(const union ofp_action *in, size_t n_in,
2619 const flow_t *flow, struct ofproto *ofproto,
2620 const struct ofpbuf *packet,
2621 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2622 uint16_t *nf_output_iface)
2624 tag_type no_tags = 0;
2625 struct action_xlate_ctx ctx;
2626 COVERAGE_INC(ofproto_ofp2odp);
2627 odp_actions_init(out);
2630 ctx.ofproto = ofproto;
2631 ctx.packet = packet;
2633 ctx.tags = tags ? tags : &no_tags;
2634 ctx.may_set_up_flow = true;
2635 ctx.nf_output_iface = NF_OUT_DROP;
2636 do_xlate_actions(in, n_in, &ctx);
2637 remove_pop_action(&ctx);
2639 /* Check with in-band control to see if we're allowed to set up this
2641 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2642 ctx.may_set_up_flow = false;
2645 if (may_set_up_flow) {
2646 *may_set_up_flow = ctx.may_set_up_flow;
2648 if (nf_output_iface) {
2649 *nf_output_iface = ctx.nf_output_iface;
2651 if (odp_actions_overflow(out)) {
2652 odp_actions_init(out);
2653 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2658 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2659 * error message code (composed with ofp_mkerr()) for the caller to propagate
2660 * upward. Otherwise, returns 0.
2662 * 'oh' is used to make log messages more informative. */
2664 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2666 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2667 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2670 type_name = ofp_message_type_to_string(oh->type);
2671 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2675 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2682 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2683 struct ofp_header *oh)
2685 struct ofp_packet_out *opo;
2686 struct ofpbuf payload, *buffer;
2687 struct odp_actions actions;
2693 error = reject_slave_controller(ofconn, oh);
2698 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2702 opo = (struct ofp_packet_out *) oh;
2704 COVERAGE_INC(ofproto_packet_out);
2705 if (opo->buffer_id != htonl(UINT32_MAX)) {
2706 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2708 if (error || !buffer) {
2716 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2717 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2718 &flow, p, &payload, &actions, NULL, NULL, NULL);
2723 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2725 ofpbuf_delete(buffer);
2731 update_port_config(struct ofproto *p, struct ofport *port,
2732 uint32_t config, uint32_t mask)
2734 mask &= config ^ port->opp.config;
2735 if (mask & OFPPC_PORT_DOWN) {
2736 if (config & OFPPC_PORT_DOWN) {
2737 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2739 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2742 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2743 if (mask & REVALIDATE_BITS) {
2744 COVERAGE_INC(ofproto_costly_flags);
2745 port->opp.config ^= mask & REVALIDATE_BITS;
2746 p->need_revalidate = true;
2748 #undef REVALIDATE_BITS
2749 if (mask & OFPPC_NO_FLOOD) {
2750 port->opp.config ^= OFPPC_NO_FLOOD;
2751 refresh_port_groups(p);
2753 if (mask & OFPPC_NO_PACKET_IN) {
2754 port->opp.config ^= OFPPC_NO_PACKET_IN;
2759 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2760 struct ofp_header *oh)
2762 const struct ofp_port_mod *opm;
2763 struct ofport *port;
2766 error = reject_slave_controller(ofconn, oh);
2770 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2774 opm = (struct ofp_port_mod *) oh;
2776 port = port_array_get(&p->ports,
2777 ofp_port_to_odp_port(ntohs(opm->port_no)));
2779 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2780 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2781 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2783 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2784 if (opm->advertise) {
2785 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2791 static struct ofpbuf *
2792 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2794 struct ofp_stats_reply *osr;
2797 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2798 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2800 osr->flags = htons(0);
2804 static struct ofpbuf *
2805 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2807 return make_stats_reply(request->header.xid, request->type, body_len);
2811 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2813 struct ofpbuf *msg = *msgp;
2814 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2815 if (nbytes + msg->size > UINT16_MAX) {
2816 struct ofp_stats_reply *reply = msg->data;
2817 reply->flags = htons(OFPSF_REPLY_MORE);
2818 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2819 queue_tx(msg, ofconn, ofconn->reply_counter);
2821 return ofpbuf_put_uninit(*msgp, nbytes);
2825 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2826 struct ofp_stats_request *request)
2828 struct ofp_desc_stats *ods;
2831 msg = start_stats_reply(request, sizeof *ods);
2832 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2833 memset(ods, 0, sizeof *ods);
2834 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2835 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2836 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2837 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2838 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2839 queue_tx(msg, ofconn, ofconn->reply_counter);
2845 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2847 struct rule *rule = rule_from_cls_rule(cls_rule);
2848 int *n_subrules = n_subrules_;
2856 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2857 struct ofp_stats_request *request)
2859 struct ofp_table_stats *ots;
2861 struct odp_stats dpstats;
2862 int n_exact, n_subrules, n_wild;
2864 msg = start_stats_reply(request, sizeof *ots * 2);
2866 /* Count rules of various kinds. */
2868 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2869 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2870 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2873 dpif_get_dp_stats(p->dpif, &dpstats);
2874 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2875 memset(ots, 0, sizeof *ots);
2876 ots->table_id = TABLEID_HASH;
2877 strcpy(ots->name, "hash");
2878 ots->wildcards = htonl(0);
2879 ots->max_entries = htonl(dpstats.max_capacity);
2880 ots->active_count = htonl(n_exact);
2881 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2883 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2885 /* Classifier table. */
2886 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2887 memset(ots, 0, sizeof *ots);
2888 ots->table_id = TABLEID_CLASSIFIER;
2889 strcpy(ots->name, "classifier");
2890 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2892 ots->max_entries = htonl(65536);
2893 ots->active_count = htonl(n_wild);
2894 ots->lookup_count = htonll(0); /* XXX */
2895 ots->matched_count = htonll(0); /* XXX */
2897 queue_tx(msg, ofconn, ofconn->reply_counter);
2902 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2903 struct ofpbuf **msgp)
2905 struct netdev_stats stats;
2906 struct ofp_port_stats *ops;
2908 /* Intentionally ignore return value, since errors will set
2909 * 'stats' to all-1s, which is correct for OpenFlow, and
2910 * netdev_get_stats() will log errors. */
2911 netdev_get_stats(port->netdev, &stats);
2913 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2914 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2915 memset(ops->pad, 0, sizeof ops->pad);
2916 ops->rx_packets = htonll(stats.rx_packets);
2917 ops->tx_packets = htonll(stats.tx_packets);
2918 ops->rx_bytes = htonll(stats.rx_bytes);
2919 ops->tx_bytes = htonll(stats.tx_bytes);
2920 ops->rx_dropped = htonll(stats.rx_dropped);
2921 ops->tx_dropped = htonll(stats.tx_dropped);
2922 ops->rx_errors = htonll(stats.rx_errors);
2923 ops->tx_errors = htonll(stats.tx_errors);
2924 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2925 ops->rx_over_err = htonll(stats.rx_over_errors);
2926 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2927 ops->collisions = htonll(stats.collisions);
2931 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2932 struct ofp_stats_request *osr,
2935 struct ofp_port_stats_request *psr;
2936 struct ofp_port_stats *ops;
2938 struct ofport *port;
2939 unsigned int port_no;
2941 if (arg_size != sizeof *psr) {
2942 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2944 psr = (struct ofp_port_stats_request *) osr->body;
2946 msg = start_stats_reply(osr, sizeof *ops * 16);
2947 if (psr->port_no != htons(OFPP_NONE)) {
2948 port = port_array_get(&p->ports,
2949 ofp_port_to_odp_port(ntohs(psr->port_no)));
2951 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2954 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2955 append_port_stat(port, port_no, ofconn, &msg);
2959 queue_tx(msg, ofconn, ofconn->reply_counter);
2963 struct flow_stats_cbdata {
2964 struct ofproto *ofproto;
2965 struct ofconn *ofconn;
2970 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2971 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2972 * returned statistic include statistics for all of 'rule''s subrules. */
2974 query_stats(struct ofproto *p, struct rule *rule,
2975 uint64_t *packet_countp, uint64_t *byte_countp)
2977 uint64_t packet_count, byte_count;
2978 struct rule *subrule;
2979 struct odp_flow *odp_flows;
2982 /* Start from historical data for 'rule' itself that are no longer tracked
2983 * by the datapath. This counts, for example, subrules that have
2985 packet_count = rule->packet_count;
2986 byte_count = rule->byte_count;
2988 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2989 * wildcarded then on all of its subrules.
2991 * Also, add any statistics that are not tracked by the datapath for each
2992 * subrule. This includes, for example, statistics for packets that were
2993 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2995 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2996 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2997 if (rule->cr.wc.wildcards) {
2999 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3000 odp_flows[i++].key = subrule->cr.flow;
3001 packet_count += subrule->packet_count;
3002 byte_count += subrule->byte_count;
3005 odp_flows[0].key = rule->cr.flow;
3008 /* Fetch up-to-date statistics from the datapath and add them in. */
3009 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3011 for (i = 0; i < n_odp_flows; i++) {
3012 struct odp_flow *odp_flow = &odp_flows[i];
3013 packet_count += odp_flow->stats.n_packets;
3014 byte_count += odp_flow->stats.n_bytes;
3019 /* Return the stats to the caller. */
3020 *packet_countp = packet_count;
3021 *byte_countp = byte_count;
3025 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3027 struct rule *rule = rule_from_cls_rule(rule_);
3028 struct flow_stats_cbdata *cbdata = cbdata_;
3029 struct ofp_flow_stats *ofs;
3030 uint64_t packet_count, byte_count;
3031 size_t act_len, len;
3032 long long int tdiff = time_msec() - rule->created;
3033 uint32_t sec = tdiff / 1000;
3034 uint32_t msec = tdiff - (sec * 1000);
3036 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3040 act_len = sizeof *rule->actions * rule->n_actions;
3041 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3043 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3045 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3046 ofs->length = htons(len);
3047 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3049 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3050 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3051 ofs->duration_sec = htonl(sec);
3052 ofs->duration_nsec = htonl(msec * 1000000);
3053 ofs->cookie = rule->flow_cookie;
3054 ofs->priority = htons(rule->cr.priority);
3055 ofs->idle_timeout = htons(rule->idle_timeout);
3056 ofs->hard_timeout = htons(rule->hard_timeout);
3057 memset(ofs->pad2, 0, sizeof ofs->pad2);
3058 ofs->packet_count = htonll(packet_count);
3059 ofs->byte_count = htonll(byte_count);
3060 memcpy(ofs->actions, rule->actions, act_len);
3064 table_id_to_include(uint8_t table_id)
3066 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3067 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3068 : table_id == 0xff ? CLS_INC_ALL
3073 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3074 const struct ofp_stats_request *osr,
3077 struct ofp_flow_stats_request *fsr;
3078 struct flow_stats_cbdata cbdata;
3079 struct cls_rule target;
3081 if (arg_size != sizeof *fsr) {
3082 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3084 fsr = (struct ofp_flow_stats_request *) osr->body;
3086 COVERAGE_INC(ofproto_flows_req);
3088 cbdata.ofconn = ofconn;
3089 cbdata.out_port = fsr->out_port;
3090 cbdata.msg = start_stats_reply(osr, 1024);
3091 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3092 classifier_for_each_match(&p->cls, &target,
3093 table_id_to_include(fsr->table_id),
3094 flow_stats_cb, &cbdata);
3095 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3099 struct flow_stats_ds_cbdata {
3100 struct ofproto *ofproto;
3105 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3107 struct rule *rule = rule_from_cls_rule(rule_);
3108 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3109 struct ds *results = cbdata->results;
3110 struct ofp_match match;
3111 uint64_t packet_count, byte_count;
3112 size_t act_len = sizeof *rule->actions * rule->n_actions;
3114 /* Don't report on subrules. */
3115 if (rule->super != NULL) {
3119 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3120 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3121 cbdata->ofproto->tun_id_from_cookie, &match);
3123 ds_put_format(results, "duration=%llds, ",
3124 (time_msec() - rule->created) / 1000);
3125 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3126 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3127 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3128 ofp_print_match(results, &match, true);
3129 ofp_print_actions(results, &rule->actions->header, act_len);
3130 ds_put_cstr(results, "\n");
3133 /* Adds a pretty-printed description of all flows to 'results', including
3134 * those marked hidden by secchan (e.g., by in-band control). */
3136 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3138 struct ofp_match match;
3139 struct cls_rule target;
3140 struct flow_stats_ds_cbdata cbdata;
3142 memset(&match, 0, sizeof match);
3143 match.wildcards = htonl(OVSFW_ALL);
3146 cbdata.results = results;
3148 cls_rule_from_match(&match, 0, false, 0, &target);
3149 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3150 flow_stats_ds_cb, &cbdata);
3153 struct aggregate_stats_cbdata {
3154 struct ofproto *ofproto;
3156 uint64_t packet_count;
3157 uint64_t byte_count;
3162 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3164 struct rule *rule = rule_from_cls_rule(rule_);
3165 struct aggregate_stats_cbdata *cbdata = cbdata_;
3166 uint64_t packet_count, byte_count;
3168 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3172 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3174 cbdata->packet_count += packet_count;
3175 cbdata->byte_count += byte_count;
3180 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3181 const struct ofp_stats_request *osr,
3184 struct ofp_aggregate_stats_request *asr;
3185 struct ofp_aggregate_stats_reply *reply;
3186 struct aggregate_stats_cbdata cbdata;
3187 struct cls_rule target;
3190 if (arg_size != sizeof *asr) {
3191 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3193 asr = (struct ofp_aggregate_stats_request *) osr->body;
3195 COVERAGE_INC(ofproto_agg_request);
3197 cbdata.out_port = asr->out_port;
3198 cbdata.packet_count = 0;
3199 cbdata.byte_count = 0;
3201 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3202 classifier_for_each_match(&p->cls, &target,
3203 table_id_to_include(asr->table_id),
3204 aggregate_stats_cb, &cbdata);
3206 msg = start_stats_reply(osr, sizeof *reply);
3207 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3208 reply->flow_count = htonl(cbdata.n_flows);
3209 reply->packet_count = htonll(cbdata.packet_count);
3210 reply->byte_count = htonll(cbdata.byte_count);
3211 queue_tx(msg, ofconn, ofconn->reply_counter);
3215 struct queue_stats_cbdata {
3216 struct ofconn *ofconn;
3222 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3223 const struct netdev_queue_stats *stats)
3225 struct ofp_queue_stats *reply;
3227 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3228 reply->port_no = htons(cbdata->port_no);
3229 memset(reply->pad, 0, sizeof reply->pad);
3230 reply->queue_id = htonl(queue_id);
3231 reply->tx_bytes = htonll(stats->tx_bytes);
3232 reply->tx_packets = htonll(stats->tx_packets);
3233 reply->tx_errors = htonll(stats->tx_errors);
3237 handle_queue_stats_dump_cb(uint32_t queue_id,
3238 struct netdev_queue_stats *stats,
3241 struct queue_stats_cbdata *cbdata = cbdata_;
3243 put_queue_stats(cbdata, queue_id, stats);
3247 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3249 struct queue_stats_cbdata *cbdata)
3251 cbdata->port_no = port_no;
3252 if (queue_id == OFPQ_ALL) {
3253 netdev_dump_queue_stats(port->netdev,
3254 handle_queue_stats_dump_cb, cbdata);
3256 struct netdev_queue_stats stats;
3258 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3259 put_queue_stats(cbdata, queue_id, &stats);
3264 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3265 const struct ofp_stats_request *osr,
3268 struct ofp_queue_stats_request *qsr;
3269 struct queue_stats_cbdata cbdata;
3270 struct ofport *port;
3271 unsigned int port_no;
3274 if (arg_size != sizeof *qsr) {
3275 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3277 qsr = (struct ofp_queue_stats_request *) osr->body;
3279 COVERAGE_INC(ofproto_queue_req);
3281 cbdata.ofconn = ofconn;
3282 cbdata.msg = start_stats_reply(osr, 128);
3284 port_no = ntohs(qsr->port_no);
3285 queue_id = ntohl(qsr->queue_id);
3286 if (port_no == OFPP_ALL) {
3287 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3288 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3290 } else if (port_no < ofproto->max_ports) {
3291 port = port_array_get(&ofproto->ports, port_no);
3293 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3296 ofpbuf_delete(cbdata.msg);
3297 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3299 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3305 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3306 struct ofp_header *oh)
3308 struct ofp_stats_request *osr;
3312 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3317 osr = (struct ofp_stats_request *) oh;
3319 switch (ntohs(osr->type)) {
3321 return handle_desc_stats_request(p, ofconn, osr);
3324 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3326 case OFPST_AGGREGATE:
3327 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3330 return handle_table_stats_request(p, ofconn, osr);
3333 return handle_port_stats_request(p, ofconn, osr, arg_size);
3336 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3339 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3342 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3346 static long long int
3347 msec_from_nsec(uint64_t sec, uint32_t nsec)
3349 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3353 update_time(struct ofproto *ofproto, struct rule *rule,
3354 const struct odp_flow_stats *stats)
3356 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3357 if (used > rule->used) {
3359 if (rule->super && used > rule->super->used) {
3360 rule->super->used = used;
3362 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3367 update_stats(struct ofproto *ofproto, struct rule *rule,
3368 const struct odp_flow_stats *stats)
3370 if (stats->n_packets) {
3371 update_time(ofproto, rule, stats);
3372 rule->packet_count += stats->n_packets;
3373 rule->byte_count += stats->n_bytes;
3374 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3379 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3380 * in which no matching flow already exists in the flow table.
3382 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3383 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3384 * code as encoded by ofp_mkerr() on failure.
3386 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3389 add_flow(struct ofproto *p, struct ofconn *ofconn,
3390 const struct ofp_flow_mod *ofm, size_t n_actions)
3392 struct ofpbuf *packet;
3397 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3401 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3403 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3404 ntohs(ofm->priority))) {
3405 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3409 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3410 n_actions, ntohs(ofm->idle_timeout),
3411 ntohs(ofm->hard_timeout), ofm->cookie,
3412 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3413 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3414 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3417 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3418 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3422 in_port = UINT16_MAX;
3425 rule_insert(p, rule, packet, in_port);
3426 ofpbuf_delete(packet);
3430 static struct rule *
3431 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3436 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3438 return rule_from_cls_rule(classifier_find_rule_exactly(
3439 &p->cls, &flow, wildcards,
3440 ntohs(ofm->priority)));
3444 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3445 struct rule *rule, const struct ofp_flow_mod *ofm)
3447 struct ofpbuf *packet;
3452 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3456 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3462 flow_extract(packet, 0, in_port, &flow);
3463 rule_execute(ofproto, rule, packet, &flow);
3464 ofpbuf_delete(packet);
3469 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3471 struct modify_flows_cbdata {
3472 struct ofproto *ofproto;
3473 const struct ofp_flow_mod *ofm;
3478 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3479 size_t n_actions, struct rule *);
3480 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3482 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3483 * encoded by ofp_mkerr() on failure.
3485 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3488 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3489 const struct ofp_flow_mod *ofm, size_t n_actions)
3491 struct modify_flows_cbdata cbdata;
3492 struct cls_rule target;
3496 cbdata.n_actions = n_actions;
3497 cbdata.match = NULL;
3499 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3502 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3503 modify_flows_cb, &cbdata);
3505 /* This credits the packet to whichever flow happened to happened to
3506 * match last. That's weird. Maybe we should do a lookup for the
3507 * flow that actually matches the packet? Who knows. */
3508 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3511 return add_flow(p, ofconn, ofm, n_actions);
3515 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3516 * code as encoded by ofp_mkerr() on failure.
3518 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3521 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3522 struct ofp_flow_mod *ofm, size_t n_actions)
3524 struct rule *rule = find_flow_strict(p, ofm);
3525 if (rule && !rule_is_hidden(rule)) {
3526 modify_flow(p, ofm, n_actions, rule);
3527 return send_buffered_packet(p, ofconn, rule, ofm);
3529 return add_flow(p, ofconn, ofm, n_actions);
3533 /* Callback for modify_flows_loose(). */
3535 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3537 struct rule *rule = rule_from_cls_rule(rule_);
3538 struct modify_flows_cbdata *cbdata = cbdata_;
3540 if (!rule_is_hidden(rule)) {
3541 cbdata->match = rule;
3542 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3546 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3547 * been identified as a flow in 'p''s flow table to be modified, by changing
3548 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3549 * ofp_action[] structures). */
3551 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3552 size_t n_actions, struct rule *rule)
3554 size_t actions_len = n_actions * sizeof *rule->actions;
3556 rule->flow_cookie = ofm->cookie;
3558 /* If the actions are the same, do nothing. */
3559 if (n_actions == rule->n_actions
3560 && !memcmp(ofm->actions, rule->actions, actions_len))
3565 /* Replace actions. */
3566 free(rule->actions);
3567 rule->actions = xmemdup(ofm->actions, actions_len);
3568 rule->n_actions = n_actions;
3570 /* Make sure that the datapath gets updated properly. */
3571 if (rule->cr.wc.wildcards) {
3572 COVERAGE_INC(ofproto_mod_wc_flow);
3573 p->need_revalidate = true;
3575 rule_update_actions(p, rule);
3581 /* OFPFC_DELETE implementation. */
3583 struct delete_flows_cbdata {
3584 struct ofproto *ofproto;
3588 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3589 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3591 /* Implements OFPFC_DELETE. */
3593 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3595 struct delete_flows_cbdata cbdata;
3596 struct cls_rule target;
3599 cbdata.out_port = ofm->out_port;
3601 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3604 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3605 delete_flows_cb, &cbdata);
3608 /* Implements OFPFC_DELETE_STRICT. */
3610 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3612 struct rule *rule = find_flow_strict(p, ofm);
3614 delete_flow(p, rule, ofm->out_port);
3618 /* Callback for delete_flows_loose(). */
3620 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3622 struct rule *rule = rule_from_cls_rule(rule_);
3623 struct delete_flows_cbdata *cbdata = cbdata_;
3625 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3628 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3629 * been identified as a flow to delete from 'p''s flow table, by deleting the
3630 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3633 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3634 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3635 * specified 'out_port'. */
3637 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3639 if (rule_is_hidden(rule)) {
3643 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3647 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3648 rule_remove(p, rule);
3652 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3653 struct ofp_flow_mod *ofm)
3655 struct ofp_match orig_match;
3659 error = reject_slave_controller(ofconn, &ofm->header);
3663 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3664 sizeof *ofm->actions, &n_actions);
3669 /* We do not support the emergency flow cache. It will hopefully
3670 * get dropped from OpenFlow in the near future. */
3671 if (ofm->flags & htons(OFPFF_EMERG)) {
3672 /* There isn't a good fit for an error code, so just state that the
3673 * flow table is full. */
3674 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3677 /* Normalize ofp->match. If normalization actually changes anything, then
3678 * log the differences. */
3679 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3680 orig_match = ofm->match;
3681 normalize_match(&ofm->match);
3682 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3683 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3684 if (!VLOG_DROP_INFO(&normal_rl)) {
3685 char *old = ofp_match_to_literal_string(&orig_match);
3686 char *new = ofp_match_to_literal_string(&ofm->match);
3687 VLOG_INFO("%s: normalization changed ofp_match, details:",
3688 rconn_get_name(ofconn->rconn));
3689 VLOG_INFO(" pre: %s", old);
3690 VLOG_INFO("post: %s", new);
3696 if (!ofm->match.wildcards) {
3697 ofm->priority = htons(UINT16_MAX);
3700 error = validate_actions((const union ofp_action *) ofm->actions,
3701 n_actions, p->max_ports);
3706 switch (ntohs(ofm->command)) {
3708 return add_flow(p, ofconn, ofm, n_actions);
3711 return modify_flows_loose(p, ofconn, ofm, n_actions);
3713 case OFPFC_MODIFY_STRICT:
3714 return modify_flow_strict(p, ofconn, ofm, n_actions);
3717 delete_flows_loose(p, ofm);
3720 case OFPFC_DELETE_STRICT:
3721 delete_flow_strict(p, ofm);
3725 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3730 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3734 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3739 p->tun_id_from_cookie = !!msg->set;
3744 handle_role_request(struct ofproto *ofproto,
3745 struct ofconn *ofconn, struct nicira_header *msg)
3747 struct nx_role_request *nrr;
3748 struct nx_role_request *reply;
3752 if (ntohs(msg->header.length) != sizeof *nrr) {
3753 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3754 ntohs(msg->header.length), sizeof *nrr);
3755 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3757 nrr = (struct nx_role_request *) msg;
3759 if (ofconn->type != OFCONN_CONTROLLER) {
3760 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3762 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3765 role = ntohl(nrr->role);
3766 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3767 && role != NX_ROLE_SLAVE) {
3768 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3770 /* There's no good error code for this. */
3771 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3774 if (role == NX_ROLE_MASTER) {
3775 struct ofconn *other;
3777 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3778 &ofproto->controllers) {
3779 if (other->role == NX_ROLE_MASTER) {
3780 other->role = NX_ROLE_SLAVE;
3784 ofconn->role = role;
3786 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3788 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3789 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3790 reply->role = htonl(role);
3791 queue_tx(buf, ofconn, ofconn->reply_counter);
3797 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3799 struct ofp_vendor_header *ovh = msg;
3800 struct nicira_header *nh;
3802 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3803 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3804 "(expected at least %zu)",
3805 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3806 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3808 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3809 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3811 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3812 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3813 "(expected at least %zu)",
3814 ntohs(ovh->header.length), sizeof(struct nicira_header));
3815 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3819 switch (ntohl(nh->subtype)) {
3820 case NXT_STATUS_REQUEST:
3821 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3824 case NXT_TUN_ID_FROM_COOKIE:
3825 return handle_tun_id_from_cookie(p, msg);
3827 case NXT_ROLE_REQUEST:
3828 return handle_role_request(p, ofconn, msg);
3831 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3835 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3837 struct ofp_header *ob;
3840 /* Currently, everything executes synchronously, so we can just
3841 * immediately send the barrier reply. */
3842 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3843 queue_tx(buf, ofconn, ofconn->reply_counter);
3848 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3849 struct ofpbuf *ofp_msg)
3851 struct ofp_header *oh = ofp_msg->data;
3854 COVERAGE_INC(ofproto_recv_openflow);
3856 case OFPT_ECHO_REQUEST:
3857 error = handle_echo_request(ofconn, oh);
3860 case OFPT_ECHO_REPLY:
3864 case OFPT_FEATURES_REQUEST:
3865 error = handle_features_request(p, ofconn, oh);
3868 case OFPT_GET_CONFIG_REQUEST:
3869 error = handle_get_config_request(p, ofconn, oh);
3872 case OFPT_SET_CONFIG:
3873 error = handle_set_config(p, ofconn, ofp_msg->data);
3876 case OFPT_PACKET_OUT:
3877 error = handle_packet_out(p, ofconn, ofp_msg->data);
3881 error = handle_port_mod(p, ofconn, oh);
3885 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3888 case OFPT_STATS_REQUEST:
3889 error = handle_stats_request(p, ofconn, oh);
3893 error = handle_vendor(p, ofconn, ofp_msg->data);
3896 case OFPT_BARRIER_REQUEST:
3897 error = handle_barrier_request(ofconn, oh);
3901 if (VLOG_IS_WARN_ENABLED()) {
3902 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3903 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3906 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3911 send_error_oh(ofconn, ofp_msg->data, error);
3916 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3918 struct odp_msg *msg = packet->data;
3920 struct ofpbuf payload;
3923 payload.data = msg + 1;
3924 payload.size = msg->length - sizeof *msg;
3925 flow_extract(&payload, msg->arg, msg->port, &flow);
3927 /* Check with in-band control to see if this packet should be sent
3928 * to the local port regardless of the flow table. */
3929 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3930 union odp_action action;
3932 memset(&action, 0, sizeof(action));
3933 action.output.type = ODPAT_OUTPUT;
3934 action.output.port = ODPP_LOCAL;
3935 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3938 rule = lookup_valid_rule(p, &flow);
3940 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3941 struct ofport *port = port_array_get(&p->ports, msg->port);
3943 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3944 COVERAGE_INC(ofproto_no_packet_in);
3945 /* XXX install 'drop' flow entry */
3946 ofpbuf_delete(packet);
3950 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3953 COVERAGE_INC(ofproto_packet_in);
3954 send_packet_in(p, packet);
3958 if (rule->cr.wc.wildcards) {
3959 rule = rule_create_subrule(p, rule, &flow);
3960 rule_make_actions(p, rule, packet);
3962 if (!rule->may_install) {
3963 /* The rule is not installable, that is, we need to process every
3964 * packet, so process the current packet and set its actions into
3966 rule_make_actions(p, rule, packet);
3968 /* XXX revalidate rule if it needs it */
3972 rule_execute(p, rule, &payload, &flow);
3973 rule_reinstall(p, rule);
3975 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3977 * Extra-special case for fail-open mode.
3979 * We are in fail-open mode and the packet matched the fail-open rule,
3980 * but we are connected to a controller too. We should send the packet
3981 * up to the controller in the hope that it will try to set up a flow
3982 * and thereby allow us to exit fail-open.
3984 * See the top-level comment in fail-open.c for more information.
3986 send_packet_in(p, packet);
3988 ofpbuf_delete(packet);
3993 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3995 struct odp_msg *msg = packet->data;
3997 switch (msg->type) {
3998 case _ODPL_ACTION_NR:
3999 COVERAGE_INC(ofproto_ctlr_action);
4000 send_packet_in(p, packet);
4003 case _ODPL_SFLOW_NR:
4005 ofproto_sflow_received(p->sflow, msg);
4007 ofpbuf_delete(packet);
4011 handle_odp_miss_msg(p, packet);
4015 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4022 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4024 struct rule *sub = rule_from_cls_rule(sub_);
4025 struct revalidate_cbdata *cbdata = cbdata_;
4027 if (cbdata->revalidate_all
4028 || (cbdata->revalidate_subrules && sub->super)
4029 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4030 revalidate_rule(cbdata->ofproto, sub);
4035 revalidate_rule(struct ofproto *p, struct rule *rule)
4037 const flow_t *flow = &rule->cr.flow;
4039 COVERAGE_INC(ofproto_revalidate_rule);
4042 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4044 rule_remove(p, rule);
4046 } else if (super != rule->super) {
4047 COVERAGE_INC(ofproto_revalidate_moved);
4048 list_remove(&rule->list);
4049 list_push_back(&super->list, &rule->list);
4050 rule->super = super;
4051 rule->hard_timeout = super->hard_timeout;
4052 rule->idle_timeout = super->idle_timeout;
4053 rule->created = super->created;
4058 rule_update_actions(p, rule);
4062 static struct ofpbuf *
4063 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4064 long long int now, uint8_t reason)
4066 struct ofp_flow_removed *ofr;
4068 long long int tdiff = now - rule->created;
4069 uint32_t sec = tdiff / 1000;
4070 uint32_t msec = tdiff - (sec * 1000);
4072 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4073 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4075 ofr->cookie = rule->flow_cookie;
4076 ofr->priority = htons(rule->cr.priority);
4077 ofr->reason = reason;
4078 ofr->duration_sec = htonl(sec);
4079 ofr->duration_nsec = htonl(msec * 1000000);
4080 ofr->idle_timeout = htons(rule->idle_timeout);
4081 ofr->packet_count = htonll(rule->packet_count);
4082 ofr->byte_count = htonll(rule->byte_count);
4088 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4090 assert(rule->installed);
4091 assert(!rule->cr.wc.wildcards);
4094 rule_remove(ofproto, rule);
4096 rule_uninstall(ofproto, rule);
4101 send_flow_removed(struct ofproto *p, struct rule *rule,
4102 long long int now, uint8_t reason)
4104 struct ofconn *ofconn;
4105 struct ofconn *prev;
4106 struct ofpbuf *buf = NULL;
4108 /* We limit the maximum number of queued flow expirations it by accounting
4109 * them under the counter for replies. That works because preventing
4110 * OpenFlow requests from being processed also prevents new flows from
4111 * being added (and expiring). (It also prevents processing OpenFlow
4112 * requests that would not add new flows, so it is imperfect.) */
4115 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4116 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4117 && ofconn_receives_async_msgs(ofconn)) {
4119 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4121 buf = compose_flow_removed(p, rule, now, reason);
4127 queue_tx(buf, prev, prev->reply_counter);
4133 expire_rule(struct cls_rule *cls_rule, void *p_)
4135 struct ofproto *p = p_;
4136 struct rule *rule = rule_from_cls_rule(cls_rule);
4137 long long int hard_expire, idle_expire, expire, now;
4139 hard_expire = (rule->hard_timeout
4140 ? rule->created + rule->hard_timeout * 1000
4142 idle_expire = (rule->idle_timeout
4143 && (rule->super || list_is_empty(&rule->list))
4144 ? rule->used + rule->idle_timeout * 1000
4146 expire = MIN(hard_expire, idle_expire);
4150 if (rule->installed && now >= rule->used + 5000) {
4151 uninstall_idle_flow(p, rule);
4152 } else if (!rule->cr.wc.wildcards) {
4153 active_timeout(p, rule);
4159 COVERAGE_INC(ofproto_expired);
4161 /* Update stats. This code will be a no-op if the rule expired
4162 * due to an idle timeout. */
4163 if (rule->cr.wc.wildcards) {
4164 struct rule *subrule, *next;
4165 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4166 rule_remove(p, subrule);
4169 rule_uninstall(p, rule);
4172 if (!rule_is_hidden(rule)) {
4173 send_flow_removed(p, rule, now,
4175 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4177 rule_remove(p, rule);
4181 active_timeout(struct ofproto *ofproto, struct rule *rule)
4183 if (ofproto->netflow && !is_controller_rule(rule) &&
4184 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4185 struct ofexpired expired;
4186 struct odp_flow odp_flow;
4188 /* Get updated flow stats. */
4189 memset(&odp_flow, 0, sizeof odp_flow);
4190 if (rule->installed) {
4191 odp_flow.key = rule->cr.flow;
4192 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4193 dpif_flow_get(ofproto->dpif, &odp_flow);
4195 if (odp_flow.stats.n_packets) {
4196 update_time(ofproto, rule, &odp_flow.stats);
4197 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
4198 odp_flow.stats.tcp_flags);
4202 expired.flow = rule->cr.flow;
4203 expired.packet_count = rule->packet_count +
4204 odp_flow.stats.n_packets;
4205 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4206 expired.used = rule->used;
4208 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4210 /* Schedule us to send the accumulated records once we have
4211 * collected all of them. */
4212 poll_immediate_wake();
4217 update_used(struct ofproto *p)
4219 struct odp_flow *flows;
4224 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4229 for (i = 0; i < n_flows; i++) {
4230 struct odp_flow *f = &flows[i];
4233 rule = rule_from_cls_rule(
4234 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4235 if (!rule || !rule->installed) {
4236 COVERAGE_INC(ofproto_unexpected_rule);
4237 dpif_flow_del(p->dpif, f);
4241 update_time(p, rule, &f->stats);
4242 rule_account(p, rule, f->stats.n_bytes);
4247 /* pinsched callback for sending 'packet' on 'ofconn'. */
4249 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4251 struct ofconn *ofconn = ofconn_;
4253 rconn_send_with_limit(ofconn->rconn, packet,
4254 ofconn->packet_in_counter, 100);
4257 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4258 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4259 * packet scheduler for sending.
4261 * 'max_len' specifies the maximum number of bytes of the packet to send on
4262 * 'ofconn' (INT_MAX specifies no limit).
4264 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4265 * ownership is transferred to this function. */
4267 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4270 struct ofproto *ofproto = ofconn->ofproto;
4271 struct ofp_packet_in *opi = packet->data;
4272 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4273 int send_len, trim_size;
4277 if (opi->reason == OFPR_ACTION) {
4278 buffer_id = UINT32_MAX;
4279 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4280 buffer_id = pktbuf_get_null();
4281 } else if (!ofconn->pktbuf) {
4282 buffer_id = UINT32_MAX;
4284 struct ofpbuf payload;
4285 payload.data = opi->data;
4286 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4287 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4290 /* Figure out how much of the packet to send. */
4291 send_len = ntohs(opi->total_len);
4292 if (buffer_id != UINT32_MAX) {
4293 send_len = MIN(send_len, ofconn->miss_send_len);
4295 send_len = MIN(send_len, max_len);
4297 /* Adjust packet length and clone if necessary. */
4298 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4300 packet = ofpbuf_clone_data(packet->data, trim_size);
4303 packet->size = trim_size;
4306 /* Update packet headers. */
4307 opi->buffer_id = htonl(buffer_id);
4308 update_openflow_length(packet);
4310 /* Hand over to packet scheduler. It might immediately call into
4311 * do_send_packet_in() or it might buffer it for a while (until a later
4312 * call to pinsched_run()). */
4313 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4314 packet, do_send_packet_in, ofconn);
4317 /* Replace struct odp_msg header in 'packet' by equivalent struct
4318 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4319 * returned by dpif_recv()).
4321 * The conversion is not complete: the caller still needs to trim any unneeded
4322 * payload off the end of the buffer, set the length in the OpenFlow header,
4323 * and set buffer_id. Those require us to know the controller settings and so
4324 * must be done on a per-controller basis.
4326 * Returns the maximum number of bytes of the packet that should be sent to
4327 * the controller (INT_MAX if no limit). */
4329 do_convert_to_packet_in(struct ofpbuf *packet)
4331 struct odp_msg *msg = packet->data;
4332 struct ofp_packet_in *opi;
4338 /* Extract relevant header fields */
4339 if (msg->type == _ODPL_ACTION_NR) {
4340 reason = OFPR_ACTION;
4343 reason = OFPR_NO_MATCH;
4346 total_len = msg->length - sizeof *msg;
4347 in_port = odp_port_to_ofp_port(msg->port);
4349 /* Repurpose packet buffer by overwriting header. */
4350 ofpbuf_pull(packet, sizeof(struct odp_msg));
4351 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4352 opi->header.version = OFP_VERSION;
4353 opi->header.type = OFPT_PACKET_IN;
4354 opi->total_len = htons(total_len);
4355 opi->in_port = htons(in_port);
4356 opi->reason = reason;
4361 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4362 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4363 * as necessary according to their individual configurations.
4365 * 'packet' must have sufficient headroom to convert it into a struct
4366 * ofp_packet_in (e.g. as returned by dpif_recv()).
4368 * Takes ownership of 'packet'. */
4370 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4372 struct ofconn *ofconn, *prev;
4375 max_len = do_convert_to_packet_in(packet);
4378 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4379 if (ofconn_receives_async_msgs(ofconn)) {
4381 schedule_packet_in(prev, packet, max_len, true);
4387 schedule_packet_in(prev, packet, max_len, false);
4389 ofpbuf_delete(packet);
4394 pick_datapath_id(const struct ofproto *ofproto)
4396 const struct ofport *port;
4398 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4400 uint8_t ea[ETH_ADDR_LEN];
4403 error = netdev_get_etheraddr(port->netdev, ea);
4405 return eth_addr_to_uint64(ea);
4407 VLOG_WARN("could not get MAC address for %s (%s)",
4408 netdev_get_name(port->netdev), strerror(error));
4410 return ofproto->fallback_dpid;
4414 pick_fallback_dpid(void)
4416 uint8_t ea[ETH_ADDR_LEN];
4417 eth_addr_nicira_random(ea);
4418 return eth_addr_to_uint64(ea);
4422 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4423 struct odp_actions *actions, tag_type *tags,
4424 uint16_t *nf_output_iface, void *ofproto_)
4426 struct ofproto *ofproto = ofproto_;
4429 /* Drop frames for reserved multicast addresses. */
4430 if (eth_addr_is_reserved(flow->dl_dst)) {
4434 /* Learn source MAC (but don't try to learn from revalidation). */
4435 if (packet != NULL) {
4436 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4438 GRAT_ARP_LOCK_NONE);
4440 /* The log messages here could actually be useful in debugging,
4441 * so keep the rate limit relatively high. */
4442 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4443 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4444 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4445 ofproto_revalidate(ofproto, rev_tag);
4449 /* Determine output port. */
4450 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4453 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4454 } else if (out_port != flow->in_port) {
4455 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4456 *nf_output_iface = out_port;
4464 static const struct ofhooks default_ofhooks = {
4466 default_normal_ofhook_cb,