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.
23 #include <netinet/in.h>
26 #include "classifier.h"
28 #include "discovery.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
38 #include "ofproto-sflow.h"
40 #include "openflow/nicira-ext.h"
41 #include "openflow/openflow.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
52 #include "stream-ssl.h"
60 #define THIS_MODULE VLM_ofproto
63 #include "sflow_api.h"
67 TABLEID_CLASSIFIER = 1
71 struct netdev *netdev;
72 struct ofp_phy_port opp; /* In host byte order. */
75 static void ofport_free(struct ofport *);
76 static void hton_ofp_phy_port(struct ofp_phy_port *);
78 static int xlate_actions(const union ofp_action *in, size_t n_in,
79 const flow_t *flow, struct ofproto *ofproto,
80 const struct ofpbuf *packet,
81 struct odp_actions *out, tag_type *tags,
82 bool *may_set_up_flow, uint16_t *nf_output_iface);
87 uint64_t flow_cookie; /* Controller-issued identifier.
88 (Kept in network-byte order.) */
89 uint16_t idle_timeout; /* In seconds from time of last use. */
90 uint16_t hard_timeout; /* In seconds from time of creation. */
91 bool send_flow_removed; /* Send a flow removed message? */
92 long long int used; /* Last-used time (0 if never used). */
93 long long int created; /* Creation time. */
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
96 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
97 tag_type tags; /* Tags (set only by hooks). */
98 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
100 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
101 * exact-match rule (having cr.wc.wildcards of 0) generated from the
102 * wildcard rule 'super'. In this case, 'list' is an element of the
105 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
106 * a list of subrules. A super-rule with no wildcards (where
107 * cr.wc.wildcards is 0) will never have any subrules. */
113 * 'n_actions' is the number of elements in the 'actions' array. A single
114 * action may take up more more than one element's worth of space.
116 * A subrule has no actions (it uses the super-rule's actions). */
118 union ofp_action *actions;
122 * A super-rule with wildcard fields never has ODP actions (since the
123 * datapath only supports exact-match flows). */
124 bool installed; /* Installed in datapath? */
125 bool may_install; /* True ordinarily; false if actions must
126 * be reassessed for every packet. */
128 union odp_action *odp_actions;
132 rule_is_hidden(const struct rule *rule)
134 /* Subrules are merely an implementation detail, so hide them from the
136 if (rule->super != NULL) {
140 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
141 * (e.g. by in-band control) and are intentionally hidden from the
143 if (rule->cr.priority > UINT16_MAX) {
150 static struct rule *rule_create(struct ofproto *, struct rule *super,
151 const union ofp_action *, size_t n_actions,
152 uint16_t idle_timeout, uint16_t hard_timeout,
153 uint64_t flow_cookie, bool send_flow_removed);
154 static void rule_free(struct rule *);
155 static void rule_destroy(struct ofproto *, struct rule *);
156 static struct rule *rule_from_cls_rule(const struct cls_rule *);
157 static void rule_insert(struct ofproto *, struct rule *,
158 struct ofpbuf *packet, uint16_t in_port);
159 static void rule_remove(struct ofproto *, struct rule *);
160 static bool rule_make_actions(struct ofproto *, struct rule *,
161 const struct ofpbuf *packet);
162 static void rule_install(struct ofproto *, struct rule *,
163 struct rule *displaced_rule);
164 static void rule_uninstall(struct ofproto *, struct rule *);
165 static void rule_post_uninstall(struct ofproto *, struct rule *);
166 static void send_flow_removed(struct ofproto *p, struct rule *rule,
167 long long int now, uint8_t reason);
169 /* ofproto supports two kinds of OpenFlow connections:
171 * - "Controller connections": Connections to ordinary OpenFlow controllers.
172 * ofproto maintains persistent connections to these controllers and by
173 * default sends them asynchronous messages such as packet-ins.
175 * - "Transient connections", e.g. from ovs-ofctl. When these connections
176 * drop, it is the other side's responsibility to reconnect them if
177 * necessary. ofproto does not send them asynchronous messages by default.
180 OFCONN_CONTROLLER, /* An OpenFlow controller. */
181 OFCONN_TRANSIENT /* A transient connection. */
184 /* An OpenFlow connection. */
186 struct ofproto *ofproto; /* The ofproto that owns this connection. */
187 struct list node; /* In struct ofproto's "all_conns" list. */
188 struct rconn *rconn; /* OpenFlow connection. */
189 enum ofconn_type type; /* Type. */
191 /* OFPT_PACKET_IN related data. */
192 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
193 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
194 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
195 int miss_send_len; /* Bytes to send of buffered packets. */
197 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
198 * requests, and the maximum number before we stop reading OpenFlow
200 #define OFCONN_REPLY_MAX 100
201 struct rconn_packet_counter *reply_counter;
203 /* type == OFCONN_CONTROLLER only. */
204 enum nx_role role; /* Role. */
205 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
206 struct discovery *discovery; /* Controller discovery object, if enabled. */
207 struct status_category *ss; /* Switch status category. */
208 enum ofproto_band band; /* In-band or out-of-band? */
211 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
212 * "schedulers" array. Their values are 0 and 1, and their meanings and values
213 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
214 * case anything ever changes, check their values here. */
215 #define N_SCHEDULERS 2
216 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
217 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
218 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
219 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
221 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
223 static void ofconn_destroy(struct ofconn *);
224 static void ofconn_run(struct ofconn *, struct ofproto *);
225 static void ofconn_wait(struct ofconn *);
226 static bool ofconn_receives_async_msgs(const struct ofconn *);
228 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
229 struct rconn_packet_counter *counter);
231 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
232 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
236 uint64_t datapath_id; /* Datapath ID. */
237 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
238 char *mfr_desc; /* Manufacturer. */
239 char *hw_desc; /* Hardware. */
240 char *sw_desc; /* Software version. */
241 char *serial_desc; /* Serial number. */
242 char *dp_desc; /* Datapath description. */
246 struct netdev_monitor *netdev_monitor;
247 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
249 struct shash port_by_name;
253 struct switch_status *switch_status;
254 struct fail_open *fail_open;
255 struct netflow *netflow;
256 struct ofproto_sflow *sflow;
258 /* In-band control. */
259 struct in_band *in_band;
260 long long int next_in_band_update;
261 struct sockaddr_in *extra_in_band_remotes;
262 size_t n_extra_remotes;
265 struct classifier cls;
266 bool need_revalidate;
267 long long int next_expiration;
268 struct tag_set revalidate_set;
269 bool tun_id_from_cookie;
271 /* OpenFlow connections. */
272 struct hmap controllers; /* Controller "struct ofconn"s. */
273 struct list all_conns; /* Contains "struct ofconn"s. */
274 struct pvconn **listeners;
276 struct pvconn **snoops;
279 /* Hooks for ovs-vswitchd. */
280 const struct ofhooks *ofhooks;
283 /* Used by default ofhooks. */
284 struct mac_learning *ml;
287 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
289 static const struct ofhooks default_ofhooks;
291 static uint64_t pick_datapath_id(const struct ofproto *);
292 static uint64_t pick_fallback_dpid(void);
294 static void update_used(struct ofproto *);
295 static void update_stats(struct ofproto *, struct rule *,
296 const struct odp_flow_stats *);
297 static void expire_rule(struct cls_rule *, void *ofproto);
298 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
299 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
300 static void revalidate_cb(struct cls_rule *rule_, void *p_);
302 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
304 static void handle_openflow(struct ofconn *, struct ofproto *,
307 static void refresh_port_groups(struct ofproto *);
309 static void update_port(struct ofproto *, const char *devname);
310 static int init_ports(struct ofproto *);
311 static void reinit_ports(struct ofproto *);
314 ofproto_create(const char *datapath, const char *datapath_type,
315 const struct ofhooks *ofhooks, void *aux,
316 struct ofproto **ofprotop)
318 struct odp_stats stats;
325 /* Connect to datapath and start listening for messages. */
326 error = dpif_open(datapath, datapath_type, &dpif);
328 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
331 error = dpif_get_dp_stats(dpif, &stats);
333 VLOG_ERR("failed to obtain stats for datapath %s: %s",
334 datapath, strerror(error));
338 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
340 VLOG_ERR("failed to listen on datapath %s: %s",
341 datapath, strerror(error));
345 dpif_flow_flush(dpif);
346 dpif_recv_purge(dpif);
348 /* Initialize settings. */
349 p = xzalloc(sizeof *p);
350 p->fallback_dpid = pick_fallback_dpid();
351 p->datapath_id = p->fallback_dpid;
352 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
353 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
354 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
355 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
356 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
358 /* Initialize datapath. */
360 p->netdev_monitor = netdev_monitor_create();
361 port_array_init(&p->ports);
362 shash_init(&p->port_by_name);
363 p->max_ports = stats.max_ports;
365 /* Initialize submodules. */
366 p->switch_status = switch_status_create(p);
372 /* Initialize flow table. */
373 classifier_init(&p->cls);
374 p->need_revalidate = false;
375 p->next_expiration = time_msec() + 1000;
376 tag_set_init(&p->revalidate_set);
378 /* Initialize OpenFlow connections. */
379 list_init(&p->all_conns);
380 hmap_init(&p->controllers);
386 /* Initialize hooks. */
388 p->ofhooks = ofhooks;
392 p->ofhooks = &default_ofhooks;
394 p->ml = mac_learning_create();
397 /* Pick final datapath ID. */
398 p->datapath_id = pick_datapath_id(p);
399 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
406 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
408 uint64_t old_dpid = p->datapath_id;
409 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
410 if (p->datapath_id != old_dpid) {
411 struct ofconn *ofconn;
413 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
415 /* Force all active connections to reconnect, since there is no way to
416 * notify a controller that the datapath ID has changed. */
417 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
418 rconn_reconnect(ofconn->rconn);
424 is_discovery_controller(const struct ofproto_controller *c)
426 return !strcmp(c->target, "discover");
430 is_in_band_controller(const struct ofproto_controller *c)
432 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
435 /* Creates a new controller in 'ofproto'. Some of the settings are initially
436 * drawn from 'c', but update_controller() needs to be called later to finish
437 * the new ofconn's configuration. */
439 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
441 struct discovery *discovery;
442 struct ofconn *ofconn;
444 if (is_discovery_controller(c)) {
445 int error = discovery_create(c->accept_re, c->update_resolv_conf,
446 ofproto->dpif, ofproto->switch_status,
455 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
456 ofconn->pktbuf = pktbuf_create();
457 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
459 ofconn->discovery = discovery;
461 rconn_connect(ofconn->rconn, c->target);
463 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
464 hash_string(c->target, 0));
467 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
468 * target or turn discovery on or off (these are done by creating new ofconns
469 * and deleting old ones), but it can update the rest of an ofconn's
472 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
474 struct ofproto *ofproto = ofconn->ofproto;
478 ofconn->band = (is_in_band_controller(c)
479 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
481 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
483 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
484 rconn_set_probe_interval(ofconn->rconn, probe_interval);
486 if (ofconn->discovery) {
487 discovery_set_update_resolv_conf(ofconn->discovery,
488 c->update_resolv_conf);
489 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
492 for (i = 0; i < N_SCHEDULERS; i++) {
493 struct pinsched **s = &ofconn->schedulers[i];
495 if (c->rate_limit > 0) {
497 *s = pinsched_create(c->rate_limit, c->burst_limit,
498 ofproto->switch_status);
500 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
503 pinsched_destroy(*s);
510 ofconn_get_target(const struct ofconn *ofconn)
512 return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn);
515 static struct ofconn *
516 find_controller_by_target(struct ofproto *ofproto, const char *target)
518 struct ofconn *ofconn;
520 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
521 hash_string(target, 0), &ofproto->controllers) {
522 if (!strcmp(ofconn_get_target(ofconn), target)) {
530 update_in_band_remotes(struct ofproto *ofproto)
532 const struct ofconn *ofconn;
533 struct sockaddr_in *addrs;
534 size_t max_addrs, n_addrs;
538 /* Allocate enough memory for as many remotes as we could possibly have. */
539 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
540 addrs = xmalloc(max_addrs * sizeof *addrs);
543 /* Add all the remotes. */
545 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
546 struct sockaddr_in *sin = &addrs[n_addrs];
548 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
549 if (sin->sin_addr.s_addr) {
550 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
553 if (ofconn->discovery) {
557 for (i = 0; i < ofproto->n_extra_remotes; i++) {
558 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
561 /* Create or update or destroy in-band.
563 * Ordinarily we only enable in-band if there's at least one remote
564 * address, but discovery needs the in-band rules for DHCP to be installed
565 * even before we know any remote addresses. */
566 if (n_addrs || discovery) {
567 if (!ofproto->in_band) {
568 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
571 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
572 ofproto->next_in_band_update = time_msec() + 1000;
574 in_band_destroy(ofproto->in_band);
575 ofproto->in_band = NULL;
583 ofproto_set_controllers(struct ofproto *p,
584 const struct ofproto_controller *controllers,
585 size_t n_controllers)
587 struct shash new_controllers;
588 enum ofproto_fail_mode fail_mode;
589 struct ofconn *ofconn, *next;
593 shash_init(&new_controllers);
594 for (i = 0; i < n_controllers; i++) {
595 const struct ofproto_controller *c = &controllers[i];
597 shash_add_once(&new_controllers, c->target, &controllers[i]);
598 if (!find_controller_by_target(p, c->target)) {
599 add_controller(p, c);
603 fail_mode = OFPROTO_FAIL_STANDALONE;
605 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
607 struct ofproto_controller *c;
609 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
611 ofconn_destroy(ofconn);
613 update_controller(ofconn, c);
617 if (c->fail == OFPROTO_FAIL_SECURE) {
618 fail_mode = OFPROTO_FAIL_SECURE;
622 shash_destroy(&new_controllers);
624 update_in_band_remotes(p);
626 if (!hmap_is_empty(&p->controllers)
627 && fail_mode == OFPROTO_FAIL_STANDALONE) {
628 struct rconn **rconns;
632 p->fail_open = fail_open_create(p, p->switch_status);
636 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
637 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
638 rconns[n++] = ofconn->rconn;
641 fail_open_set_controllers(p->fail_open, rconns, n);
642 /* p->fail_open takes ownership of 'rconns'. */
644 fail_open_destroy(p->fail_open);
648 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
649 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
650 struct ofconn, hmap_node);
651 ofconn->ss = switch_status_register(p->switch_status, "remote",
652 rconn_status_cb, ofconn->rconn);
657 any_extras_changed(const struct ofproto *ofproto,
658 const struct sockaddr_in *extras, size_t n)
662 if (n != ofproto->n_extra_remotes) {
666 for (i = 0; i < n; i++) {
667 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
668 const struct sockaddr_in *new = &extras[i];
670 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
671 old->sin_port != new->sin_port) {
679 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
680 * in-band control should guarantee access, in the same way that in-band
681 * control guarantees access to OpenFlow controllers. */
683 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
684 const struct sockaddr_in *extras, size_t n)
686 if (!any_extras_changed(ofproto, extras, n)) {
690 free(ofproto->extra_in_band_remotes);
691 ofproto->n_extra_remotes = n;
692 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
694 update_in_band_remotes(ofproto);
698 ofproto_set_desc(struct ofproto *p,
699 const char *mfr_desc, const char *hw_desc,
700 const char *sw_desc, const char *serial_desc,
703 struct ofp_desc_stats *ods;
706 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
707 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
708 sizeof ods->mfr_desc);
711 p->mfr_desc = xstrdup(mfr_desc);
714 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
715 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
716 sizeof ods->hw_desc);
719 p->hw_desc = xstrdup(hw_desc);
722 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
723 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
724 sizeof ods->sw_desc);
727 p->sw_desc = xstrdup(sw_desc);
730 if (strlen(serial_desc) >= sizeof ods->serial_num) {
731 VLOG_WARN("truncating serial_desc, must be less than %zu "
733 sizeof ods->serial_num);
735 free(p->serial_desc);
736 p->serial_desc = xstrdup(serial_desc);
739 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
740 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
741 sizeof ods->dp_desc);
744 p->dp_desc = xstrdup(dp_desc);
749 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
750 const struct svec *svec)
752 struct pvconn **pvconns = *pvconnsp;
753 size_t n_pvconns = *n_pvconnsp;
757 for (i = 0; i < n_pvconns; i++) {
758 pvconn_close(pvconns[i]);
762 pvconns = xmalloc(svec->n * sizeof *pvconns);
764 for (i = 0; i < svec->n; i++) {
765 const char *name = svec->names[i];
766 struct pvconn *pvconn;
769 error = pvconn_open(name, &pvconn);
771 pvconns[n_pvconns++] = pvconn;
773 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
781 *n_pvconnsp = n_pvconns;
787 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
789 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
793 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
795 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
799 ofproto_set_netflow(struct ofproto *ofproto,
800 const struct netflow_options *nf_options)
802 if (nf_options && nf_options->collectors.n) {
803 if (!ofproto->netflow) {
804 ofproto->netflow = netflow_create();
806 return netflow_set_options(ofproto->netflow, nf_options);
808 netflow_destroy(ofproto->netflow);
809 ofproto->netflow = NULL;
815 ofproto_set_sflow(struct ofproto *ofproto,
816 const struct ofproto_sflow_options *oso)
818 struct ofproto_sflow *os = ofproto->sflow;
821 struct ofport *ofport;
822 unsigned int odp_port;
824 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
825 refresh_port_groups(ofproto);
826 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
827 ofproto_sflow_add_port(os, odp_port,
828 netdev_get_name(ofport->netdev));
831 ofproto_sflow_set_options(os, oso);
833 ofproto_sflow_destroy(os);
834 ofproto->sflow = NULL;
839 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
843 VLOG_WARN("STP is not yet implemented");
851 ofproto_get_datapath_id(const struct ofproto *ofproto)
853 return ofproto->datapath_id;
857 ofproto_has_controller(const struct ofproto *ofproto)
859 return !hmap_is_empty(&ofproto->controllers);
863 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
867 for (i = 0; i < ofproto->n_listeners; i++) {
868 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
873 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
877 for (i = 0; i < ofproto->n_snoops; i++) {
878 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
883 ofproto_destroy(struct ofproto *p)
885 struct ofconn *ofconn, *next_ofconn;
886 struct ofport *ofport;
887 unsigned int port_no;
894 /* Destroy fail-open and in-band early, since they touch the classifier. */
895 fail_open_destroy(p->fail_open);
898 in_band_destroy(p->in_band);
900 free(p->extra_in_band_remotes);
902 ofproto_flush_flows(p);
903 classifier_destroy(&p->cls);
905 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
907 ofconn_destroy(ofconn);
909 hmap_destroy(&p->controllers);
912 netdev_monitor_destroy(p->netdev_monitor);
913 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
916 shash_destroy(&p->port_by_name);
918 switch_status_destroy(p->switch_status);
919 netflow_destroy(p->netflow);
920 ofproto_sflow_destroy(p->sflow);
922 for (i = 0; i < p->n_listeners; i++) {
923 pvconn_close(p->listeners[i]);
927 for (i = 0; i < p->n_snoops; i++) {
928 pvconn_close(p->snoops[i]);
932 mac_learning_destroy(p->ml);
937 free(p->serial_desc);
940 port_array_destroy(&p->ports);
946 ofproto_run(struct ofproto *p)
948 int error = ofproto_run1(p);
950 error = ofproto_run2(p, false);
956 process_port_change(struct ofproto *ofproto, int error, char *devname)
958 if (error == ENOBUFS) {
959 reinit_ports(ofproto);
961 update_port(ofproto, devname);
966 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
967 * means that 'ofconn' is more interesting for monitoring than a lower return
970 snoop_preference(const struct ofconn *ofconn)
972 switch (ofconn->role) {
980 /* Shouldn't happen. */
985 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
986 * Connects this vconn to a controller. */
988 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
990 struct ofconn *ofconn, *best;
992 /* Pick a controller for monitoring. */
994 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
995 if (ofconn->type == OFCONN_CONTROLLER
996 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1002 rconn_add_monitor(best->rconn, vconn);
1004 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1010 ofproto_run1(struct ofproto *p)
1012 struct ofconn *ofconn, *next_ofconn;
1017 if (shash_is_empty(&p->port_by_name)) {
1021 for (i = 0; i < 50; i++) {
1025 error = dpif_recv(p->dpif, &buf);
1027 if (error == ENODEV) {
1028 /* Someone destroyed the datapath behind our back. The caller
1029 * better destroy us and give up, because we're just going to
1030 * spin from here on out. */
1031 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1032 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1033 dpif_name(p->dpif));
1039 handle_odp_msg(p, buf);
1042 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1043 process_port_change(p, error, devname);
1045 while ((error = netdev_monitor_poll(p->netdev_monitor,
1046 &devname)) != EAGAIN) {
1047 process_port_change(p, error, devname);
1051 if (time_msec() >= p->next_in_band_update) {
1052 update_in_band_remotes(p);
1054 in_band_run(p->in_band);
1057 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1059 ofconn_run(ofconn, p);
1062 /* Fail-open maintenance. Do this after processing the ofconns since
1063 * fail-open checks the status of the controller rconn. */
1065 fail_open_run(p->fail_open);
1068 for (i = 0; i < p->n_listeners; i++) {
1069 struct vconn *vconn;
1072 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1074 ofconn_create(p, rconn_new_from_vconn("passive", vconn),
1076 } else if (retval != EAGAIN) {
1077 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1081 for (i = 0; i < p->n_snoops; i++) {
1082 struct vconn *vconn;
1085 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1087 add_snooper(p, vconn);
1088 } else if (retval != EAGAIN) {
1089 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1093 if (time_msec() >= p->next_expiration) {
1094 COVERAGE_INC(ofproto_expiration);
1095 p->next_expiration = time_msec() + 1000;
1098 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1100 /* Let the hook know that we're at a stable point: all outstanding data
1101 * in existing flows has been accounted to the account_cb. Thus, the
1102 * hook can now reasonably do operations that depend on having accurate
1103 * flow volume accounting (currently, that's just bond rebalancing). */
1104 if (p->ofhooks->account_checkpoint_cb) {
1105 p->ofhooks->account_checkpoint_cb(p->aux);
1110 netflow_run(p->netflow);
1113 ofproto_sflow_run(p->sflow);
1119 struct revalidate_cbdata {
1120 struct ofproto *ofproto;
1121 bool revalidate_all; /* Revalidate all exact-match rules? */
1122 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1123 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1127 ofproto_run2(struct ofproto *p, bool revalidate_all)
1129 if (p->need_revalidate || revalidate_all
1130 || !tag_set_is_empty(&p->revalidate_set)) {
1131 struct revalidate_cbdata cbdata;
1133 cbdata.revalidate_all = revalidate_all;
1134 cbdata.revalidate_subrules = p->need_revalidate;
1135 cbdata.revalidate_set = p->revalidate_set;
1136 tag_set_init(&p->revalidate_set);
1137 COVERAGE_INC(ofproto_revalidate);
1138 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1139 p->need_revalidate = false;
1146 ofproto_wait(struct ofproto *p)
1148 struct ofconn *ofconn;
1151 dpif_recv_wait(p->dpif);
1152 dpif_port_poll_wait(p->dpif);
1153 netdev_monitor_poll_wait(p->netdev_monitor);
1154 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1155 ofconn_wait(ofconn);
1158 poll_timer_wait(p->next_in_band_update - time_msec());
1159 in_band_wait(p->in_band);
1162 fail_open_wait(p->fail_open);
1165 ofproto_sflow_wait(p->sflow);
1167 if (!tag_set_is_empty(&p->revalidate_set)) {
1168 poll_immediate_wake();
1170 if (p->need_revalidate) {
1171 /* Shouldn't happen, but if it does just go around again. */
1172 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1173 poll_immediate_wake();
1174 } else if (p->next_expiration != LLONG_MAX) {
1175 poll_timer_wait(p->next_expiration - time_msec());
1177 for (i = 0; i < p->n_listeners; i++) {
1178 pvconn_wait(p->listeners[i]);
1180 for (i = 0; i < p->n_snoops; i++) {
1181 pvconn_wait(p->snoops[i]);
1186 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1188 tag_set_add(&ofproto->revalidate_set, tag);
1192 ofproto_get_revalidate_set(struct ofproto *ofproto)
1194 return &ofproto->revalidate_set;
1198 ofproto_is_alive(const struct ofproto *p)
1200 return !hmap_is_empty(&p->controllers);
1204 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1205 const union ofp_action *actions, size_t n_actions,
1206 const struct ofpbuf *packet)
1208 struct odp_actions odp_actions;
1211 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1217 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1219 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1220 odp_actions.n_actions, packet);
1225 ofproto_add_flow(struct ofproto *p,
1226 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1227 const union ofp_action *actions, size_t n_actions,
1231 rule = rule_create(p, NULL, actions, n_actions,
1232 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1234 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1235 rule_insert(p, rule, NULL, 0);
1239 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1240 uint32_t wildcards, unsigned int priority)
1244 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1248 rule_remove(ofproto, rule);
1253 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1255 struct rule *rule = rule_from_cls_rule(rule_);
1256 struct ofproto *ofproto = ofproto_;
1258 /* Mark the flow as not installed, even though it might really be
1259 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1260 * There is no point in uninstalling it individually since we are about to
1261 * blow away all the flows with dpif_flow_flush(). */
1262 rule->installed = false;
1264 rule_remove(ofproto, rule);
1268 ofproto_flush_flows(struct ofproto *ofproto)
1270 COVERAGE_INC(ofproto_flush);
1271 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1272 dpif_flow_flush(ofproto->dpif);
1273 if (ofproto->in_band) {
1274 in_band_flushed(ofproto->in_band);
1276 if (ofproto->fail_open) {
1277 fail_open_flushed(ofproto->fail_open);
1282 reinit_ports(struct ofproto *p)
1284 struct svec devnames;
1285 struct ofport *ofport;
1286 unsigned int port_no;
1287 struct odp_port *odp_ports;
1291 svec_init(&devnames);
1292 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1293 svec_add (&devnames, (char *) ofport->opp.name);
1295 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1296 for (i = 0; i < n_odp_ports; i++) {
1297 svec_add (&devnames, odp_ports[i].devname);
1301 svec_sort_unique(&devnames);
1302 for (i = 0; i < devnames.n; i++) {
1303 update_port(p, devnames.names[i]);
1305 svec_destroy(&devnames);
1309 refresh_port_group(struct ofproto *p, unsigned int group)
1313 struct ofport *port;
1314 unsigned int port_no;
1316 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1318 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1320 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1321 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1322 ports[n_ports++] = port_no;
1325 dpif_port_group_set(p->dpif, group, ports, n_ports);
1332 refresh_port_groups(struct ofproto *p)
1334 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1335 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1337 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1341 static struct ofport *
1342 make_ofport(const struct odp_port *odp_port)
1344 struct netdev_options netdev_options;
1345 enum netdev_flags flags;
1346 struct ofport *ofport;
1347 struct netdev *netdev;
1351 memset(&netdev_options, 0, sizeof netdev_options);
1352 netdev_options.name = odp_port->devname;
1353 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1354 netdev_options.may_open = true;
1356 error = netdev_open(&netdev_options, &netdev);
1358 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1359 "cannot be opened (%s)",
1360 odp_port->devname, odp_port->port,
1361 odp_port->devname, strerror(error));
1365 ofport = xmalloc(sizeof *ofport);
1366 ofport->netdev = netdev;
1367 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1368 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1369 memcpy(ofport->opp.name, odp_port->devname,
1370 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1371 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1373 netdev_get_flags(netdev, &flags);
1374 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1376 netdev_get_carrier(netdev, &carrier);
1377 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1379 netdev_get_features(netdev,
1380 &ofport->opp.curr, &ofport->opp.advertised,
1381 &ofport->opp.supported, &ofport->opp.peer);
1386 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1388 if (port_array_get(&p->ports, odp_port->port)) {
1389 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1392 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1393 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1402 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1404 const struct ofp_phy_port *a = &a_->opp;
1405 const struct ofp_phy_port *b = &b_->opp;
1407 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1408 return (a->port_no == b->port_no
1409 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1410 && !strcmp((char *) a->name, (char *) b->name)
1411 && a->state == b->state
1412 && a->config == b->config
1413 && a->curr == b->curr
1414 && a->advertised == b->advertised
1415 && a->supported == b->supported
1416 && a->peer == b->peer);
1420 send_port_status(struct ofproto *p, const struct ofport *ofport,
1423 /* XXX Should limit the number of queued port status change messages. */
1424 struct ofconn *ofconn;
1425 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1426 struct ofp_port_status *ops;
1429 if (!ofconn_receives_async_msgs(ofconn)) {
1433 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1434 ops->reason = reason;
1435 ops->desc = ofport->opp;
1436 hton_ofp_phy_port(&ops->desc);
1437 queue_tx(b, ofconn, NULL);
1439 if (p->ofhooks->port_changed_cb) {
1440 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1445 ofport_install(struct ofproto *p, struct ofport *ofport)
1447 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1448 const char *netdev_name = (const char *) ofport->opp.name;
1450 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1451 port_array_set(&p->ports, odp_port, ofport);
1452 shash_add(&p->port_by_name, netdev_name, ofport);
1454 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1459 ofport_remove(struct ofproto *p, struct ofport *ofport)
1461 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1463 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1464 port_array_set(&p->ports, odp_port, NULL);
1465 shash_delete(&p->port_by_name,
1466 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1468 ofproto_sflow_del_port(p->sflow, odp_port);
1473 ofport_free(struct ofport *ofport)
1476 netdev_close(ofport->netdev);
1482 update_port(struct ofproto *p, const char *devname)
1484 struct odp_port odp_port;
1485 struct ofport *old_ofport;
1486 struct ofport *new_ofport;
1489 COVERAGE_INC(ofproto_update_port);
1491 /* Query the datapath for port information. */
1492 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1494 /* Find the old ofport. */
1495 old_ofport = shash_find_data(&p->port_by_name, devname);
1498 /* There's no port named 'devname' but there might be a port with
1499 * the same port number. This could happen if a port is deleted
1500 * and then a new one added in its place very quickly, or if a port
1501 * is renamed. In the former case we want to send an OFPPR_DELETE
1502 * and an OFPPR_ADD, and in the latter case we want to send a
1503 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1504 * the old port's ifindex against the new port, or perhaps less
1505 * reliably but more portably by comparing the old port's MAC
1506 * against the new port's MAC. However, this code isn't that smart
1507 * and always sends an OFPPR_MODIFY (XXX). */
1508 old_ofport = port_array_get(&p->ports, odp_port.port);
1510 } else if (error != ENOENT && error != ENODEV) {
1511 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1512 "%s", strerror(error));
1516 /* Create a new ofport. */
1517 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1519 /* Eliminate a few pathological cases. */
1520 if (!old_ofport && !new_ofport) {
1522 } else if (old_ofport && new_ofport) {
1523 /* Most of the 'config' bits are OpenFlow soft state, but
1524 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1525 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1526 * leaves the other bits 0.) */
1527 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1529 if (ofport_equal(old_ofport, new_ofport)) {
1530 /* False alarm--no change. */
1531 ofport_free(new_ofport);
1536 /* Now deal with the normal cases. */
1538 ofport_remove(p, old_ofport);
1541 ofport_install(p, new_ofport);
1543 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1544 (!old_ofport ? OFPPR_ADD
1545 : !new_ofport ? OFPPR_DELETE
1547 ofport_free(old_ofport);
1549 /* Update port groups. */
1550 refresh_port_groups(p);
1554 init_ports(struct ofproto *p)
1556 struct odp_port *ports;
1561 error = dpif_port_list(p->dpif, &ports, &n_ports);
1566 for (i = 0; i < n_ports; i++) {
1567 const struct odp_port *odp_port = &ports[i];
1568 if (!ofport_conflicts(p, odp_port)) {
1569 struct ofport *ofport = make_ofport(odp_port);
1571 ofport_install(p, ofport);
1576 refresh_port_groups(p);
1580 static struct ofconn *
1581 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1583 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1584 ofconn->ofproto = p;
1585 list_push_back(&p->all_conns, &ofconn->node);
1586 ofconn->rconn = rconn;
1587 ofconn->type = type;
1588 ofconn->role = NX_ROLE_OTHER;
1589 ofconn->packet_in_counter = rconn_packet_counter_create ();
1590 ofconn->pktbuf = NULL;
1591 ofconn->miss_send_len = 0;
1592 ofconn->reply_counter = rconn_packet_counter_create ();
1597 ofconn_destroy(struct ofconn *ofconn)
1599 if (ofconn->type == OFCONN_CONTROLLER) {
1600 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1602 discovery_destroy(ofconn->discovery);
1604 list_remove(&ofconn->node);
1605 switch_status_unregister(ofconn->ss);
1606 rconn_destroy(ofconn->rconn);
1607 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1608 rconn_packet_counter_destroy(ofconn->reply_counter);
1609 pktbuf_destroy(ofconn->pktbuf);
1614 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1619 if (ofconn->discovery) {
1620 char *controller_name;
1621 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1622 discovery_question_connectivity(ofconn->discovery);
1624 if (discovery_run(ofconn->discovery, &controller_name)) {
1625 if (controller_name) {
1626 rconn_connect(ofconn->rconn, controller_name);
1628 rconn_disconnect(ofconn->rconn);
1633 for (i = 0; i < N_SCHEDULERS; i++) {
1634 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1637 rconn_run(ofconn->rconn);
1639 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1640 /* Limit the number of iterations to prevent other tasks from
1642 for (iteration = 0; iteration < 50; iteration++) {
1643 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1648 fail_open_maybe_recover(p->fail_open);
1650 handle_openflow(ofconn, p, of_msg);
1651 ofpbuf_delete(of_msg);
1655 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1656 ofconn_destroy(ofconn);
1661 ofconn_wait(struct ofconn *ofconn)
1665 if (ofconn->discovery) {
1666 discovery_wait(ofconn->discovery);
1668 for (i = 0; i < N_SCHEDULERS; i++) {
1669 pinsched_wait(ofconn->schedulers[i]);
1671 rconn_run_wait(ofconn->rconn);
1672 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1673 rconn_recv_wait(ofconn->rconn);
1675 COVERAGE_INC(ofproto_ofconn_stuck);
1679 /* Returns true if 'ofconn' should receive asynchronous messages. */
1681 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1683 if (ofconn->type == OFCONN_CONTROLLER) {
1684 /* Ordinary controllers always get asynchronous messages unless they
1685 * have configured themselves as "slaves". */
1686 return ofconn->role != NX_ROLE_SLAVE;
1688 /* Transient connections don't get asynchronous messages unless they
1689 * have explicitly asked for them by setting a nonzero miss send
1691 return ofconn->miss_send_len > 0;
1695 /* Caller is responsible for initializing the 'cr' member of the returned
1697 static struct rule *
1698 rule_create(struct ofproto *ofproto, struct rule *super,
1699 const union ofp_action *actions, size_t n_actions,
1700 uint16_t idle_timeout, uint16_t hard_timeout,
1701 uint64_t flow_cookie, bool send_flow_removed)
1703 struct rule *rule = xzalloc(sizeof *rule);
1704 rule->idle_timeout = idle_timeout;
1705 rule->hard_timeout = hard_timeout;
1706 rule->flow_cookie = flow_cookie;
1707 rule->used = rule->created = time_msec();
1708 rule->send_flow_removed = send_flow_removed;
1709 rule->super = super;
1711 list_push_back(&super->list, &rule->list);
1713 list_init(&rule->list);
1715 rule->n_actions = n_actions;
1716 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1717 netflow_flow_clear(&rule->nf_flow);
1718 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1723 static struct rule *
1724 rule_from_cls_rule(const struct cls_rule *cls_rule)
1726 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1730 rule_free(struct rule *rule)
1732 free(rule->actions);
1733 free(rule->odp_actions);
1737 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1738 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1739 * through all of its subrules and revalidates them, destroying any that no
1740 * longer has a super-rule (which is probably all of them).
1742 * Before calling this function, the caller must make have removed 'rule' from
1743 * the classifier. If 'rule' is an exact-match rule, the caller is also
1744 * responsible for ensuring that it has been uninstalled from the datapath. */
1746 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1749 struct rule *subrule, *next;
1750 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1751 revalidate_rule(ofproto, subrule);
1754 list_remove(&rule->list);
1760 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1762 const union ofp_action *oa;
1763 struct actions_iterator i;
1765 if (out_port == htons(OFPP_NONE)) {
1768 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1769 oa = actions_next(&i)) {
1770 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1777 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1778 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1780 * The flow that 'packet' actually contains does not need to actually match
1781 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1782 * the packet and byte counters for 'rule' will be credited for the packet sent
1783 * out whether or not the packet actually matches 'rule'.
1785 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1786 * the caller must already have accurately composed ODP actions for it given
1787 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1788 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1789 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1790 * actions and apply them to 'packet'. */
1792 rule_execute(struct ofproto *ofproto, struct rule *rule,
1793 struct ofpbuf *packet, const flow_t *flow)
1795 const union odp_action *actions;
1797 struct odp_actions a;
1799 /* Grab or compose the ODP actions.
1801 * The special case for an exact-match 'rule' where 'flow' is not the
1802 * rule's flow is important to avoid, e.g., sending a packet out its input
1803 * port simply because the ODP actions were composed for the wrong
1805 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1806 struct rule *super = rule->super ? rule->super : rule;
1807 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1808 packet, &a, NULL, 0, NULL)) {
1811 actions = a.actions;
1812 n_actions = a.n_actions;
1814 actions = rule->odp_actions;
1815 n_actions = rule->n_odp_actions;
1818 /* Execute the ODP actions. */
1819 if (!dpif_execute(ofproto->dpif, flow->in_port,
1820 actions, n_actions, packet)) {
1821 struct odp_flow_stats stats;
1822 flow_extract_stats(flow, packet, &stats);
1823 update_stats(ofproto, rule, &stats);
1824 rule->used = time_msec();
1825 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1830 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1833 struct rule *displaced_rule;
1835 /* Insert the rule in the classifier. */
1836 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1837 if (!rule->cr.wc.wildcards) {
1838 rule_make_actions(p, rule, packet);
1841 /* Send the packet and credit it to the rule. */
1844 flow_extract(packet, 0, in_port, &flow);
1845 rule_execute(p, rule, packet, &flow);
1848 /* Install the rule in the datapath only after sending the packet, to
1849 * avoid packet reordering. */
1850 if (rule->cr.wc.wildcards) {
1851 COVERAGE_INC(ofproto_add_wc_flow);
1852 p->need_revalidate = true;
1854 rule_install(p, rule, displaced_rule);
1857 /* Free the rule that was displaced, if any. */
1858 if (displaced_rule) {
1859 rule_destroy(p, displaced_rule);
1863 static struct rule *
1864 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1867 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1868 rule->idle_timeout, rule->hard_timeout,
1870 COVERAGE_INC(ofproto_subrule_create);
1871 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1872 : rule->cr.priority), &subrule->cr);
1873 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1879 rule_remove(struct ofproto *ofproto, struct rule *rule)
1881 if (rule->cr.wc.wildcards) {
1882 COVERAGE_INC(ofproto_del_wc_flow);
1883 ofproto->need_revalidate = true;
1885 rule_uninstall(ofproto, rule);
1887 classifier_remove(&ofproto->cls, &rule->cr);
1888 rule_destroy(ofproto, rule);
1891 /* Returns true if the actions changed, false otherwise. */
1893 rule_make_actions(struct ofproto *p, struct rule *rule,
1894 const struct ofpbuf *packet)
1896 const struct rule *super;
1897 struct odp_actions a;
1900 assert(!rule->cr.wc.wildcards);
1902 super = rule->super ? rule->super : rule;
1904 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1905 packet, &a, &rule->tags, &rule->may_install,
1906 &rule->nf_flow.output_iface);
1908 actions_len = a.n_actions * sizeof *a.actions;
1909 if (rule->n_odp_actions != a.n_actions
1910 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1911 COVERAGE_INC(ofproto_odp_unchanged);
1912 free(rule->odp_actions);
1913 rule->n_odp_actions = a.n_actions;
1914 rule->odp_actions = xmemdup(a.actions, actions_len);
1922 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1923 struct odp_flow_put *put)
1925 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1926 put->flow.key = rule->cr.flow;
1927 put->flow.actions = rule->odp_actions;
1928 put->flow.n_actions = rule->n_odp_actions;
1929 put->flow.flags = 0;
1931 return dpif_flow_put(ofproto->dpif, put);
1935 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1937 assert(!rule->cr.wc.wildcards);
1939 if (rule->may_install) {
1940 struct odp_flow_put put;
1941 if (!do_put_flow(p, rule,
1942 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1944 rule->installed = true;
1945 if (displaced_rule) {
1946 update_stats(p, displaced_rule, &put.flow.stats);
1947 rule_post_uninstall(p, displaced_rule);
1950 } else if (displaced_rule) {
1951 rule_uninstall(p, displaced_rule);
1956 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1958 if (rule->installed) {
1959 struct odp_flow_put put;
1960 COVERAGE_INC(ofproto_dp_missed);
1961 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1963 rule_install(ofproto, rule, NULL);
1968 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1970 bool actions_changed;
1971 uint16_t new_out_iface, old_out_iface;
1973 old_out_iface = rule->nf_flow.output_iface;
1974 actions_changed = rule_make_actions(ofproto, rule, NULL);
1976 if (rule->may_install) {
1977 if (rule->installed) {
1978 if (actions_changed) {
1979 struct odp_flow_put put;
1980 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1981 | ODPPF_ZERO_STATS, &put);
1982 update_stats(ofproto, rule, &put.flow.stats);
1984 /* Temporarily set the old output iface so that NetFlow
1985 * messages have the correct output interface for the old
1987 new_out_iface = rule->nf_flow.output_iface;
1988 rule->nf_flow.output_iface = old_out_iface;
1989 rule_post_uninstall(ofproto, rule);
1990 rule->nf_flow.output_iface = new_out_iface;
1993 rule_install(ofproto, rule, NULL);
1996 rule_uninstall(ofproto, rule);
2001 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2003 uint64_t total_bytes = rule->byte_count + extra_bytes;
2005 if (ofproto->ofhooks->account_flow_cb
2006 && total_bytes > rule->accounted_bytes)
2008 ofproto->ofhooks->account_flow_cb(
2009 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2010 total_bytes - rule->accounted_bytes, ofproto->aux);
2011 rule->accounted_bytes = total_bytes;
2016 rule_uninstall(struct ofproto *p, struct rule *rule)
2018 assert(!rule->cr.wc.wildcards);
2019 if (rule->installed) {
2020 struct odp_flow odp_flow;
2022 odp_flow.key = rule->cr.flow;
2023 odp_flow.actions = NULL;
2024 odp_flow.n_actions = 0;
2026 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2027 update_stats(p, rule, &odp_flow.stats);
2029 rule->installed = false;
2031 rule_post_uninstall(p, rule);
2036 is_controller_rule(struct rule *rule)
2038 /* If the only action is send to the controller then don't report
2039 * NetFlow expiration messages since it is just part of the control
2040 * logic for the network and not real traffic. */
2042 if (rule && rule->super) {
2043 struct rule *super = rule->super;
2045 return super->n_actions == 1 &&
2046 super->actions[0].type == htons(OFPAT_OUTPUT) &&
2047 super->actions[0].output.port == htons(OFPP_CONTROLLER);
2054 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2056 struct rule *super = rule->super;
2058 rule_account(ofproto, rule, 0);
2060 if (ofproto->netflow && !is_controller_rule(rule)) {
2061 struct ofexpired expired;
2062 expired.flow = rule->cr.flow;
2063 expired.packet_count = rule->packet_count;
2064 expired.byte_count = rule->byte_count;
2065 expired.used = rule->used;
2066 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2069 super->packet_count += rule->packet_count;
2070 super->byte_count += rule->byte_count;
2072 /* Reset counters to prevent double counting if the rule ever gets
2074 rule->packet_count = 0;
2075 rule->byte_count = 0;
2076 rule->accounted_bytes = 0;
2078 netflow_flow_clear(&rule->nf_flow);
2083 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2084 struct rconn_packet_counter *counter)
2086 update_openflow_length(msg);
2087 if (rconn_send(ofconn->rconn, msg, counter)) {
2093 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2094 int error, const void *data, size_t len)
2097 struct ofp_error_msg *oem;
2099 if (!(error >> 16)) {
2100 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2105 COVERAGE_INC(ofproto_error);
2106 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2107 oh ? oh->xid : 0, &buf);
2108 oem->type = htons((unsigned int) error >> 16);
2109 oem->code = htons(error & 0xffff);
2110 memcpy(oem->data, data, len);
2111 queue_tx(buf, ofconn, ofconn->reply_counter);
2115 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2118 size_t oh_length = ntohs(oh->length);
2119 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2123 hton_ofp_phy_port(struct ofp_phy_port *opp)
2125 opp->port_no = htons(opp->port_no);
2126 opp->config = htonl(opp->config);
2127 opp->state = htonl(opp->state);
2128 opp->curr = htonl(opp->curr);
2129 opp->advertised = htonl(opp->advertised);
2130 opp->supported = htonl(opp->supported);
2131 opp->peer = htonl(opp->peer);
2135 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2137 struct ofp_header *rq = oh;
2138 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2143 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2144 struct ofp_header *oh)
2146 struct ofp_switch_features *osf;
2148 unsigned int port_no;
2149 struct ofport *port;
2151 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2152 osf->datapath_id = htonll(p->datapath_id);
2153 osf->n_buffers = htonl(pktbuf_capacity());
2155 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2156 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2157 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2158 (1u << OFPAT_SET_VLAN_VID) |
2159 (1u << OFPAT_SET_VLAN_PCP) |
2160 (1u << OFPAT_STRIP_VLAN) |
2161 (1u << OFPAT_SET_DL_SRC) |
2162 (1u << OFPAT_SET_DL_DST) |
2163 (1u << OFPAT_SET_NW_SRC) |
2164 (1u << OFPAT_SET_NW_DST) |
2165 (1u << OFPAT_SET_NW_TOS) |
2166 (1u << OFPAT_SET_TP_SRC) |
2167 (1u << OFPAT_SET_TP_DST));
2169 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2170 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2173 queue_tx(buf, ofconn, ofconn->reply_counter);
2178 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2179 struct ofp_header *oh)
2182 struct ofp_switch_config *osc;
2186 /* Figure out flags. */
2187 dpif_get_drop_frags(p->dpif, &drop_frags);
2188 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2191 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2192 osc->flags = htons(flags);
2193 osc->miss_send_len = htons(ofconn->miss_send_len);
2194 queue_tx(buf, ofconn, ofconn->reply_counter);
2200 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2201 struct ofp_switch_config *osc)
2206 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2210 flags = ntohs(osc->flags);
2212 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2213 switch (flags & OFPC_FRAG_MASK) {
2214 case OFPC_FRAG_NORMAL:
2215 dpif_set_drop_frags(p->dpif, false);
2217 case OFPC_FRAG_DROP:
2218 dpif_set_drop_frags(p->dpif, true);
2221 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2227 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2233 add_output_group_action(struct odp_actions *actions, uint16_t group,
2234 uint16_t *nf_output_iface)
2236 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2238 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2239 *nf_output_iface = NF_OUT_FLOOD;
2244 add_controller_action(struct odp_actions *actions,
2245 const struct ofp_action_output *oao)
2247 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2248 a->controller.arg = ntohs(oao->max_len);
2251 struct action_xlate_ctx {
2253 flow_t flow; /* Flow to which these actions correspond. */
2254 int recurse; /* Recursion level, via xlate_table_action. */
2255 struct ofproto *ofproto;
2256 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2257 * null pointer if we are revalidating
2258 * without a packet to refer to. */
2261 struct odp_actions *out; /* Datapath actions. */
2262 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2263 bool may_set_up_flow; /* True ordinarily; false if the actions must
2264 * be reassessed for every packet. */
2265 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2268 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2269 struct action_xlate_ctx *ctx);
2272 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2274 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2277 if (ofport->opp.config & OFPPC_NO_FWD) {
2278 /* Forwarding disabled on port. */
2283 * We don't have an ofport record for this port, but it doesn't hurt to
2284 * allow forwarding to it anyhow. Maybe such a port will appear later
2285 * and we're pre-populating the flow table.
2289 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2290 ctx->nf_output_iface = port;
2293 static struct rule *
2294 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2297 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2299 /* The rule we found might not be valid, since we could be in need of
2300 * revalidation. If it is not valid, don't return it. */
2303 && ofproto->need_revalidate
2304 && !revalidate_rule(ofproto, rule)) {
2305 COVERAGE_INC(ofproto_invalidated);
2313 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2315 if (!ctx->recurse) {
2316 uint16_t old_in_port;
2319 /* Look up a flow with 'in_port' as the input port. Then restore the
2320 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2321 * have surprising behavior). */
2322 old_in_port = ctx->flow.in_port;
2323 ctx->flow.in_port = in_port;
2324 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2325 ctx->flow.in_port = old_in_port;
2333 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2340 xlate_output_action(struct action_xlate_ctx *ctx,
2341 const struct ofp_action_output *oao)
2344 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2346 ctx->nf_output_iface = NF_OUT_DROP;
2348 switch (ntohs(oao->port)) {
2350 add_output_action(ctx, ctx->flow.in_port);
2353 xlate_table_action(ctx, ctx->flow.in_port);
2356 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2357 ctx->out, ctx->tags,
2358 &ctx->nf_output_iface,
2359 ctx->ofproto->aux)) {
2360 COVERAGE_INC(ofproto_uninstallable);
2361 ctx->may_set_up_flow = false;
2365 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2366 &ctx->nf_output_iface);
2369 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2371 case OFPP_CONTROLLER:
2372 add_controller_action(ctx->out, oao);
2375 add_output_action(ctx, ODPP_LOCAL);
2378 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2379 if (odp_port != ctx->flow.in_port) {
2380 add_output_action(ctx, odp_port);
2385 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2386 ctx->nf_output_iface = NF_OUT_FLOOD;
2387 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2388 ctx->nf_output_iface = prev_nf_output_iface;
2389 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2390 ctx->nf_output_iface != NF_OUT_FLOOD) {
2391 ctx->nf_output_iface = NF_OUT_MULTI;
2396 xlate_nicira_action(struct action_xlate_ctx *ctx,
2397 const struct nx_action_header *nah)
2399 const struct nx_action_resubmit *nar;
2400 const struct nx_action_set_tunnel *nast;
2401 union odp_action *oa;
2402 int subtype = ntohs(nah->subtype);
2404 assert(nah->vendor == htonl(NX_VENDOR_ID));
2406 case NXAST_RESUBMIT:
2407 nar = (const struct nx_action_resubmit *) nah;
2408 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2411 case NXAST_SET_TUNNEL:
2412 nast = (const struct nx_action_set_tunnel *) nah;
2413 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2414 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2417 /* If you add a new action here that modifies flow data, don't forget to
2418 * update the flow key in ctx->flow in the same key. */
2421 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2427 do_xlate_actions(const union ofp_action *in, size_t n_in,
2428 struct action_xlate_ctx *ctx)
2430 struct actions_iterator iter;
2431 const union ofp_action *ia;
2432 const struct ofport *port;
2434 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2435 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2436 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2437 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2438 /* Drop this flow. */
2442 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2443 uint16_t type = ntohs(ia->type);
2444 union odp_action *oa;
2448 xlate_output_action(ctx, &ia->output);
2451 case OFPAT_SET_VLAN_VID:
2452 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2453 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2456 case OFPAT_SET_VLAN_PCP:
2457 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2458 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2461 case OFPAT_STRIP_VLAN:
2462 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2463 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2464 ctx->flow.dl_vlan_pcp = 0;
2467 case OFPAT_SET_DL_SRC:
2468 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2469 memcpy(oa->dl_addr.dl_addr,
2470 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2471 memcpy(ctx->flow.dl_src,
2472 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2475 case OFPAT_SET_DL_DST:
2476 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2477 memcpy(oa->dl_addr.dl_addr,
2478 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2479 memcpy(ctx->flow.dl_dst,
2480 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2483 case OFPAT_SET_NW_SRC:
2484 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2485 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2488 case OFPAT_SET_NW_DST:
2489 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2490 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2493 case OFPAT_SET_NW_TOS:
2494 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2495 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2498 case OFPAT_SET_TP_SRC:
2499 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2500 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2503 case OFPAT_SET_TP_DST:
2504 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2505 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2509 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2513 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2520 xlate_actions(const union ofp_action *in, size_t n_in,
2521 const flow_t *flow, struct ofproto *ofproto,
2522 const struct ofpbuf *packet,
2523 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2524 uint16_t *nf_output_iface)
2526 tag_type no_tags = 0;
2527 struct action_xlate_ctx ctx;
2528 COVERAGE_INC(ofproto_ofp2odp);
2529 odp_actions_init(out);
2532 ctx.ofproto = ofproto;
2533 ctx.packet = packet;
2535 ctx.tags = tags ? tags : &no_tags;
2536 ctx.may_set_up_flow = true;
2537 ctx.nf_output_iface = NF_OUT_DROP;
2538 do_xlate_actions(in, n_in, &ctx);
2540 /* Check with in-band control to see if we're allowed to set up this
2542 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2543 ctx.may_set_up_flow = false;
2546 if (may_set_up_flow) {
2547 *may_set_up_flow = ctx.may_set_up_flow;
2549 if (nf_output_iface) {
2550 *nf_output_iface = ctx.nf_output_iface;
2552 if (odp_actions_overflow(out)) {
2553 odp_actions_init(out);
2554 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2559 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2560 * error message code (composed with ofp_mkerr()) for the caller to propagate
2561 * upward. Otherwise, returns 0.
2563 * 'oh' is used to make log messages more informative. */
2565 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2567 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2568 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2571 type_name = ofp_message_type_to_string(oh->type);
2572 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2576 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2583 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2584 struct ofp_header *oh)
2586 struct ofp_packet_out *opo;
2587 struct ofpbuf payload, *buffer;
2588 struct odp_actions actions;
2594 error = reject_slave_controller(ofconn, oh);
2599 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2603 opo = (struct ofp_packet_out *) oh;
2605 COVERAGE_INC(ofproto_packet_out);
2606 if (opo->buffer_id != htonl(UINT32_MAX)) {
2607 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2609 if (error || !buffer) {
2617 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2618 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2619 &flow, p, &payload, &actions, NULL, NULL, NULL);
2624 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2626 ofpbuf_delete(buffer);
2632 update_port_config(struct ofproto *p, struct ofport *port,
2633 uint32_t config, uint32_t mask)
2635 mask &= config ^ port->opp.config;
2636 if (mask & OFPPC_PORT_DOWN) {
2637 if (config & OFPPC_PORT_DOWN) {
2638 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2640 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2643 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2644 if (mask & REVALIDATE_BITS) {
2645 COVERAGE_INC(ofproto_costly_flags);
2646 port->opp.config ^= mask & REVALIDATE_BITS;
2647 p->need_revalidate = true;
2649 #undef REVALIDATE_BITS
2650 if (mask & OFPPC_NO_FLOOD) {
2651 port->opp.config ^= OFPPC_NO_FLOOD;
2652 refresh_port_groups(p);
2654 if (mask & OFPPC_NO_PACKET_IN) {
2655 port->opp.config ^= OFPPC_NO_PACKET_IN;
2660 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2661 struct ofp_header *oh)
2663 const struct ofp_port_mod *opm;
2664 struct ofport *port;
2667 error = reject_slave_controller(ofconn, oh);
2671 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2675 opm = (struct ofp_port_mod *) oh;
2677 port = port_array_get(&p->ports,
2678 ofp_port_to_odp_port(ntohs(opm->port_no)));
2680 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2681 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2682 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2684 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2685 if (opm->advertise) {
2686 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2692 static struct ofpbuf *
2693 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2695 struct ofp_stats_reply *osr;
2698 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2699 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2701 osr->flags = htons(0);
2705 static struct ofpbuf *
2706 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2708 return make_stats_reply(request->header.xid, request->type, body_len);
2712 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2714 struct ofpbuf *msg = *msgp;
2715 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2716 if (nbytes + msg->size > UINT16_MAX) {
2717 struct ofp_stats_reply *reply = msg->data;
2718 reply->flags = htons(OFPSF_REPLY_MORE);
2719 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2720 queue_tx(msg, ofconn, ofconn->reply_counter);
2722 return ofpbuf_put_uninit(*msgp, nbytes);
2726 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2727 struct ofp_stats_request *request)
2729 struct ofp_desc_stats *ods;
2732 msg = start_stats_reply(request, sizeof *ods);
2733 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2734 memset(ods, 0, sizeof *ods);
2735 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2736 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2737 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2738 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2739 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2740 queue_tx(msg, ofconn, ofconn->reply_counter);
2746 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2748 struct rule *rule = rule_from_cls_rule(cls_rule);
2749 int *n_subrules = n_subrules_;
2757 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2758 struct ofp_stats_request *request)
2760 struct ofp_table_stats *ots;
2762 struct odp_stats dpstats;
2763 int n_exact, n_subrules, n_wild;
2765 msg = start_stats_reply(request, sizeof *ots * 2);
2767 /* Count rules of various kinds. */
2769 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2770 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2771 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2774 dpif_get_dp_stats(p->dpif, &dpstats);
2775 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2776 memset(ots, 0, sizeof *ots);
2777 ots->table_id = TABLEID_HASH;
2778 strcpy(ots->name, "hash");
2779 ots->wildcards = htonl(0);
2780 ots->max_entries = htonl(dpstats.max_capacity);
2781 ots->active_count = htonl(n_exact);
2782 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2784 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2786 /* Classifier table. */
2787 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2788 memset(ots, 0, sizeof *ots);
2789 ots->table_id = TABLEID_CLASSIFIER;
2790 strcpy(ots->name, "classifier");
2791 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2793 ots->max_entries = htonl(65536);
2794 ots->active_count = htonl(n_wild);
2795 ots->lookup_count = htonll(0); /* XXX */
2796 ots->matched_count = htonll(0); /* XXX */
2798 queue_tx(msg, ofconn, ofconn->reply_counter);
2803 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2804 struct ofpbuf **msgp)
2806 struct netdev_stats stats;
2807 struct ofp_port_stats *ops;
2809 /* Intentionally ignore return value, since errors will set
2810 * 'stats' to all-1s, which is correct for OpenFlow, and
2811 * netdev_get_stats() will log errors. */
2812 netdev_get_stats(port->netdev, &stats);
2814 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2815 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2816 memset(ops->pad, 0, sizeof ops->pad);
2817 ops->rx_packets = htonll(stats.rx_packets);
2818 ops->tx_packets = htonll(stats.tx_packets);
2819 ops->rx_bytes = htonll(stats.rx_bytes);
2820 ops->tx_bytes = htonll(stats.tx_bytes);
2821 ops->rx_dropped = htonll(stats.rx_dropped);
2822 ops->tx_dropped = htonll(stats.tx_dropped);
2823 ops->rx_errors = htonll(stats.rx_errors);
2824 ops->tx_errors = htonll(stats.tx_errors);
2825 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2826 ops->rx_over_err = htonll(stats.rx_over_errors);
2827 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2828 ops->collisions = htonll(stats.collisions);
2832 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2833 struct ofp_stats_request *osr,
2836 struct ofp_port_stats_request *psr;
2837 struct ofp_port_stats *ops;
2839 struct ofport *port;
2840 unsigned int port_no;
2842 if (arg_size != sizeof *psr) {
2843 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2845 psr = (struct ofp_port_stats_request *) osr->body;
2847 msg = start_stats_reply(osr, sizeof *ops * 16);
2848 if (psr->port_no != htons(OFPP_NONE)) {
2849 port = port_array_get(&p->ports,
2850 ofp_port_to_odp_port(ntohs(psr->port_no)));
2852 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2855 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2856 append_port_stat(port, port_no, ofconn, &msg);
2860 queue_tx(msg, ofconn, ofconn->reply_counter);
2864 struct flow_stats_cbdata {
2865 struct ofproto *ofproto;
2866 struct ofconn *ofconn;
2871 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2872 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2873 * returned statistic include statistics for all of 'rule''s subrules. */
2875 query_stats(struct ofproto *p, struct rule *rule,
2876 uint64_t *packet_countp, uint64_t *byte_countp)
2878 uint64_t packet_count, byte_count;
2879 struct rule *subrule;
2880 struct odp_flow *odp_flows;
2883 /* Start from historical data for 'rule' itself that are no longer tracked
2884 * by the datapath. This counts, for example, subrules that have
2886 packet_count = rule->packet_count;
2887 byte_count = rule->byte_count;
2889 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2890 * wildcarded then on all of its subrules.
2892 * Also, add any statistics that are not tracked by the datapath for each
2893 * subrule. This includes, for example, statistics for packets that were
2894 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2896 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2897 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2898 if (rule->cr.wc.wildcards) {
2900 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2901 odp_flows[i++].key = subrule->cr.flow;
2902 packet_count += subrule->packet_count;
2903 byte_count += subrule->byte_count;
2906 odp_flows[0].key = rule->cr.flow;
2909 /* Fetch up-to-date statistics from the datapath and add them in. */
2910 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2912 for (i = 0; i < n_odp_flows; i++) {
2913 struct odp_flow *odp_flow = &odp_flows[i];
2914 packet_count += odp_flow->stats.n_packets;
2915 byte_count += odp_flow->stats.n_bytes;
2920 /* Return the stats to the caller. */
2921 *packet_countp = packet_count;
2922 *byte_countp = byte_count;
2926 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2928 struct rule *rule = rule_from_cls_rule(rule_);
2929 struct flow_stats_cbdata *cbdata = cbdata_;
2930 struct ofp_flow_stats *ofs;
2931 uint64_t packet_count, byte_count;
2932 size_t act_len, len;
2933 long long int tdiff = time_msec() - rule->created;
2934 uint32_t sec = tdiff / 1000;
2935 uint32_t msec = tdiff - (sec * 1000);
2937 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2941 act_len = sizeof *rule->actions * rule->n_actions;
2942 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2944 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2946 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2947 ofs->length = htons(len);
2948 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2950 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2951 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2952 ofs->duration_sec = htonl(sec);
2953 ofs->duration_nsec = htonl(msec * 1000000);
2954 ofs->cookie = rule->flow_cookie;
2955 ofs->priority = htons(rule->cr.priority);
2956 ofs->idle_timeout = htons(rule->idle_timeout);
2957 ofs->hard_timeout = htons(rule->hard_timeout);
2958 memset(ofs->pad2, 0, sizeof ofs->pad2);
2959 ofs->packet_count = htonll(packet_count);
2960 ofs->byte_count = htonll(byte_count);
2961 memcpy(ofs->actions, rule->actions, act_len);
2965 table_id_to_include(uint8_t table_id)
2967 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2968 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2969 : table_id == 0xff ? CLS_INC_ALL
2974 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2975 const struct ofp_stats_request *osr,
2978 struct ofp_flow_stats_request *fsr;
2979 struct flow_stats_cbdata cbdata;
2980 struct cls_rule target;
2982 if (arg_size != sizeof *fsr) {
2983 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2985 fsr = (struct ofp_flow_stats_request *) osr->body;
2987 COVERAGE_INC(ofproto_flows_req);
2989 cbdata.ofconn = ofconn;
2990 cbdata.out_port = fsr->out_port;
2991 cbdata.msg = start_stats_reply(osr, 1024);
2992 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2993 classifier_for_each_match(&p->cls, &target,
2994 table_id_to_include(fsr->table_id),
2995 flow_stats_cb, &cbdata);
2996 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3000 struct flow_stats_ds_cbdata {
3001 struct ofproto *ofproto;
3006 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3008 struct rule *rule = rule_from_cls_rule(rule_);
3009 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3010 struct ds *results = cbdata->results;
3011 struct ofp_match match;
3012 uint64_t packet_count, byte_count;
3013 size_t act_len = sizeof *rule->actions * rule->n_actions;
3015 /* Don't report on subrules. */
3016 if (rule->super != NULL) {
3020 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3021 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3022 cbdata->ofproto->tun_id_from_cookie, &match);
3024 ds_put_format(results, "duration=%llds, ",
3025 (time_msec() - rule->created) / 1000);
3026 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3027 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3028 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3029 ofp_print_match(results, &match, true);
3030 ofp_print_actions(results, &rule->actions->header, act_len);
3031 ds_put_cstr(results, "\n");
3034 /* Adds a pretty-printed description of all flows to 'results', including
3035 * those marked hidden by secchan (e.g., by in-band control). */
3037 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3039 struct ofp_match match;
3040 struct cls_rule target;
3041 struct flow_stats_ds_cbdata cbdata;
3043 memset(&match, 0, sizeof match);
3044 match.wildcards = htonl(OVSFW_ALL);
3047 cbdata.results = results;
3049 cls_rule_from_match(&match, 0, false, 0, &target);
3050 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3051 flow_stats_ds_cb, &cbdata);
3054 struct aggregate_stats_cbdata {
3055 struct ofproto *ofproto;
3057 uint64_t packet_count;
3058 uint64_t byte_count;
3063 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3065 struct rule *rule = rule_from_cls_rule(rule_);
3066 struct aggregate_stats_cbdata *cbdata = cbdata_;
3067 uint64_t packet_count, byte_count;
3069 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3073 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3075 cbdata->packet_count += packet_count;
3076 cbdata->byte_count += byte_count;
3081 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3082 const struct ofp_stats_request *osr,
3085 struct ofp_aggregate_stats_request *asr;
3086 struct ofp_aggregate_stats_reply *reply;
3087 struct aggregate_stats_cbdata cbdata;
3088 struct cls_rule target;
3091 if (arg_size != sizeof *asr) {
3092 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3094 asr = (struct ofp_aggregate_stats_request *) osr->body;
3096 COVERAGE_INC(ofproto_agg_request);
3098 cbdata.out_port = asr->out_port;
3099 cbdata.packet_count = 0;
3100 cbdata.byte_count = 0;
3102 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3103 classifier_for_each_match(&p->cls, &target,
3104 table_id_to_include(asr->table_id),
3105 aggregate_stats_cb, &cbdata);
3107 msg = start_stats_reply(osr, sizeof *reply);
3108 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3109 reply->flow_count = htonl(cbdata.n_flows);
3110 reply->packet_count = htonll(cbdata.packet_count);
3111 reply->byte_count = htonll(cbdata.byte_count);
3112 queue_tx(msg, ofconn, ofconn->reply_counter);
3117 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3118 struct ofp_header *oh)
3120 struct ofp_stats_request *osr;
3124 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3129 osr = (struct ofp_stats_request *) oh;
3131 switch (ntohs(osr->type)) {
3133 return handle_desc_stats_request(p, ofconn, osr);
3136 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3138 case OFPST_AGGREGATE:
3139 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3142 return handle_table_stats_request(p, ofconn, osr);
3145 return handle_port_stats_request(p, ofconn, osr, arg_size);
3148 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3151 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3155 static long long int
3156 msec_from_nsec(uint64_t sec, uint32_t nsec)
3158 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3162 update_time(struct ofproto *ofproto, struct rule *rule,
3163 const struct odp_flow_stats *stats)
3165 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3166 if (used > rule->used) {
3168 if (rule->super && used > rule->super->used) {
3169 rule->super->used = used;
3171 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3176 update_stats(struct ofproto *ofproto, struct rule *rule,
3177 const struct odp_flow_stats *stats)
3179 if (stats->n_packets) {
3180 update_time(ofproto, rule, stats);
3181 rule->packet_count += stats->n_packets;
3182 rule->byte_count += stats->n_bytes;
3183 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3188 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3189 * in which no matching flow already exists in the flow table.
3191 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3192 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3193 * code as encoded by ofp_mkerr() on failure.
3195 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3198 add_flow(struct ofproto *p, struct ofconn *ofconn,
3199 const struct ofp_flow_mod *ofm, size_t n_actions)
3201 struct ofpbuf *packet;
3206 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3210 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3212 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3213 ntohs(ofm->priority))) {
3214 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3218 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3219 n_actions, ntohs(ofm->idle_timeout),
3220 ntohs(ofm->hard_timeout), ofm->cookie,
3221 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3222 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3223 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3226 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3227 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3231 in_port = UINT16_MAX;
3234 rule_insert(p, rule, packet, in_port);
3235 ofpbuf_delete(packet);
3239 static struct rule *
3240 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3245 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3247 return rule_from_cls_rule(classifier_find_rule_exactly(
3248 &p->cls, &flow, wildcards,
3249 ntohs(ofm->priority)));
3253 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3254 struct rule *rule, const struct ofp_flow_mod *ofm)
3256 struct ofpbuf *packet;
3261 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3265 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3271 flow_extract(packet, 0, in_port, &flow);
3272 rule_execute(ofproto, rule, packet, &flow);
3273 ofpbuf_delete(packet);
3278 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3280 struct modify_flows_cbdata {
3281 struct ofproto *ofproto;
3282 const struct ofp_flow_mod *ofm;
3287 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3288 size_t n_actions, struct rule *);
3289 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3291 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3292 * encoded by ofp_mkerr() on failure.
3294 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3297 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3298 const struct ofp_flow_mod *ofm, size_t n_actions)
3300 struct modify_flows_cbdata cbdata;
3301 struct cls_rule target;
3305 cbdata.n_actions = n_actions;
3306 cbdata.match = NULL;
3308 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3311 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3312 modify_flows_cb, &cbdata);
3314 /* This credits the packet to whichever flow happened to happened to
3315 * match last. That's weird. Maybe we should do a lookup for the
3316 * flow that actually matches the packet? Who knows. */
3317 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3320 return add_flow(p, ofconn, ofm, n_actions);
3324 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3325 * code as encoded by ofp_mkerr() on failure.
3327 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3330 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3331 struct ofp_flow_mod *ofm, size_t n_actions)
3333 struct rule *rule = find_flow_strict(p, ofm);
3334 if (rule && !rule_is_hidden(rule)) {
3335 modify_flow(p, ofm, n_actions, rule);
3336 return send_buffered_packet(p, ofconn, rule, ofm);
3338 return add_flow(p, ofconn, ofm, n_actions);
3342 /* Callback for modify_flows_loose(). */
3344 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3346 struct rule *rule = rule_from_cls_rule(rule_);
3347 struct modify_flows_cbdata *cbdata = cbdata_;
3349 if (!rule_is_hidden(rule)) {
3350 cbdata->match = rule;
3351 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3355 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3356 * been identified as a flow in 'p''s flow table to be modified, by changing
3357 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3358 * ofp_action[] structures). */
3360 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3361 size_t n_actions, struct rule *rule)
3363 size_t actions_len = n_actions * sizeof *rule->actions;
3365 rule->flow_cookie = ofm->cookie;
3367 /* If the actions are the same, do nothing. */
3368 if (n_actions == rule->n_actions
3369 && !memcmp(ofm->actions, rule->actions, actions_len))
3374 /* Replace actions. */
3375 free(rule->actions);
3376 rule->actions = xmemdup(ofm->actions, actions_len);
3377 rule->n_actions = n_actions;
3379 /* Make sure that the datapath gets updated properly. */
3380 if (rule->cr.wc.wildcards) {
3381 COVERAGE_INC(ofproto_mod_wc_flow);
3382 p->need_revalidate = true;
3384 rule_update_actions(p, rule);
3390 /* OFPFC_DELETE implementation. */
3392 struct delete_flows_cbdata {
3393 struct ofproto *ofproto;
3397 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3398 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3400 /* Implements OFPFC_DELETE. */
3402 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3404 struct delete_flows_cbdata cbdata;
3405 struct cls_rule target;
3408 cbdata.out_port = ofm->out_port;
3410 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3413 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3414 delete_flows_cb, &cbdata);
3417 /* Implements OFPFC_DELETE_STRICT. */
3419 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3421 struct rule *rule = find_flow_strict(p, ofm);
3423 delete_flow(p, rule, ofm->out_port);
3427 /* Callback for delete_flows_loose(). */
3429 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3431 struct rule *rule = rule_from_cls_rule(rule_);
3432 struct delete_flows_cbdata *cbdata = cbdata_;
3434 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3437 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3438 * been identified as a flow to delete from 'p''s flow table, by deleting the
3439 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3442 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3443 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3444 * specified 'out_port'. */
3446 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3448 if (rule_is_hidden(rule)) {
3452 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3456 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3457 rule_remove(p, rule);
3461 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3462 struct ofp_flow_mod *ofm)
3467 error = reject_slave_controller(ofconn, &ofm->header);
3471 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3472 sizeof *ofm->actions, &n_actions);
3477 /* We do not support the emergency flow cache. It will hopefully
3478 * get dropped from OpenFlow in the near future. */
3479 if (ofm->flags & htons(OFPFF_EMERG)) {
3480 /* There isn't a good fit for an error code, so just state that the
3481 * flow table is full. */
3482 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3485 normalize_match(&ofm->match);
3486 if (!ofm->match.wildcards) {
3487 ofm->priority = htons(UINT16_MAX);
3490 error = validate_actions((const union ofp_action *) ofm->actions,
3491 n_actions, p->max_ports);
3496 switch (ntohs(ofm->command)) {
3498 return add_flow(p, ofconn, ofm, n_actions);
3501 return modify_flows_loose(p, ofconn, ofm, n_actions);
3503 case OFPFC_MODIFY_STRICT:
3504 return modify_flow_strict(p, ofconn, ofm, n_actions);
3507 delete_flows_loose(p, ofm);
3510 case OFPFC_DELETE_STRICT:
3511 delete_flow_strict(p, ofm);
3515 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3520 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3524 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3529 p->tun_id_from_cookie = !!msg->set;
3534 handle_role_request(struct ofproto *ofproto,
3535 struct ofconn *ofconn, struct nicira_header *msg)
3537 struct nx_role_request *nrr;
3538 struct nx_role_request *reply;
3542 if (ntohs(msg->header.length) != sizeof *nrr) {
3543 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3544 ntohs(msg->header.length), sizeof *nrr);
3545 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3547 nrr = (struct nx_role_request *) msg;
3549 if (ofconn->type != OFCONN_CONTROLLER) {
3550 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3552 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3555 role = ntohl(nrr->role);
3556 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3557 && role != NX_ROLE_SLAVE) {
3558 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3560 /* There's no good error code for this. */
3561 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3564 if (role == NX_ROLE_MASTER) {
3565 struct ofconn *other;
3567 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3568 &ofproto->controllers) {
3569 if (other->role == NX_ROLE_MASTER) {
3570 other->role = NX_ROLE_SLAVE;
3574 ofconn->role = role;
3576 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3578 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3579 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3580 reply->role = htonl(role);
3581 queue_tx(buf, ofconn, ofconn->reply_counter);
3587 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3589 struct ofp_vendor_header *ovh = msg;
3590 struct nicira_header *nh;
3592 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3593 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3594 "(expected at least %zu)",
3595 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3596 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3598 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3599 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3601 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3602 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3603 "(expected at least %zu)",
3604 ntohs(ovh->header.length), sizeof(struct nicira_header));
3605 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3609 switch (ntohl(nh->subtype)) {
3610 case NXT_STATUS_REQUEST:
3611 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3614 case NXT_TUN_ID_FROM_COOKIE:
3615 return handle_tun_id_from_cookie(p, msg);
3617 case NXT_ROLE_REQUEST:
3618 return handle_role_request(p, ofconn, msg);
3621 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3625 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3627 struct ofp_header *ob;
3630 /* Currently, everything executes synchronously, so we can just
3631 * immediately send the barrier reply. */
3632 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3633 queue_tx(buf, ofconn, ofconn->reply_counter);
3638 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3639 struct ofpbuf *ofp_msg)
3641 struct ofp_header *oh = ofp_msg->data;
3644 COVERAGE_INC(ofproto_recv_openflow);
3646 case OFPT_ECHO_REQUEST:
3647 error = handle_echo_request(ofconn, oh);
3650 case OFPT_ECHO_REPLY:
3654 case OFPT_FEATURES_REQUEST:
3655 error = handle_features_request(p, ofconn, oh);
3658 case OFPT_GET_CONFIG_REQUEST:
3659 error = handle_get_config_request(p, ofconn, oh);
3662 case OFPT_SET_CONFIG:
3663 error = handle_set_config(p, ofconn, ofp_msg->data);
3666 case OFPT_PACKET_OUT:
3667 error = handle_packet_out(p, ofconn, ofp_msg->data);
3671 error = handle_port_mod(p, ofconn, oh);
3675 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3678 case OFPT_STATS_REQUEST:
3679 error = handle_stats_request(p, ofconn, oh);
3683 error = handle_vendor(p, ofconn, ofp_msg->data);
3686 case OFPT_BARRIER_REQUEST:
3687 error = handle_barrier_request(ofconn, oh);
3691 if (VLOG_IS_WARN_ENABLED()) {
3692 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3693 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3696 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3701 send_error_oh(ofconn, ofp_msg->data, error);
3706 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3708 struct odp_msg *msg = packet->data;
3710 struct ofpbuf payload;
3713 payload.data = msg + 1;
3714 payload.size = msg->length - sizeof *msg;
3715 flow_extract(&payload, msg->arg, msg->port, &flow);
3717 /* Check with in-band control to see if this packet should be sent
3718 * to the local port regardless of the flow table. */
3719 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3720 union odp_action action;
3722 memset(&action, 0, sizeof(action));
3723 action.output.type = ODPAT_OUTPUT;
3724 action.output.port = ODPP_LOCAL;
3725 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3728 rule = lookup_valid_rule(p, &flow);
3730 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3731 struct ofport *port = port_array_get(&p->ports, msg->port);
3733 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3734 COVERAGE_INC(ofproto_no_packet_in);
3735 /* XXX install 'drop' flow entry */
3736 ofpbuf_delete(packet);
3740 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3743 COVERAGE_INC(ofproto_packet_in);
3744 send_packet_in(p, packet);
3748 if (rule->cr.wc.wildcards) {
3749 rule = rule_create_subrule(p, rule, &flow);
3750 rule_make_actions(p, rule, packet);
3752 if (!rule->may_install) {
3753 /* The rule is not installable, that is, we need to process every
3754 * packet, so process the current packet and set its actions into
3756 rule_make_actions(p, rule, packet);
3758 /* XXX revalidate rule if it needs it */
3762 rule_execute(p, rule, &payload, &flow);
3763 rule_reinstall(p, rule);
3765 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3767 * Extra-special case for fail-open mode.
3769 * We are in fail-open mode and the packet matched the fail-open rule,
3770 * but we are connected to a controller too. We should send the packet
3771 * up to the controller in the hope that it will try to set up a flow
3772 * and thereby allow us to exit fail-open.
3774 * See the top-level comment in fail-open.c for more information.
3776 send_packet_in(p, packet);
3778 ofpbuf_delete(packet);
3783 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3785 struct odp_msg *msg = packet->data;
3787 switch (msg->type) {
3788 case _ODPL_ACTION_NR:
3789 COVERAGE_INC(ofproto_ctlr_action);
3790 send_packet_in(p, packet);
3793 case _ODPL_SFLOW_NR:
3795 ofproto_sflow_received(p->sflow, msg);
3797 ofpbuf_delete(packet);
3801 handle_odp_miss_msg(p, packet);
3805 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3812 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3814 struct rule *sub = rule_from_cls_rule(sub_);
3815 struct revalidate_cbdata *cbdata = cbdata_;
3817 if (cbdata->revalidate_all
3818 || (cbdata->revalidate_subrules && sub->super)
3819 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3820 revalidate_rule(cbdata->ofproto, sub);
3825 revalidate_rule(struct ofproto *p, struct rule *rule)
3827 const flow_t *flow = &rule->cr.flow;
3829 COVERAGE_INC(ofproto_revalidate_rule);
3832 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3834 rule_remove(p, rule);
3836 } else if (super != rule->super) {
3837 COVERAGE_INC(ofproto_revalidate_moved);
3838 list_remove(&rule->list);
3839 list_push_back(&super->list, &rule->list);
3840 rule->super = super;
3841 rule->hard_timeout = super->hard_timeout;
3842 rule->idle_timeout = super->idle_timeout;
3843 rule->created = super->created;
3848 rule_update_actions(p, rule);
3852 static struct ofpbuf *
3853 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3854 long long int now, uint8_t reason)
3856 struct ofp_flow_removed *ofr;
3858 long long int tdiff = now - rule->created;
3859 uint32_t sec = tdiff / 1000;
3860 uint32_t msec = tdiff - (sec * 1000);
3862 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3863 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3865 ofr->cookie = rule->flow_cookie;
3866 ofr->priority = htons(rule->cr.priority);
3867 ofr->reason = reason;
3868 ofr->duration_sec = htonl(sec);
3869 ofr->duration_nsec = htonl(msec * 1000000);
3870 ofr->idle_timeout = htons(rule->idle_timeout);
3871 ofr->packet_count = htonll(rule->packet_count);
3872 ofr->byte_count = htonll(rule->byte_count);
3878 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3880 assert(rule->installed);
3881 assert(!rule->cr.wc.wildcards);
3884 rule_remove(ofproto, rule);
3886 rule_uninstall(ofproto, rule);
3891 send_flow_removed(struct ofproto *p, struct rule *rule,
3892 long long int now, uint8_t reason)
3894 struct ofconn *ofconn;
3895 struct ofconn *prev;
3896 struct ofpbuf *buf = NULL;
3898 /* We limit the maximum number of queued flow expirations it by accounting
3899 * them under the counter for replies. That works because preventing
3900 * OpenFlow requests from being processed also prevents new flows from
3901 * being added (and expiring). (It also prevents processing OpenFlow
3902 * requests that would not add new flows, so it is imperfect.) */
3905 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3906 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
3907 && ofconn_receives_async_msgs(ofconn)) {
3909 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3911 buf = compose_flow_removed(p, rule, now, reason);
3917 queue_tx(buf, prev, prev->reply_counter);
3923 expire_rule(struct cls_rule *cls_rule, void *p_)
3925 struct ofproto *p = p_;
3926 struct rule *rule = rule_from_cls_rule(cls_rule);
3927 long long int hard_expire, idle_expire, expire, now;
3929 hard_expire = (rule->hard_timeout
3930 ? rule->created + rule->hard_timeout * 1000
3932 idle_expire = (rule->idle_timeout
3933 && (rule->super || list_is_empty(&rule->list))
3934 ? rule->used + rule->idle_timeout * 1000
3936 expire = MIN(hard_expire, idle_expire);
3940 if (rule->installed && now >= rule->used + 5000) {
3941 uninstall_idle_flow(p, rule);
3942 } else if (!rule->cr.wc.wildcards) {
3943 active_timeout(p, rule);
3949 COVERAGE_INC(ofproto_expired);
3951 /* Update stats. This code will be a no-op if the rule expired
3952 * due to an idle timeout. */
3953 if (rule->cr.wc.wildcards) {
3954 struct rule *subrule, *next;
3955 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3956 rule_remove(p, subrule);
3959 rule_uninstall(p, rule);
3962 if (!rule_is_hidden(rule)) {
3963 send_flow_removed(p, rule, now,
3965 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3967 rule_remove(p, rule);
3971 active_timeout(struct ofproto *ofproto, struct rule *rule)
3973 if (ofproto->netflow && !is_controller_rule(rule) &&
3974 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3975 struct ofexpired expired;
3976 struct odp_flow odp_flow;
3978 /* Get updated flow stats. */
3979 memset(&odp_flow, 0, sizeof odp_flow);
3980 if (rule->installed) {
3981 odp_flow.key = rule->cr.flow;
3982 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3983 dpif_flow_get(ofproto->dpif, &odp_flow);
3985 if (odp_flow.stats.n_packets) {
3986 update_time(ofproto, rule, &odp_flow.stats);
3987 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3988 odp_flow.stats.tcp_flags);
3992 expired.flow = rule->cr.flow;
3993 expired.packet_count = rule->packet_count +
3994 odp_flow.stats.n_packets;
3995 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3996 expired.used = rule->used;
3998 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4000 /* Schedule us to send the accumulated records once we have
4001 * collected all of them. */
4002 poll_immediate_wake();
4007 update_used(struct ofproto *p)
4009 struct odp_flow *flows;
4014 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4019 for (i = 0; i < n_flows; i++) {
4020 struct odp_flow *f = &flows[i];
4023 rule = rule_from_cls_rule(
4024 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4025 if (!rule || !rule->installed) {
4026 COVERAGE_INC(ofproto_unexpected_rule);
4027 dpif_flow_del(p->dpif, f);
4031 update_time(p, rule, &f->stats);
4032 rule_account(p, rule, f->stats.n_bytes);
4037 /* pinsched callback for sending 'packet' on 'ofconn'. */
4039 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4041 struct ofconn *ofconn = ofconn_;
4043 rconn_send_with_limit(ofconn->rconn, packet,
4044 ofconn->packet_in_counter, 100);
4047 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4048 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4049 * packet scheduler for sending.
4051 * 'max_len' specifies the maximum number of bytes of the packet to send on
4052 * 'ofconn' (INT_MAX specifies no limit).
4054 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4055 * ownership is transferred to this function. */
4057 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4060 struct ofproto *ofproto = ofconn->ofproto;
4061 struct ofp_packet_in *opi = packet->data;
4062 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4063 int send_len, trim_size;
4067 if (opi->reason == OFPR_ACTION) {
4068 buffer_id = UINT32_MAX;
4069 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4070 buffer_id = pktbuf_get_null();
4071 } else if (!ofconn->pktbuf) {
4072 buffer_id = UINT32_MAX;
4074 struct ofpbuf payload;
4075 payload.data = opi->data;
4076 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4077 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4080 /* Figure out how much of the packet to send. */
4081 send_len = ntohs(opi->total_len);
4082 if (buffer_id != UINT32_MAX) {
4083 send_len = MIN(send_len, ofconn->miss_send_len);
4085 send_len = MIN(send_len, max_len);
4087 /* Adjust packet length and clone if necessary. */
4088 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4090 packet = ofpbuf_clone_data(packet->data, trim_size);
4093 packet->size = trim_size;
4096 /* Update packet headers. */
4097 opi->buffer_id = htonl(buffer_id);
4098 update_openflow_length(packet);
4100 /* Hand over to packet scheduler. It might immediately call into
4101 * do_send_packet_in() or it might buffer it for a while (until a later
4102 * call to pinsched_run()). */
4103 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4104 packet, do_send_packet_in, ofconn);
4107 /* Replace struct odp_msg header in 'packet' by equivalent struct
4108 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4109 * returned by dpif_recv()).
4111 * The conversion is not complete: the caller still needs to trim any unneeded
4112 * payload off the end of the buffer, set the length in the OpenFlow header,
4113 * and set buffer_id. Those require us to know the controller settings and so
4114 * must be done on a per-controller basis.
4116 * Returns the maximum number of bytes of the packet that should be sent to
4117 * the controller (INT_MAX if no limit). */
4119 do_convert_to_packet_in(struct ofpbuf *packet)
4121 struct odp_msg *msg = packet->data;
4122 struct ofp_packet_in *opi;
4128 /* Extract relevant header fields */
4129 if (msg->type == _ODPL_ACTION_NR) {
4130 reason = OFPR_ACTION;
4133 reason = OFPR_NO_MATCH;
4136 total_len = msg->length - sizeof *msg;
4137 in_port = odp_port_to_ofp_port(msg->port);
4139 /* Repurpose packet buffer by overwriting header. */
4140 ofpbuf_pull(packet, sizeof(struct odp_msg));
4141 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4142 opi->header.version = OFP_VERSION;
4143 opi->header.type = OFPT_PACKET_IN;
4144 opi->total_len = htons(total_len);
4145 opi->in_port = htons(in_port);
4146 opi->reason = reason;
4151 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4152 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4153 * as necessary according to their individual configurations.
4155 * 'packet' must have sufficient headroom to convert it into a struct
4156 * ofp_packet_in (e.g. as returned by dpif_recv()).
4158 * Takes ownership of 'packet'. */
4160 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4162 struct ofconn *ofconn, *prev;
4165 max_len = do_convert_to_packet_in(packet);
4168 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4169 if (ofconn_receives_async_msgs(ofconn)) {
4171 schedule_packet_in(prev, packet, max_len, true);
4177 schedule_packet_in(prev, packet, max_len, false);
4179 ofpbuf_delete(packet);
4184 pick_datapath_id(const struct ofproto *ofproto)
4186 const struct ofport *port;
4188 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4190 uint8_t ea[ETH_ADDR_LEN];
4193 error = netdev_get_etheraddr(port->netdev, ea);
4195 return eth_addr_to_uint64(ea);
4197 VLOG_WARN("could not get MAC address for %s (%s)",
4198 netdev_get_name(port->netdev), strerror(error));
4200 return ofproto->fallback_dpid;
4204 pick_fallback_dpid(void)
4206 uint8_t ea[ETH_ADDR_LEN];
4207 eth_addr_nicira_random(ea);
4208 return eth_addr_to_uint64(ea);
4212 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4213 struct odp_actions *actions, tag_type *tags,
4214 uint16_t *nf_output_iface, void *ofproto_)
4216 struct ofproto *ofproto = ofproto_;
4219 /* Drop frames for reserved multicast addresses. */
4220 if (eth_addr_is_reserved(flow->dl_dst)) {
4224 /* Learn source MAC (but don't try to learn from revalidation). */
4225 if (packet != NULL) {
4226 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4229 /* The log messages here could actually be useful in debugging,
4230 * so keep the rate limit relatively high. */
4231 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4232 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4233 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4234 ofproto_revalidate(ofproto, rev_tag);
4238 /* Determine output port. */
4239 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
4241 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4242 } else if (out_port != flow->in_port) {
4243 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4244 *nf_output_iface = out_port;
4252 static const struct ofhooks default_ofhooks = {
4254 default_normal_ofhook_cb,