2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
34 #include "mac-learning.h"
38 #include "ofp-print.h"
39 #include "ofproto-sflow.h"
41 #include "openflow/nicira-ext.h"
42 #include "openflow/openflow.h"
43 #include "openvswitch/datapath-protocol.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
53 #include "stream-ssl.h"
61 #define THIS_MODULE VLM_ofproto
64 #include "sflow_api.h"
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_set_up_flow, uint16_t *nf_output_iface);
88 uint64_t flow_cookie; /* Controller-issued identifier.
89 (Kept in network-byte order.) */
90 uint16_t idle_timeout; /* In seconds from time of last use. */
91 uint16_t hard_timeout; /* In seconds from time of creation. */
92 bool send_flow_removed; /* Send a flow removed message? */
93 long long int used; /* Last-used time (0 if never used). */
94 long long int created; /* Creation time. */
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
97 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
98 tag_type tags; /* Tags (set only by hooks). */
99 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
101 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
102 * exact-match rule (having cr.wc.wildcards of 0) generated from the
103 * wildcard rule 'super'. In this case, 'list' is an element of the
106 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
107 * a list of subrules. A super-rule with no wildcards (where
108 * cr.wc.wildcards is 0) will never have any subrules. */
114 * 'n_actions' is the number of elements in the 'actions' array. A single
115 * action may take up more more than one element's worth of space.
117 * A subrule has no actions (it uses the super-rule's actions). */
119 union ofp_action *actions;
123 * A super-rule with wildcard fields never has ODP actions (since the
124 * datapath only supports exact-match flows). */
125 bool installed; /* Installed in datapath? */
126 bool may_install; /* True ordinarily; false if actions must
127 * be reassessed for every packet. */
129 union odp_action *odp_actions;
133 rule_is_hidden(const struct rule *rule)
135 /* Subrules are merely an implementation detail, so hide them from the
137 if (rule->super != NULL) {
141 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
142 * (e.g. by in-band control) and are intentionally hidden from the
144 if (rule->cr.priority > UINT16_MAX) {
151 static struct rule *rule_create(struct ofproto *, struct rule *super,
152 const union ofp_action *, size_t n_actions,
153 uint16_t idle_timeout, uint16_t hard_timeout,
154 uint64_t flow_cookie, bool send_flow_removed);
155 static void rule_free(struct rule *);
156 static void rule_destroy(struct ofproto *, struct rule *);
157 static struct rule *rule_from_cls_rule(const struct cls_rule *);
158 static void rule_insert(struct ofproto *, struct rule *,
159 struct ofpbuf *packet, uint16_t in_port);
160 static void rule_remove(struct ofproto *, struct rule *);
161 static bool rule_make_actions(struct ofproto *, struct rule *,
162 const struct ofpbuf *packet);
163 static void rule_install(struct ofproto *, struct rule *,
164 struct rule *displaced_rule);
165 static void rule_uninstall(struct ofproto *, struct rule *);
166 static void rule_post_uninstall(struct ofproto *, struct rule *);
167 static void send_flow_removed(struct ofproto *p, struct rule *rule,
168 long long int now, uint8_t reason);
170 /* ofproto supports two kinds of OpenFlow connections:
172 * - "Controller connections": Connections to ordinary OpenFlow controllers.
173 * ofproto maintains persistent connections to these controllers and by
174 * default sends them asynchronous messages such as packet-ins.
176 * - "Transient connections", e.g. from ovs-ofctl. When these connections
177 * drop, it is the other side's responsibility to reconnect them if
178 * necessary. ofproto does not send them asynchronous messages by default.
181 OFCONN_CONTROLLER, /* An OpenFlow controller. */
182 OFCONN_TRANSIENT /* A transient connection. */
185 /* An OpenFlow connection. */
187 struct ofproto *ofproto; /* The ofproto that owns this connection. */
188 struct list node; /* In struct ofproto's "all_conns" list. */
189 struct rconn *rconn; /* OpenFlow connection. */
190 enum ofconn_type type; /* Type. */
192 /* OFPT_PACKET_IN related data. */
193 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
194 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
195 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
196 int miss_send_len; /* Bytes to send of buffered packets. */
198 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
199 * requests, and the maximum number before we stop reading OpenFlow
201 #define OFCONN_REPLY_MAX 100
202 struct rconn_packet_counter *reply_counter;
204 /* type == OFCONN_CONTROLLER only. */
205 enum nx_role role; /* Role. */
206 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
207 struct discovery *discovery; /* Controller discovery object, if enabled. */
208 struct status_category *ss; /* Switch status category. */
209 enum ofproto_band band; /* In-band or out-of-band? */
212 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
213 * "schedulers" array. Their values are 0 and 1, and their meanings and values
214 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
215 * case anything ever changes, check their values here. */
216 #define N_SCHEDULERS 2
217 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
218 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
219 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
220 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
222 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
224 static void ofconn_destroy(struct ofconn *);
225 static void ofconn_run(struct ofconn *, struct ofproto *);
226 static void ofconn_wait(struct ofconn *);
227 static bool ofconn_receives_async_msgs(const struct ofconn *);
228 static char *ofconn_make_name(const struct ofproto *, const char *target);
230 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
231 struct rconn_packet_counter *counter);
233 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
234 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
238 uint64_t datapath_id; /* Datapath ID. */
239 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
240 char *mfr_desc; /* Manufacturer. */
241 char *hw_desc; /* Hardware. */
242 char *sw_desc; /* Software version. */
243 char *serial_desc; /* Serial number. */
244 char *dp_desc; /* Datapath description. */
248 struct netdev_monitor *netdev_monitor;
249 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
251 struct shash port_by_name;
255 struct switch_status *switch_status;
256 struct fail_open *fail_open;
257 struct netflow *netflow;
258 struct ofproto_sflow *sflow;
260 /* In-band control. */
261 struct in_band *in_band;
262 long long int next_in_band_update;
263 struct sockaddr_in *extra_in_band_remotes;
264 size_t n_extra_remotes;
267 struct classifier cls;
268 bool need_revalidate;
269 long long int next_expiration;
270 struct tag_set revalidate_set;
271 bool tun_id_from_cookie;
273 /* OpenFlow connections. */
274 struct hmap controllers; /* Controller "struct ofconn"s. */
275 struct list all_conns; /* Contains "struct ofconn"s. */
276 struct pvconn **listeners;
278 struct pvconn **snoops;
281 /* Hooks for ovs-vswitchd. */
282 const struct ofhooks *ofhooks;
285 /* Used by default ofhooks. */
286 struct mac_learning *ml;
289 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
291 static const struct ofhooks default_ofhooks;
293 static uint64_t pick_datapath_id(const struct ofproto *);
294 static uint64_t pick_fallback_dpid(void);
296 static void update_used(struct ofproto *);
297 static void update_stats(struct ofproto *, struct rule *,
298 const struct odp_flow_stats *);
299 static void expire_rule(struct cls_rule *, void *ofproto);
300 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
301 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
302 static void revalidate_cb(struct cls_rule *rule_, void *p_);
304 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
306 static void handle_openflow(struct ofconn *, struct ofproto *,
309 static void refresh_port_groups(struct ofproto *);
311 static void update_port(struct ofproto *, const char *devname);
312 static int init_ports(struct ofproto *);
313 static void reinit_ports(struct ofproto *);
316 ofproto_create(const char *datapath, const char *datapath_type,
317 const struct ofhooks *ofhooks, void *aux,
318 struct ofproto **ofprotop)
320 struct odp_stats stats;
327 /* Connect to datapath and start listening for messages. */
328 error = dpif_open(datapath, datapath_type, &dpif);
330 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
333 error = dpif_get_dp_stats(dpif, &stats);
335 VLOG_ERR("failed to obtain stats for datapath %s: %s",
336 datapath, strerror(error));
340 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
342 VLOG_ERR("failed to listen on datapath %s: %s",
343 datapath, strerror(error));
347 dpif_flow_flush(dpif);
348 dpif_recv_purge(dpif);
350 /* Initialize settings. */
351 p = xzalloc(sizeof *p);
352 p->fallback_dpid = pick_fallback_dpid();
353 p->datapath_id = p->fallback_dpid;
354 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
355 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
356 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
357 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
358 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
360 /* Initialize datapath. */
362 p->netdev_monitor = netdev_monitor_create();
363 port_array_init(&p->ports);
364 shash_init(&p->port_by_name);
365 p->max_ports = stats.max_ports;
367 /* Initialize submodules. */
368 p->switch_status = switch_status_create(p);
374 /* Initialize flow table. */
375 classifier_init(&p->cls);
376 p->need_revalidate = false;
377 p->next_expiration = time_msec() + 1000;
378 tag_set_init(&p->revalidate_set);
380 /* Initialize OpenFlow connections. */
381 list_init(&p->all_conns);
382 hmap_init(&p->controllers);
388 /* Initialize hooks. */
390 p->ofhooks = ofhooks;
394 p->ofhooks = &default_ofhooks;
396 p->ml = mac_learning_create();
399 /* Pick final datapath ID. */
400 p->datapath_id = pick_datapath_id(p);
401 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
408 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
410 uint64_t old_dpid = p->datapath_id;
411 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
412 if (p->datapath_id != old_dpid) {
413 struct ofconn *ofconn;
415 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
417 /* Force all active connections to reconnect, since there is no way to
418 * notify a controller that the datapath ID has changed. */
419 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
420 rconn_reconnect(ofconn->rconn);
426 is_discovery_controller(const struct ofproto_controller *c)
428 return !strcmp(c->target, "discover");
432 is_in_band_controller(const struct ofproto_controller *c)
434 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
437 /* Creates a new controller in 'ofproto'. Some of the settings are initially
438 * drawn from 'c', but update_controller() needs to be called later to finish
439 * the new ofconn's configuration. */
441 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
443 struct discovery *discovery;
444 struct ofconn *ofconn;
446 if (is_discovery_controller(c)) {
447 int error = discovery_create(c->accept_re, c->update_resolv_conf,
448 ofproto->dpif, ofproto->switch_status,
457 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
458 ofconn->pktbuf = pktbuf_create();
459 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
461 ofconn->discovery = discovery;
463 char *name = ofconn_make_name(ofproto, c->target);
464 rconn_connect(ofconn->rconn, c->target, name);
467 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
468 hash_string(c->target, 0));
471 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
472 * target or turn discovery on or off (these are done by creating new ofconns
473 * and deleting old ones), but it can update the rest of an ofconn's
476 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
478 struct ofproto *ofproto = ofconn->ofproto;
482 ofconn->band = (is_in_band_controller(c)
483 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
485 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
487 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
488 rconn_set_probe_interval(ofconn->rconn, probe_interval);
490 if (ofconn->discovery) {
491 discovery_set_update_resolv_conf(ofconn->discovery,
492 c->update_resolv_conf);
493 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
496 for (i = 0; i < N_SCHEDULERS; i++) {
497 struct pinsched **s = &ofconn->schedulers[i];
499 if (c->rate_limit > 0) {
501 *s = pinsched_create(c->rate_limit, c->burst_limit,
502 ofproto->switch_status);
504 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
507 pinsched_destroy(*s);
514 ofconn_get_target(const struct ofconn *ofconn)
516 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
519 static struct ofconn *
520 find_controller_by_target(struct ofproto *ofproto, const char *target)
522 struct ofconn *ofconn;
524 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
525 hash_string(target, 0), &ofproto->controllers) {
526 if (!strcmp(ofconn_get_target(ofconn), target)) {
534 update_in_band_remotes(struct ofproto *ofproto)
536 const struct ofconn *ofconn;
537 struct sockaddr_in *addrs;
538 size_t max_addrs, n_addrs;
542 /* Allocate enough memory for as many remotes as we could possibly have. */
543 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
544 addrs = xmalloc(max_addrs * sizeof *addrs);
547 /* Add all the remotes. */
549 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
550 struct sockaddr_in *sin = &addrs[n_addrs];
552 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
556 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
557 if (sin->sin_addr.s_addr) {
558 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
561 if (ofconn->discovery) {
565 for (i = 0; i < ofproto->n_extra_remotes; i++) {
566 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
569 /* Create or update or destroy in-band.
571 * Ordinarily we only enable in-band if there's at least one remote
572 * address, but discovery needs the in-band rules for DHCP to be installed
573 * even before we know any remote addresses. */
574 if (n_addrs || discovery) {
575 if (!ofproto->in_band) {
576 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
579 if (ofproto->in_band) {
580 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
582 ofproto->next_in_band_update = time_msec() + 1000;
584 in_band_destroy(ofproto->in_band);
585 ofproto->in_band = NULL;
593 ofproto_set_controllers(struct ofproto *p,
594 const struct ofproto_controller *controllers,
595 size_t n_controllers)
597 struct shash new_controllers;
598 enum ofproto_fail_mode fail_mode;
599 struct ofconn *ofconn, *next;
603 shash_init(&new_controllers);
604 for (i = 0; i < n_controllers; i++) {
605 const struct ofproto_controller *c = &controllers[i];
607 shash_add_once(&new_controllers, c->target, &controllers[i]);
608 if (!find_controller_by_target(p, c->target)) {
609 add_controller(p, c);
613 fail_mode = OFPROTO_FAIL_STANDALONE;
615 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
617 struct ofproto_controller *c;
619 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
621 ofconn_destroy(ofconn);
623 update_controller(ofconn, c);
627 if (c->fail == OFPROTO_FAIL_SECURE) {
628 fail_mode = OFPROTO_FAIL_SECURE;
632 shash_destroy(&new_controllers);
634 update_in_band_remotes(p);
636 if (!hmap_is_empty(&p->controllers)
637 && fail_mode == OFPROTO_FAIL_STANDALONE) {
638 struct rconn **rconns;
642 p->fail_open = fail_open_create(p, p->switch_status);
646 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
647 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
648 rconns[n++] = ofconn->rconn;
651 fail_open_set_controllers(p->fail_open, rconns, n);
652 /* p->fail_open takes ownership of 'rconns'. */
654 fail_open_destroy(p->fail_open);
658 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
659 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
660 struct ofconn, hmap_node);
661 ofconn->ss = switch_status_register(p->switch_status, "remote",
662 rconn_status_cb, ofconn->rconn);
667 any_extras_changed(const struct ofproto *ofproto,
668 const struct sockaddr_in *extras, size_t n)
672 if (n != ofproto->n_extra_remotes) {
676 for (i = 0; i < n; i++) {
677 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
678 const struct sockaddr_in *new = &extras[i];
680 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
681 old->sin_port != new->sin_port) {
689 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
690 * in-band control should guarantee access, in the same way that in-band
691 * control guarantees access to OpenFlow controllers. */
693 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
694 const struct sockaddr_in *extras, size_t n)
696 if (!any_extras_changed(ofproto, extras, n)) {
700 free(ofproto->extra_in_band_remotes);
701 ofproto->n_extra_remotes = n;
702 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
704 update_in_band_remotes(ofproto);
708 ofproto_set_desc(struct ofproto *p,
709 const char *mfr_desc, const char *hw_desc,
710 const char *sw_desc, const char *serial_desc,
713 struct ofp_desc_stats *ods;
716 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
717 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
718 sizeof ods->mfr_desc);
721 p->mfr_desc = xstrdup(mfr_desc);
724 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
725 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
726 sizeof ods->hw_desc);
729 p->hw_desc = xstrdup(hw_desc);
732 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
733 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
734 sizeof ods->sw_desc);
737 p->sw_desc = xstrdup(sw_desc);
740 if (strlen(serial_desc) >= sizeof ods->serial_num) {
741 VLOG_WARN("truncating serial_desc, must be less than %zu "
743 sizeof ods->serial_num);
745 free(p->serial_desc);
746 p->serial_desc = xstrdup(serial_desc);
749 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
750 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
751 sizeof ods->dp_desc);
754 p->dp_desc = xstrdup(dp_desc);
759 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
760 const struct svec *svec)
762 struct pvconn **pvconns = *pvconnsp;
763 size_t n_pvconns = *n_pvconnsp;
767 for (i = 0; i < n_pvconns; i++) {
768 pvconn_close(pvconns[i]);
772 pvconns = xmalloc(svec->n * sizeof *pvconns);
774 for (i = 0; i < svec->n; i++) {
775 const char *name = svec->names[i];
776 struct pvconn *pvconn;
779 error = pvconn_open(name, &pvconn);
781 pvconns[n_pvconns++] = pvconn;
783 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
791 *n_pvconnsp = n_pvconns;
797 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
799 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
803 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
805 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
809 ofproto_set_netflow(struct ofproto *ofproto,
810 const struct netflow_options *nf_options)
812 if (nf_options && nf_options->collectors.n) {
813 if (!ofproto->netflow) {
814 ofproto->netflow = netflow_create();
816 return netflow_set_options(ofproto->netflow, nf_options);
818 netflow_destroy(ofproto->netflow);
819 ofproto->netflow = NULL;
825 ofproto_set_sflow(struct ofproto *ofproto,
826 const struct ofproto_sflow_options *oso)
828 struct ofproto_sflow *os = ofproto->sflow;
831 struct ofport *ofport;
832 unsigned int odp_port;
834 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
835 refresh_port_groups(ofproto);
836 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
837 ofproto_sflow_add_port(os, odp_port,
838 netdev_get_name(ofport->netdev));
841 ofproto_sflow_set_options(os, oso);
843 ofproto_sflow_destroy(os);
844 ofproto->sflow = NULL;
849 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
853 VLOG_WARN("STP is not yet implemented");
861 ofproto_get_datapath_id(const struct ofproto *ofproto)
863 return ofproto->datapath_id;
867 ofproto_has_controller(const struct ofproto *ofproto)
869 return !hmap_is_empty(&ofproto->controllers);
873 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
877 for (i = 0; i < ofproto->n_listeners; i++) {
878 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
883 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
887 for (i = 0; i < ofproto->n_snoops; i++) {
888 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
893 ofproto_destroy(struct ofproto *p)
895 struct ofconn *ofconn, *next_ofconn;
896 struct ofport *ofport;
897 unsigned int port_no;
904 /* Destroy fail-open and in-band early, since they touch the classifier. */
905 fail_open_destroy(p->fail_open);
908 in_band_destroy(p->in_band);
910 free(p->extra_in_band_remotes);
912 ofproto_flush_flows(p);
913 classifier_destroy(&p->cls);
915 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
917 ofconn_destroy(ofconn);
919 hmap_destroy(&p->controllers);
922 netdev_monitor_destroy(p->netdev_monitor);
923 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
926 shash_destroy(&p->port_by_name);
928 switch_status_destroy(p->switch_status);
929 netflow_destroy(p->netflow);
930 ofproto_sflow_destroy(p->sflow);
932 for (i = 0; i < p->n_listeners; i++) {
933 pvconn_close(p->listeners[i]);
937 for (i = 0; i < p->n_snoops; i++) {
938 pvconn_close(p->snoops[i]);
942 mac_learning_destroy(p->ml);
947 free(p->serial_desc);
950 port_array_destroy(&p->ports);
956 ofproto_run(struct ofproto *p)
958 int error = ofproto_run1(p);
960 error = ofproto_run2(p, false);
966 process_port_change(struct ofproto *ofproto, int error, char *devname)
968 if (error == ENOBUFS) {
969 reinit_ports(ofproto);
971 update_port(ofproto, devname);
976 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
977 * means that 'ofconn' is more interesting for monitoring than a lower return
980 snoop_preference(const struct ofconn *ofconn)
982 switch (ofconn->role) {
990 /* Shouldn't happen. */
995 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
996 * Connects this vconn to a controller. */
998 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1000 struct ofconn *ofconn, *best;
1002 /* Pick a controller for monitoring. */
1004 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1005 if (ofconn->type == OFCONN_CONTROLLER
1006 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1012 rconn_add_monitor(best->rconn, vconn);
1014 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1020 ofproto_run1(struct ofproto *p)
1022 struct ofconn *ofconn, *next_ofconn;
1027 if (shash_is_empty(&p->port_by_name)) {
1031 for (i = 0; i < 50; i++) {
1035 error = dpif_recv(p->dpif, &buf);
1037 if (error == ENODEV) {
1038 /* Someone destroyed the datapath behind our back. The caller
1039 * better destroy us and give up, because we're just going to
1040 * spin from here on out. */
1041 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1042 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1043 dpif_name(p->dpif));
1049 handle_odp_msg(p, buf);
1052 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1053 process_port_change(p, error, devname);
1055 while ((error = netdev_monitor_poll(p->netdev_monitor,
1056 &devname)) != EAGAIN) {
1057 process_port_change(p, error, devname);
1061 if (time_msec() >= p->next_in_band_update) {
1062 update_in_band_remotes(p);
1064 in_band_run(p->in_band);
1067 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1069 ofconn_run(ofconn, p);
1072 /* Fail-open maintenance. Do this after processing the ofconns since
1073 * fail-open checks the status of the controller rconn. */
1075 fail_open_run(p->fail_open);
1078 for (i = 0; i < p->n_listeners; i++) {
1079 struct vconn *vconn;
1082 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1084 struct rconn *rconn;
1087 rconn = rconn_create(60, 0);
1088 name = ofconn_make_name(p, vconn_get_name(vconn));
1089 rconn_connect_unreliably(rconn, vconn, name);
1092 ofconn_create(p, rconn, OFCONN_TRANSIENT);
1093 } else if (retval != EAGAIN) {
1094 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1098 for (i = 0; i < p->n_snoops; i++) {
1099 struct vconn *vconn;
1102 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1104 add_snooper(p, vconn);
1105 } else if (retval != EAGAIN) {
1106 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1110 if (time_msec() >= p->next_expiration) {
1111 COVERAGE_INC(ofproto_expiration);
1112 p->next_expiration = time_msec() + 1000;
1115 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1117 /* Let the hook know that we're at a stable point: all outstanding data
1118 * in existing flows has been accounted to the account_cb. Thus, the
1119 * hook can now reasonably do operations that depend on having accurate
1120 * flow volume accounting (currently, that's just bond rebalancing). */
1121 if (p->ofhooks->account_checkpoint_cb) {
1122 p->ofhooks->account_checkpoint_cb(p->aux);
1127 netflow_run(p->netflow);
1130 ofproto_sflow_run(p->sflow);
1136 struct revalidate_cbdata {
1137 struct ofproto *ofproto;
1138 bool revalidate_all; /* Revalidate all exact-match rules? */
1139 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1140 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1144 ofproto_run2(struct ofproto *p, bool revalidate_all)
1146 if (p->need_revalidate || revalidate_all
1147 || !tag_set_is_empty(&p->revalidate_set)) {
1148 struct revalidate_cbdata cbdata;
1150 cbdata.revalidate_all = revalidate_all;
1151 cbdata.revalidate_subrules = p->need_revalidate;
1152 cbdata.revalidate_set = p->revalidate_set;
1153 tag_set_init(&p->revalidate_set);
1154 COVERAGE_INC(ofproto_revalidate);
1155 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1156 p->need_revalidate = false;
1163 ofproto_wait(struct ofproto *p)
1165 struct ofconn *ofconn;
1168 dpif_recv_wait(p->dpif);
1169 dpif_port_poll_wait(p->dpif);
1170 netdev_monitor_poll_wait(p->netdev_monitor);
1171 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1172 ofconn_wait(ofconn);
1175 poll_timer_wait_until(p->next_in_band_update);
1176 in_band_wait(p->in_band);
1179 fail_open_wait(p->fail_open);
1182 ofproto_sflow_wait(p->sflow);
1184 if (!tag_set_is_empty(&p->revalidate_set)) {
1185 poll_immediate_wake();
1187 if (p->need_revalidate) {
1188 /* Shouldn't happen, but if it does just go around again. */
1189 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1190 poll_immediate_wake();
1191 } else if (p->next_expiration != LLONG_MAX) {
1192 poll_timer_wait_until(p->next_expiration);
1194 for (i = 0; i < p->n_listeners; i++) {
1195 pvconn_wait(p->listeners[i]);
1197 for (i = 0; i < p->n_snoops; i++) {
1198 pvconn_wait(p->snoops[i]);
1203 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1205 tag_set_add(&ofproto->revalidate_set, tag);
1209 ofproto_get_revalidate_set(struct ofproto *ofproto)
1211 return &ofproto->revalidate_set;
1215 ofproto_is_alive(const struct ofproto *p)
1217 return !hmap_is_empty(&p->controllers);
1221 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1222 const union ofp_action *actions, size_t n_actions,
1223 const struct ofpbuf *packet)
1225 struct odp_actions odp_actions;
1228 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1234 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1236 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1237 odp_actions.n_actions, packet);
1242 ofproto_add_flow(struct ofproto *p,
1243 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1244 const union ofp_action *actions, size_t n_actions,
1248 rule = rule_create(p, NULL, actions, n_actions,
1249 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1251 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1252 rule_insert(p, rule, NULL, 0);
1256 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1257 uint32_t wildcards, unsigned int priority)
1261 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1265 rule_remove(ofproto, rule);
1270 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1272 struct rule *rule = rule_from_cls_rule(rule_);
1273 struct ofproto *ofproto = ofproto_;
1275 /* Mark the flow as not installed, even though it might really be
1276 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1277 * There is no point in uninstalling it individually since we are about to
1278 * blow away all the flows with dpif_flow_flush(). */
1279 rule->installed = false;
1281 rule_remove(ofproto, rule);
1285 ofproto_flush_flows(struct ofproto *ofproto)
1287 COVERAGE_INC(ofproto_flush);
1288 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1289 dpif_flow_flush(ofproto->dpif);
1290 if (ofproto->in_band) {
1291 in_band_flushed(ofproto->in_band);
1293 if (ofproto->fail_open) {
1294 fail_open_flushed(ofproto->fail_open);
1299 reinit_ports(struct ofproto *p)
1301 struct svec devnames;
1302 struct ofport *ofport;
1303 unsigned int port_no;
1304 struct odp_port *odp_ports;
1308 svec_init(&devnames);
1309 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1310 svec_add (&devnames, (char *) ofport->opp.name);
1312 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1313 for (i = 0; i < n_odp_ports; i++) {
1314 svec_add (&devnames, odp_ports[i].devname);
1318 svec_sort_unique(&devnames);
1319 for (i = 0; i < devnames.n; i++) {
1320 update_port(p, devnames.names[i]);
1322 svec_destroy(&devnames);
1326 refresh_port_group(struct ofproto *p, unsigned int group)
1330 struct ofport *port;
1331 unsigned int port_no;
1333 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1335 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1337 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1338 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1339 ports[n_ports++] = port_no;
1342 dpif_port_group_set(p->dpif, group, ports, n_ports);
1349 refresh_port_groups(struct ofproto *p)
1351 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1352 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1354 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1358 static struct ofport *
1359 make_ofport(const struct odp_port *odp_port)
1361 struct netdev_options netdev_options;
1362 enum netdev_flags flags;
1363 struct ofport *ofport;
1364 struct netdev *netdev;
1368 memset(&netdev_options, 0, sizeof netdev_options);
1369 netdev_options.name = odp_port->devname;
1370 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1372 error = netdev_open(&netdev_options, &netdev);
1374 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1375 "cannot be opened (%s)",
1376 odp_port->devname, odp_port->port,
1377 odp_port->devname, strerror(error));
1381 ofport = xmalloc(sizeof *ofport);
1382 ofport->netdev = netdev;
1383 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1384 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1385 memcpy(ofport->opp.name, odp_port->devname,
1386 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1387 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1389 netdev_get_flags(netdev, &flags);
1390 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1392 netdev_get_carrier(netdev, &carrier);
1393 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1395 netdev_get_features(netdev,
1396 &ofport->opp.curr, &ofport->opp.advertised,
1397 &ofport->opp.supported, &ofport->opp.peer);
1402 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1404 if (port_array_get(&p->ports, odp_port->port)) {
1405 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1408 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1409 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1418 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1420 const struct ofp_phy_port *a = &a_->opp;
1421 const struct ofp_phy_port *b = &b_->opp;
1423 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1424 return (a->port_no == b->port_no
1425 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1426 && !strcmp((char *) a->name, (char *) b->name)
1427 && a->state == b->state
1428 && a->config == b->config
1429 && a->curr == b->curr
1430 && a->advertised == b->advertised
1431 && a->supported == b->supported
1432 && a->peer == b->peer);
1436 send_port_status(struct ofproto *p, const struct ofport *ofport,
1439 /* XXX Should limit the number of queued port status change messages. */
1440 struct ofconn *ofconn;
1441 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1442 struct ofp_port_status *ops;
1445 if (!ofconn_receives_async_msgs(ofconn)) {
1449 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1450 ops->reason = reason;
1451 ops->desc = ofport->opp;
1452 hton_ofp_phy_port(&ops->desc);
1453 queue_tx(b, ofconn, NULL);
1455 if (p->ofhooks->port_changed_cb) {
1456 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1461 ofport_install(struct ofproto *p, struct ofport *ofport)
1463 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1464 const char *netdev_name = (const char *) ofport->opp.name;
1466 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1467 port_array_set(&p->ports, odp_port, ofport);
1468 shash_add(&p->port_by_name, netdev_name, ofport);
1470 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1475 ofport_remove(struct ofproto *p, struct ofport *ofport)
1477 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1479 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1480 port_array_set(&p->ports, odp_port, NULL);
1481 shash_delete(&p->port_by_name,
1482 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1484 ofproto_sflow_del_port(p->sflow, odp_port);
1489 ofport_free(struct ofport *ofport)
1492 netdev_close(ofport->netdev);
1498 update_port(struct ofproto *p, const char *devname)
1500 struct odp_port odp_port;
1501 struct ofport *old_ofport;
1502 struct ofport *new_ofport;
1505 COVERAGE_INC(ofproto_update_port);
1507 /* Query the datapath for port information. */
1508 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1510 /* Find the old ofport. */
1511 old_ofport = shash_find_data(&p->port_by_name, devname);
1514 /* There's no port named 'devname' but there might be a port with
1515 * the same port number. This could happen if a port is deleted
1516 * and then a new one added in its place very quickly, or if a port
1517 * is renamed. In the former case we want to send an OFPPR_DELETE
1518 * and an OFPPR_ADD, and in the latter case we want to send a
1519 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1520 * the old port's ifindex against the new port, or perhaps less
1521 * reliably but more portably by comparing the old port's MAC
1522 * against the new port's MAC. However, this code isn't that smart
1523 * and always sends an OFPPR_MODIFY (XXX). */
1524 old_ofport = port_array_get(&p->ports, odp_port.port);
1526 } else if (error != ENOENT && error != ENODEV) {
1527 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1528 "%s", strerror(error));
1532 /* Create a new ofport. */
1533 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1535 /* Eliminate a few pathological cases. */
1536 if (!old_ofport && !new_ofport) {
1538 } else if (old_ofport && new_ofport) {
1539 /* Most of the 'config' bits are OpenFlow soft state, but
1540 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1541 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1542 * leaves the other bits 0.) */
1543 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1545 if (ofport_equal(old_ofport, new_ofport)) {
1546 /* False alarm--no change. */
1547 ofport_free(new_ofport);
1552 /* Now deal with the normal cases. */
1554 ofport_remove(p, old_ofport);
1557 ofport_install(p, new_ofport);
1559 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1560 (!old_ofport ? OFPPR_ADD
1561 : !new_ofport ? OFPPR_DELETE
1563 ofport_free(old_ofport);
1565 /* Update port groups. */
1566 refresh_port_groups(p);
1570 init_ports(struct ofproto *p)
1572 struct odp_port *ports;
1577 error = dpif_port_list(p->dpif, &ports, &n_ports);
1582 for (i = 0; i < n_ports; i++) {
1583 const struct odp_port *odp_port = &ports[i];
1584 if (!ofport_conflicts(p, odp_port)) {
1585 struct ofport *ofport = make_ofport(odp_port);
1587 ofport_install(p, ofport);
1592 refresh_port_groups(p);
1596 static struct ofconn *
1597 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1599 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1600 ofconn->ofproto = p;
1601 list_push_back(&p->all_conns, &ofconn->node);
1602 ofconn->rconn = rconn;
1603 ofconn->type = type;
1604 ofconn->role = NX_ROLE_OTHER;
1605 ofconn->packet_in_counter = rconn_packet_counter_create ();
1606 ofconn->pktbuf = NULL;
1607 ofconn->miss_send_len = 0;
1608 ofconn->reply_counter = rconn_packet_counter_create ();
1613 ofconn_destroy(struct ofconn *ofconn)
1615 if (ofconn->type == OFCONN_CONTROLLER) {
1616 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1618 discovery_destroy(ofconn->discovery);
1620 list_remove(&ofconn->node);
1621 switch_status_unregister(ofconn->ss);
1622 rconn_destroy(ofconn->rconn);
1623 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1624 rconn_packet_counter_destroy(ofconn->reply_counter);
1625 pktbuf_destroy(ofconn->pktbuf);
1630 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1635 if (ofconn->discovery) {
1636 char *controller_name;
1637 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1638 discovery_question_connectivity(ofconn->discovery);
1640 if (discovery_run(ofconn->discovery, &controller_name)) {
1641 if (controller_name) {
1642 char *ofconn_name = ofconn_make_name(p, controller_name);
1643 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1646 rconn_disconnect(ofconn->rconn);
1651 for (i = 0; i < N_SCHEDULERS; i++) {
1652 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1655 rconn_run(ofconn->rconn);
1657 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1658 /* Limit the number of iterations to prevent other tasks from
1660 for (iteration = 0; iteration < 50; iteration++) {
1661 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1666 fail_open_maybe_recover(p->fail_open);
1668 handle_openflow(ofconn, p, of_msg);
1669 ofpbuf_delete(of_msg);
1673 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1674 ofconn_destroy(ofconn);
1679 ofconn_wait(struct ofconn *ofconn)
1683 if (ofconn->discovery) {
1684 discovery_wait(ofconn->discovery);
1686 for (i = 0; i < N_SCHEDULERS; i++) {
1687 pinsched_wait(ofconn->schedulers[i]);
1689 rconn_run_wait(ofconn->rconn);
1690 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1691 rconn_recv_wait(ofconn->rconn);
1693 COVERAGE_INC(ofproto_ofconn_stuck);
1697 /* Returns true if 'ofconn' should receive asynchronous messages. */
1699 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1701 if (ofconn->type == OFCONN_CONTROLLER) {
1702 /* Ordinary controllers always get asynchronous messages unless they
1703 * have configured themselves as "slaves". */
1704 return ofconn->role != NX_ROLE_SLAVE;
1706 /* Transient connections don't get asynchronous messages unless they
1707 * have explicitly asked for them by setting a nonzero miss send
1709 return ofconn->miss_send_len > 0;
1713 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1714 * and 'target', suitable for use in log messages for identifying the
1717 * The name is dynamically allocated. The caller should free it (with free())
1718 * when it is no longer needed. */
1720 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1722 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1725 /* Caller is responsible for initializing the 'cr' member of the returned
1727 static struct rule *
1728 rule_create(struct ofproto *ofproto, struct rule *super,
1729 const union ofp_action *actions, size_t n_actions,
1730 uint16_t idle_timeout, uint16_t hard_timeout,
1731 uint64_t flow_cookie, bool send_flow_removed)
1733 struct rule *rule = xzalloc(sizeof *rule);
1734 rule->idle_timeout = idle_timeout;
1735 rule->hard_timeout = hard_timeout;
1736 rule->flow_cookie = flow_cookie;
1737 rule->used = rule->created = time_msec();
1738 rule->send_flow_removed = send_flow_removed;
1739 rule->super = super;
1741 list_push_back(&super->list, &rule->list);
1743 list_init(&rule->list);
1745 rule->n_actions = n_actions;
1746 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1747 netflow_flow_clear(&rule->nf_flow);
1748 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1753 static struct rule *
1754 rule_from_cls_rule(const struct cls_rule *cls_rule)
1756 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1760 rule_free(struct rule *rule)
1762 free(rule->actions);
1763 free(rule->odp_actions);
1767 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1768 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1769 * through all of its subrules and revalidates them, destroying any that no
1770 * longer has a super-rule (which is probably all of them).
1772 * Before calling this function, the caller must make have removed 'rule' from
1773 * the classifier. If 'rule' is an exact-match rule, the caller is also
1774 * responsible for ensuring that it has been uninstalled from the datapath. */
1776 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1779 struct rule *subrule, *next;
1780 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1781 revalidate_rule(ofproto, subrule);
1784 list_remove(&rule->list);
1790 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1792 const union ofp_action *oa;
1793 struct actions_iterator i;
1795 if (out_port == htons(OFPP_NONE)) {
1798 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1799 oa = actions_next(&i)) {
1800 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1807 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1808 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1810 * The flow that 'packet' actually contains does not need to actually match
1811 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1812 * the packet and byte counters for 'rule' will be credited for the packet sent
1813 * out whether or not the packet actually matches 'rule'.
1815 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1816 * the caller must already have accurately composed ODP actions for it given
1817 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1818 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1819 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1820 * actions and apply them to 'packet'. */
1822 rule_execute(struct ofproto *ofproto, struct rule *rule,
1823 struct ofpbuf *packet, const flow_t *flow)
1825 const union odp_action *actions;
1827 struct odp_actions a;
1829 /* Grab or compose the ODP actions.
1831 * The special case for an exact-match 'rule' where 'flow' is not the
1832 * rule's flow is important to avoid, e.g., sending a packet out its input
1833 * port simply because the ODP actions were composed for the wrong
1835 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1836 struct rule *super = rule->super ? rule->super : rule;
1837 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1838 packet, &a, NULL, 0, NULL)) {
1841 actions = a.actions;
1842 n_actions = a.n_actions;
1844 actions = rule->odp_actions;
1845 n_actions = rule->n_odp_actions;
1848 /* Execute the ODP actions. */
1849 if (!dpif_execute(ofproto->dpif, flow->in_port,
1850 actions, n_actions, packet)) {
1851 struct odp_flow_stats stats;
1852 flow_extract_stats(flow, packet, &stats);
1853 update_stats(ofproto, rule, &stats);
1854 rule->used = time_msec();
1855 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1860 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1863 struct rule *displaced_rule;
1865 /* Insert the rule in the classifier. */
1866 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1867 if (!rule->cr.wc.wildcards) {
1868 rule_make_actions(p, rule, packet);
1871 /* Send the packet and credit it to the rule. */
1874 flow_extract(packet, 0, in_port, &flow);
1875 rule_execute(p, rule, packet, &flow);
1878 /* Install the rule in the datapath only after sending the packet, to
1879 * avoid packet reordering. */
1880 if (rule->cr.wc.wildcards) {
1881 COVERAGE_INC(ofproto_add_wc_flow);
1882 p->need_revalidate = true;
1884 rule_install(p, rule, displaced_rule);
1887 /* Free the rule that was displaced, if any. */
1888 if (displaced_rule) {
1889 rule_destroy(p, displaced_rule);
1893 static struct rule *
1894 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1897 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1898 rule->idle_timeout, rule->hard_timeout,
1900 COVERAGE_INC(ofproto_subrule_create);
1901 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1902 : rule->cr.priority), &subrule->cr);
1903 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1909 rule_remove(struct ofproto *ofproto, struct rule *rule)
1911 if (rule->cr.wc.wildcards) {
1912 COVERAGE_INC(ofproto_del_wc_flow);
1913 ofproto->need_revalidate = true;
1915 rule_uninstall(ofproto, rule);
1917 classifier_remove(&ofproto->cls, &rule->cr);
1918 rule_destroy(ofproto, rule);
1921 /* Returns true if the actions changed, false otherwise. */
1923 rule_make_actions(struct ofproto *p, struct rule *rule,
1924 const struct ofpbuf *packet)
1926 const struct rule *super;
1927 struct odp_actions a;
1930 assert(!rule->cr.wc.wildcards);
1932 super = rule->super ? rule->super : rule;
1934 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1935 packet, &a, &rule->tags, &rule->may_install,
1936 &rule->nf_flow.output_iface);
1938 actions_len = a.n_actions * sizeof *a.actions;
1939 if (rule->n_odp_actions != a.n_actions
1940 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1941 COVERAGE_INC(ofproto_odp_unchanged);
1942 free(rule->odp_actions);
1943 rule->n_odp_actions = a.n_actions;
1944 rule->odp_actions = xmemdup(a.actions, actions_len);
1952 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1953 struct odp_flow_put *put)
1955 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1956 put->flow.key = rule->cr.flow;
1957 put->flow.actions = rule->odp_actions;
1958 put->flow.n_actions = rule->n_odp_actions;
1959 put->flow.flags = 0;
1961 return dpif_flow_put(ofproto->dpif, put);
1965 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1967 assert(!rule->cr.wc.wildcards);
1969 if (rule->may_install) {
1970 struct odp_flow_put put;
1971 if (!do_put_flow(p, rule,
1972 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1974 rule->installed = true;
1975 if (displaced_rule) {
1976 update_stats(p, displaced_rule, &put.flow.stats);
1977 rule_post_uninstall(p, displaced_rule);
1980 } else if (displaced_rule) {
1981 rule_uninstall(p, displaced_rule);
1986 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1988 if (rule->installed) {
1989 struct odp_flow_put put;
1990 COVERAGE_INC(ofproto_dp_missed);
1991 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1993 rule_install(ofproto, rule, NULL);
1998 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2000 bool actions_changed;
2001 uint16_t new_out_iface, old_out_iface;
2003 old_out_iface = rule->nf_flow.output_iface;
2004 actions_changed = rule_make_actions(ofproto, rule, NULL);
2006 if (rule->may_install) {
2007 if (rule->installed) {
2008 if (actions_changed) {
2009 struct odp_flow_put put;
2010 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2011 | ODPPF_ZERO_STATS, &put);
2012 update_stats(ofproto, rule, &put.flow.stats);
2014 /* Temporarily set the old output iface so that NetFlow
2015 * messages have the correct output interface for the old
2017 new_out_iface = rule->nf_flow.output_iface;
2018 rule->nf_flow.output_iface = old_out_iface;
2019 rule_post_uninstall(ofproto, rule);
2020 rule->nf_flow.output_iface = new_out_iface;
2023 rule_install(ofproto, rule, NULL);
2026 rule_uninstall(ofproto, rule);
2031 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2033 uint64_t total_bytes = rule->byte_count + extra_bytes;
2035 if (ofproto->ofhooks->account_flow_cb
2036 && total_bytes > rule->accounted_bytes)
2038 ofproto->ofhooks->account_flow_cb(
2039 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2040 total_bytes - rule->accounted_bytes, ofproto->aux);
2041 rule->accounted_bytes = total_bytes;
2046 rule_uninstall(struct ofproto *p, struct rule *rule)
2048 assert(!rule->cr.wc.wildcards);
2049 if (rule->installed) {
2050 struct odp_flow odp_flow;
2052 odp_flow.key = rule->cr.flow;
2053 odp_flow.actions = NULL;
2054 odp_flow.n_actions = 0;
2056 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2057 update_stats(p, rule, &odp_flow.stats);
2059 rule->installed = false;
2061 rule_post_uninstall(p, rule);
2066 is_controller_rule(struct rule *rule)
2068 /* If the only action is send to the controller then don't report
2069 * NetFlow expiration messages since it is just part of the control
2070 * logic for the network and not real traffic. */
2072 if (rule && rule->super) {
2073 struct rule *super = rule->super;
2075 return super->n_actions == 1 &&
2076 super->actions[0].type == htons(OFPAT_OUTPUT) &&
2077 super->actions[0].output.port == htons(OFPP_CONTROLLER);
2084 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2086 struct rule *super = rule->super;
2088 rule_account(ofproto, rule, 0);
2090 if (ofproto->netflow && !is_controller_rule(rule)) {
2091 struct ofexpired expired;
2092 expired.flow = rule->cr.flow;
2093 expired.packet_count = rule->packet_count;
2094 expired.byte_count = rule->byte_count;
2095 expired.used = rule->used;
2096 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2099 super->packet_count += rule->packet_count;
2100 super->byte_count += rule->byte_count;
2102 /* Reset counters to prevent double counting if the rule ever gets
2104 rule->packet_count = 0;
2105 rule->byte_count = 0;
2106 rule->accounted_bytes = 0;
2108 netflow_flow_clear(&rule->nf_flow);
2113 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2114 struct rconn_packet_counter *counter)
2116 update_openflow_length(msg);
2117 if (rconn_send(ofconn->rconn, msg, counter)) {
2123 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2124 int error, const void *data, size_t len)
2127 struct ofp_error_msg *oem;
2129 if (!(error >> 16)) {
2130 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2135 COVERAGE_INC(ofproto_error);
2136 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2137 oh ? oh->xid : 0, &buf);
2138 oem->type = htons((unsigned int) error >> 16);
2139 oem->code = htons(error & 0xffff);
2140 memcpy(oem->data, data, len);
2141 queue_tx(buf, ofconn, ofconn->reply_counter);
2145 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2148 size_t oh_length = ntohs(oh->length);
2149 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2153 hton_ofp_phy_port(struct ofp_phy_port *opp)
2155 opp->port_no = htons(opp->port_no);
2156 opp->config = htonl(opp->config);
2157 opp->state = htonl(opp->state);
2158 opp->curr = htonl(opp->curr);
2159 opp->advertised = htonl(opp->advertised);
2160 opp->supported = htonl(opp->supported);
2161 opp->peer = htonl(opp->peer);
2165 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2167 struct ofp_header *rq = oh;
2168 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2173 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2174 struct ofp_header *oh)
2176 struct ofp_switch_features *osf;
2178 unsigned int port_no;
2179 struct ofport *port;
2181 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2182 osf->datapath_id = htonll(p->datapath_id);
2183 osf->n_buffers = htonl(pktbuf_capacity());
2185 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2186 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2187 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2188 (1u << OFPAT_SET_VLAN_VID) |
2189 (1u << OFPAT_SET_VLAN_PCP) |
2190 (1u << OFPAT_STRIP_VLAN) |
2191 (1u << OFPAT_SET_DL_SRC) |
2192 (1u << OFPAT_SET_DL_DST) |
2193 (1u << OFPAT_SET_NW_SRC) |
2194 (1u << OFPAT_SET_NW_DST) |
2195 (1u << OFPAT_SET_NW_TOS) |
2196 (1u << OFPAT_SET_TP_SRC) |
2197 (1u << OFPAT_SET_TP_DST));
2199 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2200 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2203 queue_tx(buf, ofconn, ofconn->reply_counter);
2208 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2209 struct ofp_header *oh)
2212 struct ofp_switch_config *osc;
2216 /* Figure out flags. */
2217 dpif_get_drop_frags(p->dpif, &drop_frags);
2218 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2221 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2222 osc->flags = htons(flags);
2223 osc->miss_send_len = htons(ofconn->miss_send_len);
2224 queue_tx(buf, ofconn, ofconn->reply_counter);
2230 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2231 struct ofp_switch_config *osc)
2236 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2240 flags = ntohs(osc->flags);
2242 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2243 switch (flags & OFPC_FRAG_MASK) {
2244 case OFPC_FRAG_NORMAL:
2245 dpif_set_drop_frags(p->dpif, false);
2247 case OFPC_FRAG_DROP:
2248 dpif_set_drop_frags(p->dpif, true);
2251 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2257 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2263 add_output_group_action(struct odp_actions *actions, uint16_t group,
2264 uint16_t *nf_output_iface)
2266 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2268 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2269 *nf_output_iface = NF_OUT_FLOOD;
2274 add_controller_action(struct odp_actions *actions,
2275 const struct ofp_action_output *oao)
2277 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2278 a->controller.arg = ntohs(oao->max_len);
2281 struct action_xlate_ctx {
2283 flow_t flow; /* Flow to which these actions correspond. */
2284 int recurse; /* Recursion level, via xlate_table_action. */
2285 struct ofproto *ofproto;
2286 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2287 * null pointer if we are revalidating
2288 * without a packet to refer to. */
2291 struct odp_actions *out; /* Datapath actions. */
2292 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2293 bool may_set_up_flow; /* True ordinarily; false if the actions must
2294 * be reassessed for every packet. */
2295 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2298 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2299 struct action_xlate_ctx *ctx);
2302 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2304 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2307 if (ofport->opp.config & OFPPC_NO_FWD) {
2308 /* Forwarding disabled on port. */
2313 * We don't have an ofport record for this port, but it doesn't hurt to
2314 * allow forwarding to it anyhow. Maybe such a port will appear later
2315 * and we're pre-populating the flow table.
2319 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2320 ctx->nf_output_iface = port;
2323 static struct rule *
2324 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2327 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2329 /* The rule we found might not be valid, since we could be in need of
2330 * revalidation. If it is not valid, don't return it. */
2333 && ofproto->need_revalidate
2334 && !revalidate_rule(ofproto, rule)) {
2335 COVERAGE_INC(ofproto_invalidated);
2343 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2345 if (!ctx->recurse) {
2346 uint16_t old_in_port;
2349 /* Look up a flow with 'in_port' as the input port. Then restore the
2350 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2351 * have surprising behavior). */
2352 old_in_port = ctx->flow.in_port;
2353 ctx->flow.in_port = in_port;
2354 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2355 ctx->flow.in_port = old_in_port;
2363 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2370 xlate_output_action(struct action_xlate_ctx *ctx,
2371 const struct ofp_action_output *oao)
2374 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2376 ctx->nf_output_iface = NF_OUT_DROP;
2378 switch (ntohs(oao->port)) {
2380 add_output_action(ctx, ctx->flow.in_port);
2383 xlate_table_action(ctx, ctx->flow.in_port);
2386 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2387 ctx->out, ctx->tags,
2388 &ctx->nf_output_iface,
2389 ctx->ofproto->aux)) {
2390 COVERAGE_INC(ofproto_uninstallable);
2391 ctx->may_set_up_flow = false;
2395 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2396 &ctx->nf_output_iface);
2399 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2401 case OFPP_CONTROLLER:
2402 add_controller_action(ctx->out, oao);
2405 add_output_action(ctx, ODPP_LOCAL);
2408 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2409 if (odp_port != ctx->flow.in_port) {
2410 add_output_action(ctx, odp_port);
2415 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2416 ctx->nf_output_iface = NF_OUT_FLOOD;
2417 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2418 ctx->nf_output_iface = prev_nf_output_iface;
2419 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2420 ctx->nf_output_iface != NF_OUT_FLOOD) {
2421 ctx->nf_output_iface = NF_OUT_MULTI;
2426 xlate_nicira_action(struct action_xlate_ctx *ctx,
2427 const struct nx_action_header *nah)
2429 const struct nx_action_resubmit *nar;
2430 const struct nx_action_set_tunnel *nast;
2431 union odp_action *oa;
2432 int subtype = ntohs(nah->subtype);
2434 assert(nah->vendor == htonl(NX_VENDOR_ID));
2436 case NXAST_RESUBMIT:
2437 nar = (const struct nx_action_resubmit *) nah;
2438 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2441 case NXAST_SET_TUNNEL:
2442 nast = (const struct nx_action_set_tunnel *) nah;
2443 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2444 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2447 /* If you add a new action here that modifies flow data, don't forget to
2448 * update the flow key in ctx->flow in the same key. */
2451 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2457 do_xlate_actions(const union ofp_action *in, size_t n_in,
2458 struct action_xlate_ctx *ctx)
2460 struct actions_iterator iter;
2461 const union ofp_action *ia;
2462 const struct ofport *port;
2464 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2465 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2466 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2467 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2468 /* Drop this flow. */
2472 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2473 uint16_t type = ntohs(ia->type);
2474 union odp_action *oa;
2478 xlate_output_action(ctx, &ia->output);
2481 case OFPAT_SET_VLAN_VID:
2482 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2483 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2486 case OFPAT_SET_VLAN_PCP:
2487 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2488 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2491 case OFPAT_STRIP_VLAN:
2492 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2493 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2494 ctx->flow.dl_vlan_pcp = 0;
2497 case OFPAT_SET_DL_SRC:
2498 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2499 memcpy(oa->dl_addr.dl_addr,
2500 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2501 memcpy(ctx->flow.dl_src,
2502 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2505 case OFPAT_SET_DL_DST:
2506 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2507 memcpy(oa->dl_addr.dl_addr,
2508 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2509 memcpy(ctx->flow.dl_dst,
2510 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2513 case OFPAT_SET_NW_SRC:
2514 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2515 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2518 case OFPAT_SET_NW_DST:
2519 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2520 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2523 case OFPAT_SET_NW_TOS:
2524 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2525 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2528 case OFPAT_SET_TP_SRC:
2529 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2530 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2533 case OFPAT_SET_TP_DST:
2534 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2535 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2539 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2543 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2550 xlate_actions(const union ofp_action *in, size_t n_in,
2551 const flow_t *flow, struct ofproto *ofproto,
2552 const struct ofpbuf *packet,
2553 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2554 uint16_t *nf_output_iface)
2556 tag_type no_tags = 0;
2557 struct action_xlate_ctx ctx;
2558 COVERAGE_INC(ofproto_ofp2odp);
2559 odp_actions_init(out);
2562 ctx.ofproto = ofproto;
2563 ctx.packet = packet;
2565 ctx.tags = tags ? tags : &no_tags;
2566 ctx.may_set_up_flow = true;
2567 ctx.nf_output_iface = NF_OUT_DROP;
2568 do_xlate_actions(in, n_in, &ctx);
2570 /* Check with in-band control to see if we're allowed to set up this
2572 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2573 ctx.may_set_up_flow = false;
2576 if (may_set_up_flow) {
2577 *may_set_up_flow = ctx.may_set_up_flow;
2579 if (nf_output_iface) {
2580 *nf_output_iface = ctx.nf_output_iface;
2582 if (odp_actions_overflow(out)) {
2583 odp_actions_init(out);
2584 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2589 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2590 * error message code (composed with ofp_mkerr()) for the caller to propagate
2591 * upward. Otherwise, returns 0.
2593 * 'oh' is used to make log messages more informative. */
2595 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2597 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2598 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2601 type_name = ofp_message_type_to_string(oh->type);
2602 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2606 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2613 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2614 struct ofp_header *oh)
2616 struct ofp_packet_out *opo;
2617 struct ofpbuf payload, *buffer;
2618 struct odp_actions actions;
2624 error = reject_slave_controller(ofconn, oh);
2629 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2633 opo = (struct ofp_packet_out *) oh;
2635 COVERAGE_INC(ofproto_packet_out);
2636 if (opo->buffer_id != htonl(UINT32_MAX)) {
2637 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2639 if (error || !buffer) {
2647 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2648 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2649 &flow, p, &payload, &actions, NULL, NULL, NULL);
2654 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2656 ofpbuf_delete(buffer);
2662 update_port_config(struct ofproto *p, struct ofport *port,
2663 uint32_t config, uint32_t mask)
2665 mask &= config ^ port->opp.config;
2666 if (mask & OFPPC_PORT_DOWN) {
2667 if (config & OFPPC_PORT_DOWN) {
2668 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2670 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2673 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2674 if (mask & REVALIDATE_BITS) {
2675 COVERAGE_INC(ofproto_costly_flags);
2676 port->opp.config ^= mask & REVALIDATE_BITS;
2677 p->need_revalidate = true;
2679 #undef REVALIDATE_BITS
2680 if (mask & OFPPC_NO_FLOOD) {
2681 port->opp.config ^= OFPPC_NO_FLOOD;
2682 refresh_port_groups(p);
2684 if (mask & OFPPC_NO_PACKET_IN) {
2685 port->opp.config ^= OFPPC_NO_PACKET_IN;
2690 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2691 struct ofp_header *oh)
2693 const struct ofp_port_mod *opm;
2694 struct ofport *port;
2697 error = reject_slave_controller(ofconn, oh);
2701 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2705 opm = (struct ofp_port_mod *) oh;
2707 port = port_array_get(&p->ports,
2708 ofp_port_to_odp_port(ntohs(opm->port_no)));
2710 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2711 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2712 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2714 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2715 if (opm->advertise) {
2716 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2722 static struct ofpbuf *
2723 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2725 struct ofp_stats_reply *osr;
2728 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2729 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2731 osr->flags = htons(0);
2735 static struct ofpbuf *
2736 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2738 return make_stats_reply(request->header.xid, request->type, body_len);
2742 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2744 struct ofpbuf *msg = *msgp;
2745 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2746 if (nbytes + msg->size > UINT16_MAX) {
2747 struct ofp_stats_reply *reply = msg->data;
2748 reply->flags = htons(OFPSF_REPLY_MORE);
2749 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2750 queue_tx(msg, ofconn, ofconn->reply_counter);
2752 return ofpbuf_put_uninit(*msgp, nbytes);
2756 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2757 struct ofp_stats_request *request)
2759 struct ofp_desc_stats *ods;
2762 msg = start_stats_reply(request, sizeof *ods);
2763 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2764 memset(ods, 0, sizeof *ods);
2765 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2766 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2767 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2768 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2769 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2770 queue_tx(msg, ofconn, ofconn->reply_counter);
2776 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2778 struct rule *rule = rule_from_cls_rule(cls_rule);
2779 int *n_subrules = n_subrules_;
2787 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2788 struct ofp_stats_request *request)
2790 struct ofp_table_stats *ots;
2792 struct odp_stats dpstats;
2793 int n_exact, n_subrules, n_wild;
2795 msg = start_stats_reply(request, sizeof *ots * 2);
2797 /* Count rules of various kinds. */
2799 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2800 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2801 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2804 dpif_get_dp_stats(p->dpif, &dpstats);
2805 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2806 memset(ots, 0, sizeof *ots);
2807 ots->table_id = TABLEID_HASH;
2808 strcpy(ots->name, "hash");
2809 ots->wildcards = htonl(0);
2810 ots->max_entries = htonl(dpstats.max_capacity);
2811 ots->active_count = htonl(n_exact);
2812 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2814 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2816 /* Classifier table. */
2817 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2818 memset(ots, 0, sizeof *ots);
2819 ots->table_id = TABLEID_CLASSIFIER;
2820 strcpy(ots->name, "classifier");
2821 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2823 ots->max_entries = htonl(65536);
2824 ots->active_count = htonl(n_wild);
2825 ots->lookup_count = htonll(0); /* XXX */
2826 ots->matched_count = htonll(0); /* XXX */
2828 queue_tx(msg, ofconn, ofconn->reply_counter);
2833 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2834 struct ofpbuf **msgp)
2836 struct netdev_stats stats;
2837 struct ofp_port_stats *ops;
2839 /* Intentionally ignore return value, since errors will set
2840 * 'stats' to all-1s, which is correct for OpenFlow, and
2841 * netdev_get_stats() will log errors. */
2842 netdev_get_stats(port->netdev, &stats);
2844 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2845 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2846 memset(ops->pad, 0, sizeof ops->pad);
2847 ops->rx_packets = htonll(stats.rx_packets);
2848 ops->tx_packets = htonll(stats.tx_packets);
2849 ops->rx_bytes = htonll(stats.rx_bytes);
2850 ops->tx_bytes = htonll(stats.tx_bytes);
2851 ops->rx_dropped = htonll(stats.rx_dropped);
2852 ops->tx_dropped = htonll(stats.tx_dropped);
2853 ops->rx_errors = htonll(stats.rx_errors);
2854 ops->tx_errors = htonll(stats.tx_errors);
2855 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2856 ops->rx_over_err = htonll(stats.rx_over_errors);
2857 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2858 ops->collisions = htonll(stats.collisions);
2862 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2863 struct ofp_stats_request *osr,
2866 struct ofp_port_stats_request *psr;
2867 struct ofp_port_stats *ops;
2869 struct ofport *port;
2870 unsigned int port_no;
2872 if (arg_size != sizeof *psr) {
2873 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2875 psr = (struct ofp_port_stats_request *) osr->body;
2877 msg = start_stats_reply(osr, sizeof *ops * 16);
2878 if (psr->port_no != htons(OFPP_NONE)) {
2879 port = port_array_get(&p->ports,
2880 ofp_port_to_odp_port(ntohs(psr->port_no)));
2882 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2885 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2886 append_port_stat(port, port_no, ofconn, &msg);
2890 queue_tx(msg, ofconn, ofconn->reply_counter);
2894 struct flow_stats_cbdata {
2895 struct ofproto *ofproto;
2896 struct ofconn *ofconn;
2901 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2902 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2903 * returned statistic include statistics for all of 'rule''s subrules. */
2905 query_stats(struct ofproto *p, struct rule *rule,
2906 uint64_t *packet_countp, uint64_t *byte_countp)
2908 uint64_t packet_count, byte_count;
2909 struct rule *subrule;
2910 struct odp_flow *odp_flows;
2913 /* Start from historical data for 'rule' itself that are no longer tracked
2914 * by the datapath. This counts, for example, subrules that have
2916 packet_count = rule->packet_count;
2917 byte_count = rule->byte_count;
2919 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2920 * wildcarded then on all of its subrules.
2922 * Also, add any statistics that are not tracked by the datapath for each
2923 * subrule. This includes, for example, statistics for packets that were
2924 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2926 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2927 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2928 if (rule->cr.wc.wildcards) {
2930 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2931 odp_flows[i++].key = subrule->cr.flow;
2932 packet_count += subrule->packet_count;
2933 byte_count += subrule->byte_count;
2936 odp_flows[0].key = rule->cr.flow;
2939 /* Fetch up-to-date statistics from the datapath and add them in. */
2940 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2942 for (i = 0; i < n_odp_flows; i++) {
2943 struct odp_flow *odp_flow = &odp_flows[i];
2944 packet_count += odp_flow->stats.n_packets;
2945 byte_count += odp_flow->stats.n_bytes;
2950 /* Return the stats to the caller. */
2951 *packet_countp = packet_count;
2952 *byte_countp = byte_count;
2956 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2958 struct rule *rule = rule_from_cls_rule(rule_);
2959 struct flow_stats_cbdata *cbdata = cbdata_;
2960 struct ofp_flow_stats *ofs;
2961 uint64_t packet_count, byte_count;
2962 size_t act_len, len;
2963 long long int tdiff = time_msec() - rule->created;
2964 uint32_t sec = tdiff / 1000;
2965 uint32_t msec = tdiff - (sec * 1000);
2967 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2971 act_len = sizeof *rule->actions * rule->n_actions;
2972 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2974 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2976 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2977 ofs->length = htons(len);
2978 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2980 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2981 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2982 ofs->duration_sec = htonl(sec);
2983 ofs->duration_nsec = htonl(msec * 1000000);
2984 ofs->cookie = rule->flow_cookie;
2985 ofs->priority = htons(rule->cr.priority);
2986 ofs->idle_timeout = htons(rule->idle_timeout);
2987 ofs->hard_timeout = htons(rule->hard_timeout);
2988 memset(ofs->pad2, 0, sizeof ofs->pad2);
2989 ofs->packet_count = htonll(packet_count);
2990 ofs->byte_count = htonll(byte_count);
2991 memcpy(ofs->actions, rule->actions, act_len);
2995 table_id_to_include(uint8_t table_id)
2997 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2998 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2999 : table_id == 0xff ? CLS_INC_ALL
3004 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3005 const struct ofp_stats_request *osr,
3008 struct ofp_flow_stats_request *fsr;
3009 struct flow_stats_cbdata cbdata;
3010 struct cls_rule target;
3012 if (arg_size != sizeof *fsr) {
3013 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3015 fsr = (struct ofp_flow_stats_request *) osr->body;
3017 COVERAGE_INC(ofproto_flows_req);
3019 cbdata.ofconn = ofconn;
3020 cbdata.out_port = fsr->out_port;
3021 cbdata.msg = start_stats_reply(osr, 1024);
3022 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3023 classifier_for_each_match(&p->cls, &target,
3024 table_id_to_include(fsr->table_id),
3025 flow_stats_cb, &cbdata);
3026 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3030 struct flow_stats_ds_cbdata {
3031 struct ofproto *ofproto;
3036 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3038 struct rule *rule = rule_from_cls_rule(rule_);
3039 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3040 struct ds *results = cbdata->results;
3041 struct ofp_match match;
3042 uint64_t packet_count, byte_count;
3043 size_t act_len = sizeof *rule->actions * rule->n_actions;
3045 /* Don't report on subrules. */
3046 if (rule->super != NULL) {
3050 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3051 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3052 cbdata->ofproto->tun_id_from_cookie, &match);
3054 ds_put_format(results, "duration=%llds, ",
3055 (time_msec() - rule->created) / 1000);
3056 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3057 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3058 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3059 ofp_print_match(results, &match, true);
3060 ofp_print_actions(results, &rule->actions->header, act_len);
3061 ds_put_cstr(results, "\n");
3064 /* Adds a pretty-printed description of all flows to 'results', including
3065 * those marked hidden by secchan (e.g., by in-band control). */
3067 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3069 struct ofp_match match;
3070 struct cls_rule target;
3071 struct flow_stats_ds_cbdata cbdata;
3073 memset(&match, 0, sizeof match);
3074 match.wildcards = htonl(OVSFW_ALL);
3077 cbdata.results = results;
3079 cls_rule_from_match(&match, 0, false, 0, &target);
3080 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3081 flow_stats_ds_cb, &cbdata);
3084 struct aggregate_stats_cbdata {
3085 struct ofproto *ofproto;
3087 uint64_t packet_count;
3088 uint64_t byte_count;
3093 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3095 struct rule *rule = rule_from_cls_rule(rule_);
3096 struct aggregate_stats_cbdata *cbdata = cbdata_;
3097 uint64_t packet_count, byte_count;
3099 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3103 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3105 cbdata->packet_count += packet_count;
3106 cbdata->byte_count += byte_count;
3111 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3112 const struct ofp_stats_request *osr,
3115 struct ofp_aggregate_stats_request *asr;
3116 struct ofp_aggregate_stats_reply *reply;
3117 struct aggregate_stats_cbdata cbdata;
3118 struct cls_rule target;
3121 if (arg_size != sizeof *asr) {
3122 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3124 asr = (struct ofp_aggregate_stats_request *) osr->body;
3126 COVERAGE_INC(ofproto_agg_request);
3128 cbdata.out_port = asr->out_port;
3129 cbdata.packet_count = 0;
3130 cbdata.byte_count = 0;
3132 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3133 classifier_for_each_match(&p->cls, &target,
3134 table_id_to_include(asr->table_id),
3135 aggregate_stats_cb, &cbdata);
3137 msg = start_stats_reply(osr, sizeof *reply);
3138 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3139 reply->flow_count = htonl(cbdata.n_flows);
3140 reply->packet_count = htonll(cbdata.packet_count);
3141 reply->byte_count = htonll(cbdata.byte_count);
3142 queue_tx(msg, ofconn, ofconn->reply_counter);
3147 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3148 struct ofp_header *oh)
3150 struct ofp_stats_request *osr;
3154 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3159 osr = (struct ofp_stats_request *) oh;
3161 switch (ntohs(osr->type)) {
3163 return handle_desc_stats_request(p, ofconn, osr);
3166 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3168 case OFPST_AGGREGATE:
3169 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3172 return handle_table_stats_request(p, ofconn, osr);
3175 return handle_port_stats_request(p, ofconn, osr, arg_size);
3178 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3181 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3185 static long long int
3186 msec_from_nsec(uint64_t sec, uint32_t nsec)
3188 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3192 update_time(struct ofproto *ofproto, struct rule *rule,
3193 const struct odp_flow_stats *stats)
3195 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3196 if (used > rule->used) {
3198 if (rule->super && used > rule->super->used) {
3199 rule->super->used = used;
3201 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3206 update_stats(struct ofproto *ofproto, struct rule *rule,
3207 const struct odp_flow_stats *stats)
3209 if (stats->n_packets) {
3210 update_time(ofproto, rule, stats);
3211 rule->packet_count += stats->n_packets;
3212 rule->byte_count += stats->n_bytes;
3213 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3218 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3219 * in which no matching flow already exists in the flow table.
3221 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3222 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3223 * code as encoded by ofp_mkerr() on failure.
3225 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3228 add_flow(struct ofproto *p, struct ofconn *ofconn,
3229 const struct ofp_flow_mod *ofm, size_t n_actions)
3231 struct ofpbuf *packet;
3236 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3240 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3242 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3243 ntohs(ofm->priority))) {
3244 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3248 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3249 n_actions, ntohs(ofm->idle_timeout),
3250 ntohs(ofm->hard_timeout), ofm->cookie,
3251 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3252 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3253 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3256 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3257 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3261 in_port = UINT16_MAX;
3264 rule_insert(p, rule, packet, in_port);
3265 ofpbuf_delete(packet);
3269 static struct rule *
3270 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3275 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3277 return rule_from_cls_rule(classifier_find_rule_exactly(
3278 &p->cls, &flow, wildcards,
3279 ntohs(ofm->priority)));
3283 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3284 struct rule *rule, const struct ofp_flow_mod *ofm)
3286 struct ofpbuf *packet;
3291 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3295 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3301 flow_extract(packet, 0, in_port, &flow);
3302 rule_execute(ofproto, rule, packet, &flow);
3303 ofpbuf_delete(packet);
3308 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3310 struct modify_flows_cbdata {
3311 struct ofproto *ofproto;
3312 const struct ofp_flow_mod *ofm;
3317 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3318 size_t n_actions, struct rule *);
3319 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3321 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3322 * encoded by ofp_mkerr() on failure.
3324 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3327 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3328 const struct ofp_flow_mod *ofm, size_t n_actions)
3330 struct modify_flows_cbdata cbdata;
3331 struct cls_rule target;
3335 cbdata.n_actions = n_actions;
3336 cbdata.match = NULL;
3338 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3341 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3342 modify_flows_cb, &cbdata);
3344 /* This credits the packet to whichever flow happened to happened to
3345 * match last. That's weird. Maybe we should do a lookup for the
3346 * flow that actually matches the packet? Who knows. */
3347 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3350 return add_flow(p, ofconn, ofm, n_actions);
3354 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3355 * code as encoded by ofp_mkerr() on failure.
3357 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3360 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3361 struct ofp_flow_mod *ofm, size_t n_actions)
3363 struct rule *rule = find_flow_strict(p, ofm);
3364 if (rule && !rule_is_hidden(rule)) {
3365 modify_flow(p, ofm, n_actions, rule);
3366 return send_buffered_packet(p, ofconn, rule, ofm);
3368 return add_flow(p, ofconn, ofm, n_actions);
3372 /* Callback for modify_flows_loose(). */
3374 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3376 struct rule *rule = rule_from_cls_rule(rule_);
3377 struct modify_flows_cbdata *cbdata = cbdata_;
3379 if (!rule_is_hidden(rule)) {
3380 cbdata->match = rule;
3381 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3385 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3386 * been identified as a flow in 'p''s flow table to be modified, by changing
3387 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3388 * ofp_action[] structures). */
3390 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3391 size_t n_actions, struct rule *rule)
3393 size_t actions_len = n_actions * sizeof *rule->actions;
3395 rule->flow_cookie = ofm->cookie;
3397 /* If the actions are the same, do nothing. */
3398 if (n_actions == rule->n_actions
3399 && !memcmp(ofm->actions, rule->actions, actions_len))
3404 /* Replace actions. */
3405 free(rule->actions);
3406 rule->actions = xmemdup(ofm->actions, actions_len);
3407 rule->n_actions = n_actions;
3409 /* Make sure that the datapath gets updated properly. */
3410 if (rule->cr.wc.wildcards) {
3411 COVERAGE_INC(ofproto_mod_wc_flow);
3412 p->need_revalidate = true;
3414 rule_update_actions(p, rule);
3420 /* OFPFC_DELETE implementation. */
3422 struct delete_flows_cbdata {
3423 struct ofproto *ofproto;
3427 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3428 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3430 /* Implements OFPFC_DELETE. */
3432 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3434 struct delete_flows_cbdata cbdata;
3435 struct cls_rule target;
3438 cbdata.out_port = ofm->out_port;
3440 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3443 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3444 delete_flows_cb, &cbdata);
3447 /* Implements OFPFC_DELETE_STRICT. */
3449 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3451 struct rule *rule = find_flow_strict(p, ofm);
3453 delete_flow(p, rule, ofm->out_port);
3457 /* Callback for delete_flows_loose(). */
3459 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3461 struct rule *rule = rule_from_cls_rule(rule_);
3462 struct delete_flows_cbdata *cbdata = cbdata_;
3464 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3467 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3468 * been identified as a flow to delete from 'p''s flow table, by deleting the
3469 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3472 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3473 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3474 * specified 'out_port'. */
3476 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3478 if (rule_is_hidden(rule)) {
3482 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3486 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3487 rule_remove(p, rule);
3491 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3492 struct ofp_flow_mod *ofm)
3497 error = reject_slave_controller(ofconn, &ofm->header);
3501 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3502 sizeof *ofm->actions, &n_actions);
3507 /* We do not support the emergency flow cache. It will hopefully
3508 * get dropped from OpenFlow in the near future. */
3509 if (ofm->flags & htons(OFPFF_EMERG)) {
3510 /* There isn't a good fit for an error code, so just state that the
3511 * flow table is full. */
3512 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3515 normalize_match(&ofm->match);
3516 if (!ofm->match.wildcards) {
3517 ofm->priority = htons(UINT16_MAX);
3520 error = validate_actions((const union ofp_action *) ofm->actions,
3521 n_actions, p->max_ports);
3526 switch (ntohs(ofm->command)) {
3528 return add_flow(p, ofconn, ofm, n_actions);
3531 return modify_flows_loose(p, ofconn, ofm, n_actions);
3533 case OFPFC_MODIFY_STRICT:
3534 return modify_flow_strict(p, ofconn, ofm, n_actions);
3537 delete_flows_loose(p, ofm);
3540 case OFPFC_DELETE_STRICT:
3541 delete_flow_strict(p, ofm);
3545 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3550 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3554 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3559 p->tun_id_from_cookie = !!msg->set;
3564 handle_role_request(struct ofproto *ofproto,
3565 struct ofconn *ofconn, struct nicira_header *msg)
3567 struct nx_role_request *nrr;
3568 struct nx_role_request *reply;
3572 if (ntohs(msg->header.length) != sizeof *nrr) {
3573 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3574 ntohs(msg->header.length), sizeof *nrr);
3575 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3577 nrr = (struct nx_role_request *) msg;
3579 if (ofconn->type != OFCONN_CONTROLLER) {
3580 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3582 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3585 role = ntohl(nrr->role);
3586 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3587 && role != NX_ROLE_SLAVE) {
3588 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3590 /* There's no good error code for this. */
3591 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3594 if (role == NX_ROLE_MASTER) {
3595 struct ofconn *other;
3597 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3598 &ofproto->controllers) {
3599 if (other->role == NX_ROLE_MASTER) {
3600 other->role = NX_ROLE_SLAVE;
3604 ofconn->role = role;
3606 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3608 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3609 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3610 reply->role = htonl(role);
3611 queue_tx(buf, ofconn, ofconn->reply_counter);
3617 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3619 struct ofp_vendor_header *ovh = msg;
3620 struct nicira_header *nh;
3622 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3623 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3624 "(expected at least %zu)",
3625 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3626 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3628 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3629 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3631 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3632 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3633 "(expected at least %zu)",
3634 ntohs(ovh->header.length), sizeof(struct nicira_header));
3635 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3639 switch (ntohl(nh->subtype)) {
3640 case NXT_STATUS_REQUEST:
3641 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3644 case NXT_TUN_ID_FROM_COOKIE:
3645 return handle_tun_id_from_cookie(p, msg);
3647 case NXT_ROLE_REQUEST:
3648 return handle_role_request(p, ofconn, msg);
3651 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3655 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3657 struct ofp_header *ob;
3660 /* Currently, everything executes synchronously, so we can just
3661 * immediately send the barrier reply. */
3662 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3663 queue_tx(buf, ofconn, ofconn->reply_counter);
3668 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3669 struct ofpbuf *ofp_msg)
3671 struct ofp_header *oh = ofp_msg->data;
3674 COVERAGE_INC(ofproto_recv_openflow);
3676 case OFPT_ECHO_REQUEST:
3677 error = handle_echo_request(ofconn, oh);
3680 case OFPT_ECHO_REPLY:
3684 case OFPT_FEATURES_REQUEST:
3685 error = handle_features_request(p, ofconn, oh);
3688 case OFPT_GET_CONFIG_REQUEST:
3689 error = handle_get_config_request(p, ofconn, oh);
3692 case OFPT_SET_CONFIG:
3693 error = handle_set_config(p, ofconn, ofp_msg->data);
3696 case OFPT_PACKET_OUT:
3697 error = handle_packet_out(p, ofconn, ofp_msg->data);
3701 error = handle_port_mod(p, ofconn, oh);
3705 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3708 case OFPT_STATS_REQUEST:
3709 error = handle_stats_request(p, ofconn, oh);
3713 error = handle_vendor(p, ofconn, ofp_msg->data);
3716 case OFPT_BARRIER_REQUEST:
3717 error = handle_barrier_request(ofconn, oh);
3721 if (VLOG_IS_WARN_ENABLED()) {
3722 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3723 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3726 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3731 send_error_oh(ofconn, ofp_msg->data, error);
3736 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3738 struct odp_msg *msg = packet->data;
3740 struct ofpbuf payload;
3743 payload.data = msg + 1;
3744 payload.size = msg->length - sizeof *msg;
3745 flow_extract(&payload, msg->arg, msg->port, &flow);
3747 /* Check with in-band control to see if this packet should be sent
3748 * to the local port regardless of the flow table. */
3749 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3750 union odp_action action;
3752 memset(&action, 0, sizeof(action));
3753 action.output.type = ODPAT_OUTPUT;
3754 action.output.port = ODPP_LOCAL;
3755 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3758 rule = lookup_valid_rule(p, &flow);
3760 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3761 struct ofport *port = port_array_get(&p->ports, msg->port);
3763 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3764 COVERAGE_INC(ofproto_no_packet_in);
3765 /* XXX install 'drop' flow entry */
3766 ofpbuf_delete(packet);
3770 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3773 COVERAGE_INC(ofproto_packet_in);
3774 send_packet_in(p, packet);
3778 if (rule->cr.wc.wildcards) {
3779 rule = rule_create_subrule(p, rule, &flow);
3780 rule_make_actions(p, rule, packet);
3782 if (!rule->may_install) {
3783 /* The rule is not installable, that is, we need to process every
3784 * packet, so process the current packet and set its actions into
3786 rule_make_actions(p, rule, packet);
3788 /* XXX revalidate rule if it needs it */
3792 rule_execute(p, rule, &payload, &flow);
3793 rule_reinstall(p, rule);
3795 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3797 * Extra-special case for fail-open mode.
3799 * We are in fail-open mode and the packet matched the fail-open rule,
3800 * but we are connected to a controller too. We should send the packet
3801 * up to the controller in the hope that it will try to set up a flow
3802 * and thereby allow us to exit fail-open.
3804 * See the top-level comment in fail-open.c for more information.
3806 send_packet_in(p, packet);
3808 ofpbuf_delete(packet);
3813 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3815 struct odp_msg *msg = packet->data;
3817 switch (msg->type) {
3818 case _ODPL_ACTION_NR:
3819 COVERAGE_INC(ofproto_ctlr_action);
3820 send_packet_in(p, packet);
3823 case _ODPL_SFLOW_NR:
3825 ofproto_sflow_received(p->sflow, msg);
3827 ofpbuf_delete(packet);
3831 handle_odp_miss_msg(p, packet);
3835 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3842 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3844 struct rule *sub = rule_from_cls_rule(sub_);
3845 struct revalidate_cbdata *cbdata = cbdata_;
3847 if (cbdata->revalidate_all
3848 || (cbdata->revalidate_subrules && sub->super)
3849 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3850 revalidate_rule(cbdata->ofproto, sub);
3855 revalidate_rule(struct ofproto *p, struct rule *rule)
3857 const flow_t *flow = &rule->cr.flow;
3859 COVERAGE_INC(ofproto_revalidate_rule);
3862 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3864 rule_remove(p, rule);
3866 } else if (super != rule->super) {
3867 COVERAGE_INC(ofproto_revalidate_moved);
3868 list_remove(&rule->list);
3869 list_push_back(&super->list, &rule->list);
3870 rule->super = super;
3871 rule->hard_timeout = super->hard_timeout;
3872 rule->idle_timeout = super->idle_timeout;
3873 rule->created = super->created;
3878 rule_update_actions(p, rule);
3882 static struct ofpbuf *
3883 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3884 long long int now, uint8_t reason)
3886 struct ofp_flow_removed *ofr;
3888 long long int tdiff = now - rule->created;
3889 uint32_t sec = tdiff / 1000;
3890 uint32_t msec = tdiff - (sec * 1000);
3892 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3893 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3895 ofr->cookie = rule->flow_cookie;
3896 ofr->priority = htons(rule->cr.priority);
3897 ofr->reason = reason;
3898 ofr->duration_sec = htonl(sec);
3899 ofr->duration_nsec = htonl(msec * 1000000);
3900 ofr->idle_timeout = htons(rule->idle_timeout);
3901 ofr->packet_count = htonll(rule->packet_count);
3902 ofr->byte_count = htonll(rule->byte_count);
3908 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3910 assert(rule->installed);
3911 assert(!rule->cr.wc.wildcards);
3914 rule_remove(ofproto, rule);
3916 rule_uninstall(ofproto, rule);
3921 send_flow_removed(struct ofproto *p, struct rule *rule,
3922 long long int now, uint8_t reason)
3924 struct ofconn *ofconn;
3925 struct ofconn *prev;
3926 struct ofpbuf *buf = NULL;
3928 /* We limit the maximum number of queued flow expirations it by accounting
3929 * them under the counter for replies. That works because preventing
3930 * OpenFlow requests from being processed also prevents new flows from
3931 * being added (and expiring). (It also prevents processing OpenFlow
3932 * requests that would not add new flows, so it is imperfect.) */
3935 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3936 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
3937 && ofconn_receives_async_msgs(ofconn)) {
3939 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3941 buf = compose_flow_removed(p, rule, now, reason);
3947 queue_tx(buf, prev, prev->reply_counter);
3953 expire_rule(struct cls_rule *cls_rule, void *p_)
3955 struct ofproto *p = p_;
3956 struct rule *rule = rule_from_cls_rule(cls_rule);
3957 long long int hard_expire, idle_expire, expire, now;
3959 hard_expire = (rule->hard_timeout
3960 ? rule->created + rule->hard_timeout * 1000
3962 idle_expire = (rule->idle_timeout
3963 && (rule->super || list_is_empty(&rule->list))
3964 ? rule->used + rule->idle_timeout * 1000
3966 expire = MIN(hard_expire, idle_expire);
3970 if (rule->installed && now >= rule->used + 5000) {
3971 uninstall_idle_flow(p, rule);
3972 } else if (!rule->cr.wc.wildcards) {
3973 active_timeout(p, rule);
3979 COVERAGE_INC(ofproto_expired);
3981 /* Update stats. This code will be a no-op if the rule expired
3982 * due to an idle timeout. */
3983 if (rule->cr.wc.wildcards) {
3984 struct rule *subrule, *next;
3985 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3986 rule_remove(p, subrule);
3989 rule_uninstall(p, rule);
3992 if (!rule_is_hidden(rule)) {
3993 send_flow_removed(p, rule, now,
3995 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3997 rule_remove(p, rule);
4001 active_timeout(struct ofproto *ofproto, struct rule *rule)
4003 if (ofproto->netflow && !is_controller_rule(rule) &&
4004 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4005 struct ofexpired expired;
4006 struct odp_flow odp_flow;
4008 /* Get updated flow stats. */
4009 memset(&odp_flow, 0, sizeof odp_flow);
4010 if (rule->installed) {
4011 odp_flow.key = rule->cr.flow;
4012 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4013 dpif_flow_get(ofproto->dpif, &odp_flow);
4015 if (odp_flow.stats.n_packets) {
4016 update_time(ofproto, rule, &odp_flow.stats);
4017 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
4018 odp_flow.stats.tcp_flags);
4022 expired.flow = rule->cr.flow;
4023 expired.packet_count = rule->packet_count +
4024 odp_flow.stats.n_packets;
4025 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4026 expired.used = rule->used;
4028 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4030 /* Schedule us to send the accumulated records once we have
4031 * collected all of them. */
4032 poll_immediate_wake();
4037 update_used(struct ofproto *p)
4039 struct odp_flow *flows;
4044 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4049 for (i = 0; i < n_flows; i++) {
4050 struct odp_flow *f = &flows[i];
4053 rule = rule_from_cls_rule(
4054 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4055 if (!rule || !rule->installed) {
4056 COVERAGE_INC(ofproto_unexpected_rule);
4057 dpif_flow_del(p->dpif, f);
4061 update_time(p, rule, &f->stats);
4062 rule_account(p, rule, f->stats.n_bytes);
4067 /* pinsched callback for sending 'packet' on 'ofconn'. */
4069 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4071 struct ofconn *ofconn = ofconn_;
4073 rconn_send_with_limit(ofconn->rconn, packet,
4074 ofconn->packet_in_counter, 100);
4077 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4078 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4079 * packet scheduler for sending.
4081 * 'max_len' specifies the maximum number of bytes of the packet to send on
4082 * 'ofconn' (INT_MAX specifies no limit).
4084 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4085 * ownership is transferred to this function. */
4087 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4090 struct ofproto *ofproto = ofconn->ofproto;
4091 struct ofp_packet_in *opi = packet->data;
4092 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4093 int send_len, trim_size;
4097 if (opi->reason == OFPR_ACTION) {
4098 buffer_id = UINT32_MAX;
4099 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4100 buffer_id = pktbuf_get_null();
4101 } else if (!ofconn->pktbuf) {
4102 buffer_id = UINT32_MAX;
4104 struct ofpbuf payload;
4105 payload.data = opi->data;
4106 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4107 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4110 /* Figure out how much of the packet to send. */
4111 send_len = ntohs(opi->total_len);
4112 if (buffer_id != UINT32_MAX) {
4113 send_len = MIN(send_len, ofconn->miss_send_len);
4115 send_len = MIN(send_len, max_len);
4117 /* Adjust packet length and clone if necessary. */
4118 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4120 packet = ofpbuf_clone_data(packet->data, trim_size);
4123 packet->size = trim_size;
4126 /* Update packet headers. */
4127 opi->buffer_id = htonl(buffer_id);
4128 update_openflow_length(packet);
4130 /* Hand over to packet scheduler. It might immediately call into
4131 * do_send_packet_in() or it might buffer it for a while (until a later
4132 * call to pinsched_run()). */
4133 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4134 packet, do_send_packet_in, ofconn);
4137 /* Replace struct odp_msg header in 'packet' by equivalent struct
4138 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4139 * returned by dpif_recv()).
4141 * The conversion is not complete: the caller still needs to trim any unneeded
4142 * payload off the end of the buffer, set the length in the OpenFlow header,
4143 * and set buffer_id. Those require us to know the controller settings and so
4144 * must be done on a per-controller basis.
4146 * Returns the maximum number of bytes of the packet that should be sent to
4147 * the controller (INT_MAX if no limit). */
4149 do_convert_to_packet_in(struct ofpbuf *packet)
4151 struct odp_msg *msg = packet->data;
4152 struct ofp_packet_in *opi;
4158 /* Extract relevant header fields */
4159 if (msg->type == _ODPL_ACTION_NR) {
4160 reason = OFPR_ACTION;
4163 reason = OFPR_NO_MATCH;
4166 total_len = msg->length - sizeof *msg;
4167 in_port = odp_port_to_ofp_port(msg->port);
4169 /* Repurpose packet buffer by overwriting header. */
4170 ofpbuf_pull(packet, sizeof(struct odp_msg));
4171 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4172 opi->header.version = OFP_VERSION;
4173 opi->header.type = OFPT_PACKET_IN;
4174 opi->total_len = htons(total_len);
4175 opi->in_port = htons(in_port);
4176 opi->reason = reason;
4181 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4182 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4183 * as necessary according to their individual configurations.
4185 * 'packet' must have sufficient headroom to convert it into a struct
4186 * ofp_packet_in (e.g. as returned by dpif_recv()).
4188 * Takes ownership of 'packet'. */
4190 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4192 struct ofconn *ofconn, *prev;
4195 max_len = do_convert_to_packet_in(packet);
4198 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4199 if (ofconn_receives_async_msgs(ofconn)) {
4201 schedule_packet_in(prev, packet, max_len, true);
4207 schedule_packet_in(prev, packet, max_len, false);
4209 ofpbuf_delete(packet);
4214 pick_datapath_id(const struct ofproto *ofproto)
4216 const struct ofport *port;
4218 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4220 uint8_t ea[ETH_ADDR_LEN];
4223 error = netdev_get_etheraddr(port->netdev, ea);
4225 return eth_addr_to_uint64(ea);
4227 VLOG_WARN("could not get MAC address for %s (%s)",
4228 netdev_get_name(port->netdev), strerror(error));
4230 return ofproto->fallback_dpid;
4234 pick_fallback_dpid(void)
4236 uint8_t ea[ETH_ADDR_LEN];
4237 eth_addr_nicira_random(ea);
4238 return eth_addr_to_uint64(ea);
4242 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4243 struct odp_actions *actions, tag_type *tags,
4244 uint16_t *nf_output_iface, void *ofproto_)
4246 struct ofproto *ofproto = ofproto_;
4249 /* Drop frames for reserved multicast addresses. */
4250 if (eth_addr_is_reserved(flow->dl_dst)) {
4254 /* Learn source MAC (but don't try to learn from revalidation). */
4255 if (packet != NULL) {
4256 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4258 GRAT_ARP_LOCK_NONE);
4260 /* The log messages here could actually be useful in debugging,
4261 * so keep the rate limit relatively high. */
4262 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4263 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4264 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4265 ofproto_revalidate(ofproto, rev_tag);
4269 /* Determine output port. */
4270 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4273 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4274 } else if (out_port != flow->in_port) {
4275 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4276 *nf_output_iface = out_port;
4284 static const struct ofhooks default_ofhooks = {
4286 default_normal_ofhook_cb,