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 "byte-order.h"
28 #include "classifier.h"
30 #include "discovery.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
41 #include "ofp-print.h"
43 #include "ofproto-sflow.h"
45 #include "openflow/nicira-ext.h"
46 #include "openflow/openflow.h"
47 #include "openvswitch/datapath-protocol.h"
51 #include "poll-loop.h"
55 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto);
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER = 1
74 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
75 struct netdev *netdev;
76 struct ofp_phy_port opp; /* In host byte order. */
80 static void ofport_free(struct ofport *);
81 static void hton_ofp_phy_port(struct ofp_phy_port *);
83 static int xlate_actions(const union ofp_action *in, size_t n_in,
84 const struct flow *, struct ofproto *,
85 const struct ofpbuf *packet,
86 struct odp_actions *out, tag_type *tags,
87 bool *may_set_up_flow, uint16_t *nf_output_iface);
92 ovs_be64 flow_cookie; /* Controller-issued identifier. */
93 uint16_t idle_timeout; /* In seconds from time of last use. */
94 uint16_t hard_timeout; /* In seconds from time of creation. */
95 bool send_flow_removed; /* Send a flow removed message? */
96 long long int used; /* Last-used time (0 if never used). */
97 long long int created; /* Creation time. */
98 uint64_t packet_count; /* Number of packets received. */
99 uint64_t byte_count; /* Number of bytes received. */
100 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
101 tag_type tags; /* Tags (set only by hooks). */
102 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
104 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
105 * exact-match rule (having cr.wc.wildcards of 0) generated from the
106 * wildcard rule 'super'. In this case, 'list' is an element of the
109 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
110 * a list of subrules. A super-rule with no wildcards (where
111 * cr.wc.wildcards is 0) will never have any subrules. */
117 * 'n_actions' is the number of elements in the 'actions' array. A single
118 * action may take up more more than one element's worth of space.
120 * A subrule has no actions (it uses the super-rule's actions). */
122 union ofp_action *actions;
126 * A super-rule with wildcard fields never has ODP actions (since the
127 * datapath only supports exact-match flows). */
128 bool installed; /* Installed in datapath? */
129 bool may_install; /* True ordinarily; false if actions must
130 * be reassessed for every packet. */
132 union odp_action *odp_actions;
136 rule_is_hidden(const struct rule *rule)
138 /* Subrules are merely an implementation detail, so hide them from the
140 if (rule->super != NULL) {
144 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
145 * (e.g. by in-band control) and are intentionally hidden from the
147 if (rule->cr.priority > UINT16_MAX) {
154 static struct rule *rule_create(struct ofproto *, struct rule *super,
155 const union ofp_action *, size_t n_actions,
156 uint16_t idle_timeout, uint16_t hard_timeout,
157 ovs_be64 flow_cookie, bool send_flow_removed);
158 static void rule_free(struct rule *);
159 static void rule_destroy(struct ofproto *, struct rule *);
160 static struct rule *rule_from_cls_rule(const struct cls_rule *);
161 static void rule_insert(struct ofproto *, struct rule *,
162 struct ofpbuf *packet, uint16_t in_port);
163 static void rule_remove(struct ofproto *, struct rule *);
164 static bool rule_make_actions(struct ofproto *, struct rule *,
165 const struct ofpbuf *packet);
166 static void rule_install(struct ofproto *, struct rule *,
167 struct rule *displaced_rule);
168 static void rule_uninstall(struct ofproto *, struct rule *);
169 static void rule_post_uninstall(struct ofproto *, struct rule *);
170 static void send_flow_removed(struct ofproto *p, struct rule *rule,
171 long long int now, uint8_t reason);
173 /* ofproto supports two kinds of OpenFlow connections:
175 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
176 * maintains persistent connections to these controllers and by default
177 * sends them asynchronous messages such as packet-ins.
179 * - "Service" connections, e.g. from ovs-ofctl. When these connections
180 * drop, it is the other side's responsibility to reconnect them if
181 * necessary. ofproto does not send them asynchronous messages by default.
183 * Currently, active (tcp, ssl, unix) connections are always "primary"
184 * connections and passive (ptcp, pssl, punix) connections are always "service"
185 * connections. There is no inherent reason for this, but it reflects the
189 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
190 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
193 /* A listener for incoming OpenFlow "service" connections. */
195 struct hmap_node node; /* In struct ofproto's "services" hmap. */
196 struct pvconn *pvconn; /* OpenFlow connection listener. */
198 /* These are not used by ofservice directly. They are settings for
199 * accepted "struct ofconn"s from the pvconn. */
200 int probe_interval; /* Max idle time before probing, in seconds. */
201 int rate_limit; /* Max packet-in rate in packets per second. */
202 int burst_limit; /* Limit on accumulating packet credits. */
205 static struct ofservice *ofservice_lookup(struct ofproto *,
207 static int ofservice_create(struct ofproto *,
208 const struct ofproto_controller *);
209 static void ofservice_reconfigure(struct ofservice *,
210 const struct ofproto_controller *);
211 static void ofservice_destroy(struct ofproto *, struct ofservice *);
213 /* An OpenFlow connection. */
215 struct ofproto *ofproto; /* The ofproto that owns this connection. */
216 struct list node; /* In struct ofproto's "all_conns" list. */
217 struct rconn *rconn; /* OpenFlow connection. */
218 enum ofconn_type type; /* Type. */
220 /* OFPT_PACKET_IN related data. */
221 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
222 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
223 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
224 int miss_send_len; /* Bytes to send of buffered packets. */
226 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
227 * requests, and the maximum number before we stop reading OpenFlow
229 #define OFCONN_REPLY_MAX 100
230 struct rconn_packet_counter *reply_counter;
232 /* type == OFCONN_PRIMARY only. */
233 enum nx_role role; /* Role. */
234 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
235 struct discovery *discovery; /* Controller discovery object, if enabled. */
236 struct status_category *ss; /* Switch status category. */
237 enum ofproto_band band; /* In-band or out-of-band? */
240 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
241 * "schedulers" array. Their values are 0 and 1, and their meanings and values
242 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
243 * case anything ever changes, check their values here. */
244 #define N_SCHEDULERS 2
245 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
246 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
247 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
248 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
250 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
252 static void ofconn_destroy(struct ofconn *);
253 static void ofconn_run(struct ofconn *, struct ofproto *);
254 static void ofconn_wait(struct ofconn *);
255 static bool ofconn_receives_async_msgs(const struct ofconn *);
256 static char *ofconn_make_name(const struct ofproto *, const char *target);
257 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
259 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
260 struct rconn_packet_counter *counter);
262 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
263 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
267 uint64_t datapath_id; /* Datapath ID. */
268 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
269 char *mfr_desc; /* Manufacturer. */
270 char *hw_desc; /* Hardware. */
271 char *sw_desc; /* Software version. */
272 char *serial_desc; /* Serial number. */
273 char *dp_desc; /* Datapath description. */
277 struct netdev_monitor *netdev_monitor;
278 struct hmap ports; /* Contains "struct ofport"s. */
279 struct shash port_by_name;
283 struct switch_status *switch_status;
284 struct fail_open *fail_open;
285 struct netflow *netflow;
286 struct ofproto_sflow *sflow;
288 /* In-band control. */
289 struct in_band *in_band;
290 long long int next_in_band_update;
291 struct sockaddr_in *extra_in_band_remotes;
292 size_t n_extra_remotes;
295 struct classifier cls;
296 bool need_revalidate;
297 long long int next_expiration;
298 struct tag_set revalidate_set;
299 bool tun_id_from_cookie;
301 /* OpenFlow connections. */
302 struct hmap controllers; /* Controller "struct ofconn"s. */
303 struct list all_conns; /* Contains "struct ofconn"s. */
304 enum ofproto_fail_mode fail_mode;
306 /* OpenFlow listeners. */
307 struct hmap services; /* Contains "struct ofservice"s. */
308 struct pvconn **snoops;
311 /* Hooks for ovs-vswitchd. */
312 const struct ofhooks *ofhooks;
315 /* Used by default ofhooks. */
316 struct mac_learning *ml;
319 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
321 static const struct ofhooks default_ofhooks;
323 static uint64_t pick_datapath_id(const struct ofproto *);
324 static uint64_t pick_fallback_dpid(void);
326 static int ofproto_expire(struct ofproto *);
328 static void update_stats(struct ofproto *, struct rule *,
329 const struct odp_flow_stats *);
330 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
331 static void revalidate_cb(struct cls_rule *rule_, void *p_);
333 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
335 static void handle_openflow(struct ofconn *, struct ofproto *,
338 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
339 static void update_port(struct ofproto *, const char *devname);
340 static int init_ports(struct ofproto *);
341 static void reinit_ports(struct ofproto *);
344 ofproto_create(const char *datapath, const char *datapath_type,
345 const struct ofhooks *ofhooks, void *aux,
346 struct ofproto **ofprotop)
348 struct odp_stats stats;
355 /* Connect to datapath and start listening for messages. */
356 error = dpif_open(datapath, datapath_type, &dpif);
358 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
361 error = dpif_get_dp_stats(dpif, &stats);
363 VLOG_ERR("failed to obtain stats for datapath %s: %s",
364 datapath, strerror(error));
368 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
370 VLOG_ERR("failed to listen on datapath %s: %s",
371 datapath, strerror(error));
375 dpif_flow_flush(dpif);
376 dpif_recv_purge(dpif);
378 /* Initialize settings. */
379 p = xzalloc(sizeof *p);
380 p->fallback_dpid = pick_fallback_dpid();
381 p->datapath_id = p->fallback_dpid;
382 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
383 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
384 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
385 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
386 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
388 /* Initialize datapath. */
390 p->netdev_monitor = netdev_monitor_create();
391 hmap_init(&p->ports);
392 shash_init(&p->port_by_name);
393 p->max_ports = stats.max_ports;
395 /* Initialize submodules. */
396 p->switch_status = switch_status_create(p);
402 /* Initialize flow table. */
403 classifier_init(&p->cls);
404 p->need_revalidate = false;
405 p->next_expiration = time_msec() + 1000;
406 tag_set_init(&p->revalidate_set);
408 /* Initialize OpenFlow connections. */
409 list_init(&p->all_conns);
410 hmap_init(&p->controllers);
411 hmap_init(&p->services);
415 /* Initialize hooks. */
417 p->ofhooks = ofhooks;
421 p->ofhooks = &default_ofhooks;
423 p->ml = mac_learning_create();
426 /* Pick final datapath ID. */
427 p->datapath_id = pick_datapath_id(p);
428 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
435 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
437 uint64_t old_dpid = p->datapath_id;
438 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
439 if (p->datapath_id != old_dpid) {
440 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
442 /* Force all active connections to reconnect, since there is no way to
443 * notify a controller that the datapath ID has changed. */
444 ofproto_reconnect_controllers(p);
449 is_discovery_controller(const struct ofproto_controller *c)
451 return !strcmp(c->target, "discover");
455 is_in_band_controller(const struct ofproto_controller *c)
457 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
460 /* Creates a new controller in 'ofproto'. Some of the settings are initially
461 * drawn from 'c', but update_controller() needs to be called later to finish
462 * the new ofconn's configuration. */
464 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
466 struct discovery *discovery;
467 struct ofconn *ofconn;
469 if (is_discovery_controller(c)) {
470 int error = discovery_create(c->accept_re, c->update_resolv_conf,
471 ofproto->dpif, ofproto->switch_status,
480 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
481 ofconn->pktbuf = pktbuf_create();
482 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
484 ofconn->discovery = discovery;
486 char *name = ofconn_make_name(ofproto, c->target);
487 rconn_connect(ofconn->rconn, c->target, name);
490 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
491 hash_string(c->target, 0));
494 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
495 * target or turn discovery on or off (these are done by creating new ofconns
496 * and deleting old ones), but it can update the rest of an ofconn's
499 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
503 ofconn->band = (is_in_band_controller(c)
504 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
506 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
508 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
509 rconn_set_probe_interval(ofconn->rconn, probe_interval);
511 if (ofconn->discovery) {
512 discovery_set_update_resolv_conf(ofconn->discovery,
513 c->update_resolv_conf);
514 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
517 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
521 ofconn_get_target(const struct ofconn *ofconn)
523 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
526 static struct ofconn *
527 find_controller_by_target(struct ofproto *ofproto, const char *target)
529 struct ofconn *ofconn;
531 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
532 hash_string(target, 0), &ofproto->controllers) {
533 if (!strcmp(ofconn_get_target(ofconn), target)) {
541 update_in_band_remotes(struct ofproto *ofproto)
543 const struct ofconn *ofconn;
544 struct sockaddr_in *addrs;
545 size_t max_addrs, n_addrs;
549 /* Allocate enough memory for as many remotes as we could possibly have. */
550 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
551 addrs = xmalloc(max_addrs * sizeof *addrs);
554 /* Add all the remotes. */
556 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
557 struct sockaddr_in *sin = &addrs[n_addrs];
559 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
563 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
564 if (sin->sin_addr.s_addr) {
565 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
568 if (ofconn->discovery) {
572 for (i = 0; i < ofproto->n_extra_remotes; i++) {
573 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
576 /* Create or update or destroy in-band.
578 * Ordinarily we only enable in-band if there's at least one remote
579 * address, but discovery needs the in-band rules for DHCP to be installed
580 * even before we know any remote addresses. */
581 if (n_addrs || discovery) {
582 if (!ofproto->in_band) {
583 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
586 if (ofproto->in_band) {
587 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
589 ofproto->next_in_band_update = time_msec() + 1000;
591 in_band_destroy(ofproto->in_band);
592 ofproto->in_band = NULL;
600 update_fail_open(struct ofproto *p)
602 struct ofconn *ofconn;
604 if (!hmap_is_empty(&p->controllers)
605 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
606 struct rconn **rconns;
610 p->fail_open = fail_open_create(p, p->switch_status);
614 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
615 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
616 rconns[n++] = ofconn->rconn;
619 fail_open_set_controllers(p->fail_open, rconns, n);
620 /* p->fail_open takes ownership of 'rconns'. */
622 fail_open_destroy(p->fail_open);
628 ofproto_set_controllers(struct ofproto *p,
629 const struct ofproto_controller *controllers,
630 size_t n_controllers)
632 struct shash new_controllers;
633 struct ofconn *ofconn, *next_ofconn;
634 struct ofservice *ofservice, *next_ofservice;
638 /* Create newly configured controllers and services.
639 * Create a name to ofproto_controller mapping in 'new_controllers'. */
640 shash_init(&new_controllers);
641 for (i = 0; i < n_controllers; i++) {
642 const struct ofproto_controller *c = &controllers[i];
644 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
645 if (!find_controller_by_target(p, c->target)) {
646 add_controller(p, c);
648 } else if (!pvconn_verify_name(c->target)) {
649 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
653 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
654 dpif_name(p->dpif), c->target);
658 shash_add_once(&new_controllers, c->target, &controllers[i]);
661 /* Delete controllers that are no longer configured.
662 * Update configuration of all now-existing controllers. */
664 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
665 struct ofproto_controller *c;
667 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
669 ofconn_destroy(ofconn);
671 update_controller(ofconn, c);
678 /* Delete services that are no longer configured.
679 * Update configuration of all now-existing services. */
680 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
681 struct ofproto_controller *c;
683 c = shash_find_data(&new_controllers,
684 pvconn_get_name(ofservice->pvconn));
686 ofservice_destroy(p, ofservice);
688 ofservice_reconfigure(ofservice, c);
692 shash_destroy(&new_controllers);
694 update_in_band_remotes(p);
697 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
698 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
699 struct ofconn, hmap_node);
700 ofconn->ss = switch_status_register(p->switch_status, "remote",
701 rconn_status_cb, ofconn->rconn);
706 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
708 p->fail_mode = fail_mode;
712 /* Drops the connections between 'ofproto' and all of its controllers, forcing
713 * them to reconnect. */
715 ofproto_reconnect_controllers(struct ofproto *ofproto)
717 struct ofconn *ofconn;
719 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
720 rconn_reconnect(ofconn->rconn);
725 any_extras_changed(const struct ofproto *ofproto,
726 const struct sockaddr_in *extras, size_t n)
730 if (n != ofproto->n_extra_remotes) {
734 for (i = 0; i < n; i++) {
735 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
736 const struct sockaddr_in *new = &extras[i];
738 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
739 old->sin_port != new->sin_port) {
747 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
748 * in-band control should guarantee access, in the same way that in-band
749 * control guarantees access to OpenFlow controllers. */
751 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
752 const struct sockaddr_in *extras, size_t n)
754 if (!any_extras_changed(ofproto, extras, n)) {
758 free(ofproto->extra_in_band_remotes);
759 ofproto->n_extra_remotes = n;
760 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
762 update_in_band_remotes(ofproto);
766 ofproto_set_desc(struct ofproto *p,
767 const char *mfr_desc, const char *hw_desc,
768 const char *sw_desc, const char *serial_desc,
771 struct ofp_desc_stats *ods;
774 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
775 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
776 sizeof ods->mfr_desc);
779 p->mfr_desc = xstrdup(mfr_desc);
782 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
783 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
784 sizeof ods->hw_desc);
787 p->hw_desc = xstrdup(hw_desc);
790 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
791 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
792 sizeof ods->sw_desc);
795 p->sw_desc = xstrdup(sw_desc);
798 if (strlen(serial_desc) >= sizeof ods->serial_num) {
799 VLOG_WARN("truncating serial_desc, must be less than %zu "
801 sizeof ods->serial_num);
803 free(p->serial_desc);
804 p->serial_desc = xstrdup(serial_desc);
807 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
808 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
809 sizeof ods->dp_desc);
812 p->dp_desc = xstrdup(dp_desc);
817 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
818 const struct svec *svec)
820 struct pvconn **pvconns = *pvconnsp;
821 size_t n_pvconns = *n_pvconnsp;
825 for (i = 0; i < n_pvconns; i++) {
826 pvconn_close(pvconns[i]);
830 pvconns = xmalloc(svec->n * sizeof *pvconns);
832 for (i = 0; i < svec->n; i++) {
833 const char *name = svec->names[i];
834 struct pvconn *pvconn;
837 error = pvconn_open(name, &pvconn);
839 pvconns[n_pvconns++] = pvconn;
841 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
849 *n_pvconnsp = n_pvconns;
855 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
857 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
861 ofproto_set_netflow(struct ofproto *ofproto,
862 const struct netflow_options *nf_options)
864 if (nf_options && nf_options->collectors.n) {
865 if (!ofproto->netflow) {
866 ofproto->netflow = netflow_create();
868 return netflow_set_options(ofproto->netflow, nf_options);
870 netflow_destroy(ofproto->netflow);
871 ofproto->netflow = NULL;
877 ofproto_set_sflow(struct ofproto *ofproto,
878 const struct ofproto_sflow_options *oso)
880 struct ofproto_sflow *os = ofproto->sflow;
883 struct ofport *ofport;
885 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
886 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
887 ofproto_sflow_add_port(os, ofport->odp_port,
888 netdev_get_name(ofport->netdev));
891 ofproto_sflow_set_options(os, oso);
893 ofproto_sflow_destroy(os);
894 ofproto->sflow = NULL;
899 ofproto_get_datapath_id(const struct ofproto *ofproto)
901 return ofproto->datapath_id;
905 ofproto_has_primary_controller(const struct ofproto *ofproto)
907 return !hmap_is_empty(&ofproto->controllers);
910 enum ofproto_fail_mode
911 ofproto_get_fail_mode(const struct ofproto *p)
917 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
921 for (i = 0; i < ofproto->n_snoops; i++) {
922 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
927 ofproto_destroy(struct ofproto *p)
929 struct ofservice *ofservice, *next_ofservice;
930 struct ofconn *ofconn, *next_ofconn;
931 struct ofport *ofport, *next_ofport;
938 /* Destroy fail-open and in-band early, since they touch the classifier. */
939 fail_open_destroy(p->fail_open);
942 in_band_destroy(p->in_band);
944 free(p->extra_in_band_remotes);
946 ofproto_flush_flows(p);
947 classifier_destroy(&p->cls);
949 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
950 ofconn_destroy(ofconn);
952 hmap_destroy(&p->controllers);
955 netdev_monitor_destroy(p->netdev_monitor);
956 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
957 hmap_remove(&p->ports, &ofport->hmap_node);
960 shash_destroy(&p->port_by_name);
962 switch_status_destroy(p->switch_status);
963 netflow_destroy(p->netflow);
964 ofproto_sflow_destroy(p->sflow);
966 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
967 ofservice_destroy(p, ofservice);
969 hmap_destroy(&p->services);
971 for (i = 0; i < p->n_snoops; i++) {
972 pvconn_close(p->snoops[i]);
976 mac_learning_destroy(p->ml);
981 free(p->serial_desc);
984 hmap_destroy(&p->ports);
990 ofproto_run(struct ofproto *p)
992 int error = ofproto_run1(p);
994 error = ofproto_run2(p, false);
1000 process_port_change(struct ofproto *ofproto, int error, char *devname)
1002 if (error == ENOBUFS) {
1003 reinit_ports(ofproto);
1004 } else if (!error) {
1005 update_port(ofproto, devname);
1010 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1011 * means that 'ofconn' is more interesting for monitoring than a lower return
1014 snoop_preference(const struct ofconn *ofconn)
1016 switch (ofconn->role) {
1017 case NX_ROLE_MASTER:
1024 /* Shouldn't happen. */
1029 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1030 * Connects this vconn to a controller. */
1032 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1034 struct ofconn *ofconn, *best;
1036 /* Pick a controller for monitoring. */
1038 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1039 if (ofconn->type == OFCONN_PRIMARY
1040 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1046 rconn_add_monitor(best->rconn, vconn);
1048 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1054 ofproto_run1(struct ofproto *p)
1056 struct ofconn *ofconn, *next_ofconn;
1057 struct ofservice *ofservice;
1062 if (shash_is_empty(&p->port_by_name)) {
1066 for (i = 0; i < 50; i++) {
1069 error = dpif_recv(p->dpif, &buf);
1071 if (error == ENODEV) {
1072 /* Someone destroyed the datapath behind our back. The caller
1073 * better destroy us and give up, because we're just going to
1074 * spin from here on out. */
1075 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1076 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1077 dpif_name(p->dpif));
1083 handle_odp_msg(p, buf);
1086 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1087 process_port_change(p, error, devname);
1089 while ((error = netdev_monitor_poll(p->netdev_monitor,
1090 &devname)) != EAGAIN) {
1091 process_port_change(p, error, devname);
1095 if (time_msec() >= p->next_in_band_update) {
1096 update_in_band_remotes(p);
1098 in_band_run(p->in_band);
1101 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1102 ofconn_run(ofconn, p);
1105 /* Fail-open maintenance. Do this after processing the ofconns since
1106 * fail-open checks the status of the controller rconn. */
1108 fail_open_run(p->fail_open);
1111 HMAP_FOR_EACH (ofservice, node, &p->services) {
1112 struct vconn *vconn;
1115 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1117 struct rconn *rconn;
1120 rconn = rconn_create(ofservice->probe_interval, 0);
1121 name = ofconn_make_name(p, vconn_get_name(vconn));
1122 rconn_connect_unreliably(rconn, vconn, name);
1125 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1126 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1127 ofservice->burst_limit);
1128 } else if (retval != EAGAIN) {
1129 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1133 for (i = 0; i < p->n_snoops; i++) {
1134 struct vconn *vconn;
1137 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1139 add_snooper(p, vconn);
1140 } else if (retval != EAGAIN) {
1141 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1145 if (time_msec() >= p->next_expiration) {
1146 int delay = ofproto_expire(p);
1147 p->next_expiration = time_msec() + delay;
1148 COVERAGE_INC(ofproto_expiration);
1152 netflow_run(p->netflow);
1155 ofproto_sflow_run(p->sflow);
1161 struct revalidate_cbdata {
1162 struct ofproto *ofproto;
1163 bool revalidate_all; /* Revalidate all exact-match rules? */
1164 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1165 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1169 ofproto_run2(struct ofproto *p, bool revalidate_all)
1171 if (p->need_revalidate || revalidate_all
1172 || !tag_set_is_empty(&p->revalidate_set)) {
1173 struct revalidate_cbdata cbdata;
1175 cbdata.revalidate_all = revalidate_all;
1176 cbdata.revalidate_subrules = p->need_revalidate;
1177 cbdata.revalidate_set = p->revalidate_set;
1178 tag_set_init(&p->revalidate_set);
1179 COVERAGE_INC(ofproto_revalidate);
1180 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1181 p->need_revalidate = false;
1188 ofproto_wait(struct ofproto *p)
1190 struct ofservice *ofservice;
1191 struct ofconn *ofconn;
1194 dpif_recv_wait(p->dpif);
1195 dpif_port_poll_wait(p->dpif);
1196 netdev_monitor_poll_wait(p->netdev_monitor);
1197 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1198 ofconn_wait(ofconn);
1201 poll_timer_wait_until(p->next_in_band_update);
1202 in_band_wait(p->in_band);
1205 fail_open_wait(p->fail_open);
1208 ofproto_sflow_wait(p->sflow);
1210 if (!tag_set_is_empty(&p->revalidate_set)) {
1211 poll_immediate_wake();
1213 if (p->need_revalidate) {
1214 /* Shouldn't happen, but if it does just go around again. */
1215 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1216 poll_immediate_wake();
1217 } else if (p->next_expiration != LLONG_MAX) {
1218 poll_timer_wait_until(p->next_expiration);
1220 HMAP_FOR_EACH (ofservice, node, &p->services) {
1221 pvconn_wait(ofservice->pvconn);
1223 for (i = 0; i < p->n_snoops; i++) {
1224 pvconn_wait(p->snoops[i]);
1229 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1231 tag_set_add(&ofproto->revalidate_set, tag);
1235 ofproto_get_revalidate_set(struct ofproto *ofproto)
1237 return &ofproto->revalidate_set;
1241 ofproto_is_alive(const struct ofproto *p)
1243 return !hmap_is_empty(&p->controllers);
1246 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1248 * This is almost the same as calling dpif_port_del() directly on the
1249 * datapath, but it also makes 'ofproto' close its open netdev for the port
1250 * (if any). This makes it possible to create a new netdev of a different
1251 * type under the same name, which otherwise the netdev library would refuse
1252 * to do because of the conflict. (The netdev would eventually get closed on
1253 * the next trip through ofproto_run(), but this interface is more direct.)
1255 * Returns 0 if successful, otherwise a positive errno. */
1257 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1259 struct ofport *ofport = get_port(ofproto, odp_port);
1260 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1263 error = dpif_port_del(ofproto->dpif, odp_port);
1265 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1266 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1267 } else if (ofport) {
1268 /* 'name' is ofport->opp.name and update_port() is going to destroy
1269 * 'ofport'. Just in case update_port() refers to 'name' after it
1270 * destroys 'ofport', make a copy of it around the update_port()
1272 char *devname = xstrdup(name);
1273 update_port(ofproto, devname);
1279 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1280 * true if 'odp_port' exists and should be included, false otherwise. */
1282 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1284 struct ofport *ofport = get_port(ofproto, odp_port);
1285 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1289 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1290 const union ofp_action *actions, size_t n_actions,
1291 const struct ofpbuf *packet)
1293 struct odp_actions odp_actions;
1296 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1302 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1304 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
1309 ofproto_add_flow(struct ofproto *p, const struct flow *flow,
1310 uint32_t wildcards, unsigned int priority,
1311 const union ofp_action *actions, size_t n_actions,
1315 rule = rule_create(p, NULL, actions, n_actions,
1316 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1318 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1319 rule_insert(p, rule, NULL, 0);
1323 ofproto_delete_flow(struct ofproto *ofproto, const struct flow *flow,
1324 uint32_t wildcards, unsigned int priority)
1328 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1332 rule_remove(ofproto, rule);
1337 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1339 struct rule *rule = rule_from_cls_rule(rule_);
1340 struct ofproto *ofproto = ofproto_;
1342 /* Mark the flow as not installed, even though it might really be
1343 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1344 * There is no point in uninstalling it individually since we are about to
1345 * blow away all the flows with dpif_flow_flush(). */
1346 rule->installed = false;
1348 rule_remove(ofproto, rule);
1352 ofproto_flush_flows(struct ofproto *ofproto)
1354 COVERAGE_INC(ofproto_flush);
1355 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1356 dpif_flow_flush(ofproto->dpif);
1357 if (ofproto->in_band) {
1358 in_band_flushed(ofproto->in_band);
1360 if (ofproto->fail_open) {
1361 fail_open_flushed(ofproto->fail_open);
1366 reinit_ports(struct ofproto *p)
1368 struct svec devnames;
1369 struct ofport *ofport;
1370 struct odp_port *odp_ports;
1374 COVERAGE_INC(ofproto_reinit_ports);
1376 svec_init(&devnames);
1377 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1378 svec_add (&devnames, (char *) ofport->opp.name);
1380 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1381 for (i = 0; i < n_odp_ports; i++) {
1382 svec_add (&devnames, odp_ports[i].devname);
1386 svec_sort_unique(&devnames);
1387 for (i = 0; i < devnames.n; i++) {
1388 update_port(p, devnames.names[i]);
1390 svec_destroy(&devnames);
1393 static struct ofport *
1394 make_ofport(const struct odp_port *odp_port)
1396 struct netdev_options netdev_options;
1397 enum netdev_flags flags;
1398 struct ofport *ofport;
1399 struct netdev *netdev;
1402 memset(&netdev_options, 0, sizeof netdev_options);
1403 netdev_options.name = odp_port->devname;
1404 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1406 error = netdev_open(&netdev_options, &netdev);
1408 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1409 "cannot be opened (%s)",
1410 odp_port->devname, odp_port->port,
1411 odp_port->devname, strerror(error));
1415 ofport = xmalloc(sizeof *ofport);
1416 ofport->netdev = netdev;
1417 ofport->odp_port = odp_port->port;
1418 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1419 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1420 memcpy(ofport->opp.name, odp_port->devname,
1421 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1422 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1424 netdev_get_flags(netdev, &flags);
1425 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1427 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1429 netdev_get_features(netdev,
1430 &ofport->opp.curr, &ofport->opp.advertised,
1431 &ofport->opp.supported, &ofport->opp.peer);
1436 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1438 if (get_port(p, odp_port->port)) {
1439 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1442 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1443 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1452 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1454 const struct ofp_phy_port *a = &a_->opp;
1455 const struct ofp_phy_port *b = &b_->opp;
1457 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1458 return (a->port_no == b->port_no
1459 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1460 && !strcmp((char *) a->name, (char *) b->name)
1461 && a->state == b->state
1462 && a->config == b->config
1463 && a->curr == b->curr
1464 && a->advertised == b->advertised
1465 && a->supported == b->supported
1466 && a->peer == b->peer);
1470 send_port_status(struct ofproto *p, const struct ofport *ofport,
1473 /* XXX Should limit the number of queued port status change messages. */
1474 struct ofconn *ofconn;
1475 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1476 struct ofp_port_status *ops;
1479 if (!ofconn_receives_async_msgs(ofconn)) {
1483 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1484 ops->reason = reason;
1485 ops->desc = ofport->opp;
1486 hton_ofp_phy_port(&ops->desc);
1487 queue_tx(b, ofconn, NULL);
1492 ofport_install(struct ofproto *p, struct ofport *ofport)
1494 const char *netdev_name = (const char *) ofport->opp.name;
1496 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1497 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1498 shash_add(&p->port_by_name, netdev_name, ofport);
1500 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1505 ofport_remove(struct ofproto *p, struct ofport *ofport)
1507 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1508 hmap_remove(&p->ports, &ofport->hmap_node);
1509 shash_delete(&p->port_by_name,
1510 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1512 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1517 ofport_free(struct ofport *ofport)
1520 netdev_close(ofport->netdev);
1525 static struct ofport *
1526 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1528 struct ofport *port;
1530 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1531 hash_int(odp_port, 0), &ofproto->ports) {
1532 if (port->odp_port == odp_port) {
1540 update_port(struct ofproto *p, const char *devname)
1542 struct odp_port odp_port;
1543 struct ofport *old_ofport;
1544 struct ofport *new_ofport;
1547 COVERAGE_INC(ofproto_update_port);
1549 /* Query the datapath for port information. */
1550 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1552 /* Find the old ofport. */
1553 old_ofport = shash_find_data(&p->port_by_name, devname);
1556 /* There's no port named 'devname' but there might be a port with
1557 * the same port number. This could happen if a port is deleted
1558 * and then a new one added in its place very quickly, or if a port
1559 * is renamed. In the former case we want to send an OFPPR_DELETE
1560 * and an OFPPR_ADD, and in the latter case we want to send a
1561 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1562 * the old port's ifindex against the new port, or perhaps less
1563 * reliably but more portably by comparing the old port's MAC
1564 * against the new port's MAC. However, this code isn't that smart
1565 * and always sends an OFPPR_MODIFY (XXX). */
1566 old_ofport = get_port(p, odp_port.port);
1568 } else if (error != ENOENT && error != ENODEV) {
1569 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1570 "%s", strerror(error));
1574 /* Create a new ofport. */
1575 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1577 /* Eliminate a few pathological cases. */
1578 if (!old_ofport && !new_ofport) {
1580 } else if (old_ofport && new_ofport) {
1581 /* Most of the 'config' bits are OpenFlow soft state, but
1582 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1583 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1584 * leaves the other bits 0.) */
1585 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1587 if (ofport_equal(old_ofport, new_ofport)) {
1588 /* False alarm--no change. */
1589 ofport_free(new_ofport);
1594 /* Now deal with the normal cases. */
1596 ofport_remove(p, old_ofport);
1599 ofport_install(p, new_ofport);
1601 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1602 (!old_ofport ? OFPPR_ADD
1603 : !new_ofport ? OFPPR_DELETE
1605 ofport_free(old_ofport);
1609 init_ports(struct ofproto *p)
1611 struct odp_port *ports;
1616 error = dpif_port_list(p->dpif, &ports, &n_ports);
1621 for (i = 0; i < n_ports; i++) {
1622 const struct odp_port *odp_port = &ports[i];
1623 if (!ofport_conflicts(p, odp_port)) {
1624 struct ofport *ofport = make_ofport(odp_port);
1626 ofport_install(p, ofport);
1634 static struct ofconn *
1635 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1637 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1638 ofconn->ofproto = p;
1639 list_push_back(&p->all_conns, &ofconn->node);
1640 ofconn->rconn = rconn;
1641 ofconn->type = type;
1642 ofconn->role = NX_ROLE_OTHER;
1643 ofconn->packet_in_counter = rconn_packet_counter_create ();
1644 ofconn->pktbuf = NULL;
1645 ofconn->miss_send_len = 0;
1646 ofconn->reply_counter = rconn_packet_counter_create ();
1651 ofconn_destroy(struct ofconn *ofconn)
1653 if (ofconn->type == OFCONN_PRIMARY) {
1654 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1656 discovery_destroy(ofconn->discovery);
1658 list_remove(&ofconn->node);
1659 switch_status_unregister(ofconn->ss);
1660 rconn_destroy(ofconn->rconn);
1661 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1662 rconn_packet_counter_destroy(ofconn->reply_counter);
1663 pktbuf_destroy(ofconn->pktbuf);
1668 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1673 if (ofconn->discovery) {
1674 char *controller_name;
1675 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1676 discovery_question_connectivity(ofconn->discovery);
1678 if (discovery_run(ofconn->discovery, &controller_name)) {
1679 if (controller_name) {
1680 char *ofconn_name = ofconn_make_name(p, controller_name);
1681 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1684 rconn_disconnect(ofconn->rconn);
1689 for (i = 0; i < N_SCHEDULERS; i++) {
1690 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1693 rconn_run(ofconn->rconn);
1695 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1696 /* Limit the number of iterations to prevent other tasks from
1698 for (iteration = 0; iteration < 50; iteration++) {
1699 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1704 fail_open_maybe_recover(p->fail_open);
1706 handle_openflow(ofconn, p, of_msg);
1707 ofpbuf_delete(of_msg);
1711 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1712 ofconn_destroy(ofconn);
1717 ofconn_wait(struct ofconn *ofconn)
1721 if (ofconn->discovery) {
1722 discovery_wait(ofconn->discovery);
1724 for (i = 0; i < N_SCHEDULERS; i++) {
1725 pinsched_wait(ofconn->schedulers[i]);
1727 rconn_run_wait(ofconn->rconn);
1728 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1729 rconn_recv_wait(ofconn->rconn);
1731 COVERAGE_INC(ofproto_ofconn_stuck);
1735 /* Returns true if 'ofconn' should receive asynchronous messages. */
1737 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1739 if (ofconn->type == OFCONN_PRIMARY) {
1740 /* Primary controllers always get asynchronous messages unless they
1741 * have configured themselves as "slaves". */
1742 return ofconn->role != NX_ROLE_SLAVE;
1744 /* Service connections don't get asynchronous messages unless they have
1745 * explicitly asked for them by setting a nonzero miss send length. */
1746 return ofconn->miss_send_len > 0;
1750 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1751 * and 'target', suitable for use in log messages for identifying the
1754 * The name is dynamically allocated. The caller should free it (with free())
1755 * when it is no longer needed. */
1757 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1759 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1763 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1767 for (i = 0; i < N_SCHEDULERS; i++) {
1768 struct pinsched **s = &ofconn->schedulers[i];
1772 *s = pinsched_create(rate, burst,
1773 ofconn->ofproto->switch_status);
1775 pinsched_set_limits(*s, rate, burst);
1778 pinsched_destroy(*s);
1785 ofservice_reconfigure(struct ofservice *ofservice,
1786 const struct ofproto_controller *c)
1788 ofservice->probe_interval = c->probe_interval;
1789 ofservice->rate_limit = c->rate_limit;
1790 ofservice->burst_limit = c->burst_limit;
1793 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1794 * positive errno value. */
1796 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1798 struct ofservice *ofservice;
1799 struct pvconn *pvconn;
1802 error = pvconn_open(c->target, &pvconn);
1807 ofservice = xzalloc(sizeof *ofservice);
1808 hmap_insert(&ofproto->services, &ofservice->node,
1809 hash_string(c->target, 0));
1810 ofservice->pvconn = pvconn;
1812 ofservice_reconfigure(ofservice, c);
1818 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1820 hmap_remove(&ofproto->services, &ofservice->node);
1821 pvconn_close(ofservice->pvconn);
1825 /* Finds and returns the ofservice within 'ofproto' that has the given
1826 * 'target', or a null pointer if none exists. */
1827 static struct ofservice *
1828 ofservice_lookup(struct ofproto *ofproto, const char *target)
1830 struct ofservice *ofservice;
1832 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1833 &ofproto->services) {
1834 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1841 /* Caller is responsible for initializing the 'cr' member of the returned
1843 static struct rule *
1844 rule_create(struct ofproto *ofproto, struct rule *super,
1845 const union ofp_action *actions, size_t n_actions,
1846 uint16_t idle_timeout, uint16_t hard_timeout,
1847 ovs_be64 flow_cookie, bool send_flow_removed)
1849 struct rule *rule = xzalloc(sizeof *rule);
1850 rule->idle_timeout = idle_timeout;
1851 rule->hard_timeout = hard_timeout;
1852 rule->flow_cookie = flow_cookie;
1853 rule->used = rule->created = time_msec();
1854 rule->send_flow_removed = send_flow_removed;
1855 rule->super = super;
1857 list_push_back(&super->list, &rule->list);
1859 list_init(&rule->list);
1861 if (n_actions > 0) {
1862 rule->n_actions = n_actions;
1863 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1865 netflow_flow_clear(&rule->nf_flow);
1866 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1871 static struct rule *
1872 rule_from_cls_rule(const struct cls_rule *cls_rule)
1874 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1878 rule_free(struct rule *rule)
1880 free(rule->actions);
1881 free(rule->odp_actions);
1885 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1886 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1887 * through all of its subrules and revalidates them, destroying any that no
1888 * longer has a super-rule (which is probably all of them).
1890 * Before calling this function, the caller must make have removed 'rule' from
1891 * the classifier. If 'rule' is an exact-match rule, the caller is also
1892 * responsible for ensuring that it has been uninstalled from the datapath. */
1894 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1897 struct rule *subrule, *next;
1898 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
1899 revalidate_rule(ofproto, subrule);
1902 list_remove(&rule->list);
1908 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1910 const union ofp_action *oa;
1911 struct actions_iterator i;
1913 if (out_port == htons(OFPP_NONE)) {
1916 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1917 oa = actions_next(&i)) {
1918 if (action_outputs_to_port(oa, out_port)) {
1925 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1926 * 'packet', which arrived on 'in_port'.
1928 * Takes ownership of 'packet'. */
1930 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1931 const union odp_action *actions, size_t n_actions,
1932 struct ofpbuf *packet)
1934 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1935 /* As an optimization, avoid a round-trip from userspace to kernel to
1936 * userspace. This also avoids possibly filling up kernel packet
1937 * buffers along the way. */
1938 struct odp_msg *msg;
1940 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1941 msg->type = _ODPL_ACTION_NR;
1942 msg->length = sizeof(struct odp_msg) + packet->size;
1943 msg->port = in_port;
1945 msg->arg = actions[0].controller.arg;
1947 send_packet_in(ofproto, packet);
1953 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
1954 ofpbuf_delete(packet);
1959 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1960 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1961 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1963 * The flow that 'packet' actually contains does not need to actually match
1964 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1965 * the packet and byte counters for 'rule' will be credited for the packet sent
1966 * out whether or not the packet actually matches 'rule'.
1968 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1969 * the caller must already have accurately composed ODP actions for it given
1970 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1971 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1972 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1973 * actions and apply them to 'packet'.
1975 * Takes ownership of 'packet'. */
1977 rule_execute(struct ofproto *ofproto, struct rule *rule,
1978 struct ofpbuf *packet, const struct flow *flow)
1980 const union odp_action *actions;
1981 struct odp_flow_stats stats;
1983 struct odp_actions a;
1985 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1987 /* Grab or compose the ODP actions.
1989 * The special case for an exact-match 'rule' where 'flow' is not the
1990 * rule's flow is important to avoid, e.g., sending a packet out its input
1991 * port simply because the ODP actions were composed for the wrong
1993 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1994 struct rule *super = rule->super ? rule->super : rule;
1995 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1996 packet, &a, NULL, 0, NULL)) {
1997 ofpbuf_delete(packet);
2000 actions = a.actions;
2001 n_actions = a.n_actions;
2003 actions = rule->odp_actions;
2004 n_actions = rule->n_odp_actions;
2007 /* Execute the ODP actions. */
2008 flow_extract_stats(flow, packet, &stats);
2009 if (execute_odp_actions(ofproto, flow->in_port,
2010 actions, n_actions, packet)) {
2011 update_stats(ofproto, rule, &stats);
2012 rule->used = time_msec();
2013 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2017 /* Inserts 'rule' into 'p''s flow table.
2019 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2020 * actions on it and credits the statistics for sending the packet to 'rule'.
2021 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2024 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2027 struct rule *displaced_rule;
2029 /* Insert the rule in the classifier. */
2030 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2031 if (!rule->cr.wc.wildcards) {
2032 rule_make_actions(p, rule, packet);
2035 /* Send the packet and credit it to the rule. */
2038 flow_extract(packet, 0, in_port, &flow);
2039 rule_execute(p, rule, packet, &flow);
2042 /* Install the rule in the datapath only after sending the packet, to
2043 * avoid packet reordering. */
2044 if (rule->cr.wc.wildcards) {
2045 COVERAGE_INC(ofproto_add_wc_flow);
2046 p->need_revalidate = true;
2048 rule_install(p, rule, displaced_rule);
2051 /* Free the rule that was displaced, if any. */
2052 if (displaced_rule) {
2053 rule_destroy(p, displaced_rule);
2057 static struct rule *
2058 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2059 const struct flow *flow)
2061 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2062 rule->idle_timeout, rule->hard_timeout,
2064 COVERAGE_INC(ofproto_subrule_create);
2065 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2066 : rule->cr.priority), &subrule->cr);
2068 if (classifier_insert(&ofproto->cls, &subrule->cr)) {
2076 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2078 * - If 'rule' was installed in the datapath, uninstalls it and updates
2079 * 'rule''s statistics (or its super-rule's statistics, if it is a
2080 * subrule), via rule_uninstall().
2082 * - Removes 'rule' from the classifier.
2084 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2085 * uninstalls and destroys) its subrules, via rule_destroy().
2088 rule_remove(struct ofproto *ofproto, struct rule *rule)
2090 if (rule->cr.wc.wildcards) {
2091 COVERAGE_INC(ofproto_del_wc_flow);
2092 ofproto->need_revalidate = true;
2094 rule_uninstall(ofproto, rule);
2096 classifier_remove(&ofproto->cls, &rule->cr);
2097 rule_destroy(ofproto, rule);
2100 /* Returns true if the actions changed, false otherwise. */
2102 rule_make_actions(struct ofproto *p, struct rule *rule,
2103 const struct ofpbuf *packet)
2105 const struct rule *super;
2106 struct odp_actions a;
2109 assert(!rule->cr.wc.wildcards);
2111 super = rule->super ? rule->super : rule;
2113 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2114 packet, &a, &rule->tags, &rule->may_install,
2115 &rule->nf_flow.output_iface);
2117 actions_len = a.n_actions * sizeof *a.actions;
2118 if (rule->n_odp_actions != a.n_actions
2119 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2120 COVERAGE_INC(ofproto_odp_unchanged);
2121 free(rule->odp_actions);
2122 rule->n_odp_actions = a.n_actions;
2123 rule->odp_actions = xmemdup(a.actions, actions_len);
2131 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2132 struct odp_flow_put *put)
2134 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2135 odp_flow_key_from_flow(&put->flow.key, &rule->cr.flow);
2136 put->flow.actions = rule->odp_actions;
2137 put->flow.n_actions = rule->n_odp_actions;
2138 put->flow.flags = 0;
2140 return dpif_flow_put(ofproto->dpif, put);
2144 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2146 assert(!rule->cr.wc.wildcards);
2148 if (rule->may_install) {
2149 struct odp_flow_put put;
2150 if (!do_put_flow(p, rule,
2151 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2153 rule->installed = true;
2154 if (displaced_rule) {
2155 update_stats(p, displaced_rule, &put.flow.stats);
2156 rule_post_uninstall(p, displaced_rule);
2159 } else if (displaced_rule) {
2160 rule_uninstall(p, displaced_rule);
2165 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2167 if (rule->installed) {
2168 struct odp_flow_put put;
2169 COVERAGE_INC(ofproto_dp_missed);
2170 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2172 rule_install(ofproto, rule, NULL);
2177 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2179 bool actions_changed;
2180 uint16_t new_out_iface, old_out_iface;
2182 old_out_iface = rule->nf_flow.output_iface;
2183 actions_changed = rule_make_actions(ofproto, rule, NULL);
2185 if (rule->may_install) {
2186 if (rule->installed) {
2187 if (actions_changed) {
2188 struct odp_flow_put put;
2189 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2190 | ODPPF_ZERO_STATS, &put);
2191 update_stats(ofproto, rule, &put.flow.stats);
2193 /* Temporarily set the old output iface so that NetFlow
2194 * messages have the correct output interface for the old
2196 new_out_iface = rule->nf_flow.output_iface;
2197 rule->nf_flow.output_iface = old_out_iface;
2198 rule_post_uninstall(ofproto, rule);
2199 rule->nf_flow.output_iface = new_out_iface;
2202 rule_install(ofproto, rule, NULL);
2205 rule_uninstall(ofproto, rule);
2210 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2212 uint64_t total_bytes = rule->byte_count + extra_bytes;
2214 if (ofproto->ofhooks->account_flow_cb
2215 && total_bytes > rule->accounted_bytes)
2217 ofproto->ofhooks->account_flow_cb(
2218 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2219 total_bytes - rule->accounted_bytes, ofproto->aux);
2220 rule->accounted_bytes = total_bytes;
2224 /* 'rule' must be an exact-match rule in 'p'.
2226 * If 'rule' is installed in the datapath, uninstalls it and updates's
2227 * statistics. If 'rule' is a subrule, the statistics that are updated are
2228 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2231 * If 'rule' is not installed, this function has no effect. */
2233 rule_uninstall(struct ofproto *p, struct rule *rule)
2235 assert(!rule->cr.wc.wildcards);
2236 if (rule->installed) {
2237 struct odp_flow odp_flow;
2239 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
2240 odp_flow.actions = NULL;
2241 odp_flow.n_actions = 0;
2243 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2244 update_stats(p, rule, &odp_flow.stats);
2246 rule->installed = false;
2248 rule_post_uninstall(p, rule);
2253 is_controller_rule(struct rule *rule)
2255 /* If the only action is send to the controller then don't report
2256 * NetFlow expiration messages since it is just part of the control
2257 * logic for the network and not real traffic. */
2261 && rule->super->n_actions == 1
2262 && action_outputs_to_port(&rule->super->actions[0],
2263 htons(OFPP_CONTROLLER)));
2267 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2269 struct rule *super = rule->super;
2271 rule_account(ofproto, rule, 0);
2273 if (ofproto->netflow && !is_controller_rule(rule)) {
2274 struct ofexpired expired;
2275 expired.flow = rule->cr.flow;
2276 expired.packet_count = rule->packet_count;
2277 expired.byte_count = rule->byte_count;
2278 expired.used = rule->used;
2279 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2282 super->packet_count += rule->packet_count;
2283 super->byte_count += rule->byte_count;
2285 /* Reset counters to prevent double counting if the rule ever gets
2287 rule->packet_count = 0;
2288 rule->byte_count = 0;
2289 rule->accounted_bytes = 0;
2291 netflow_flow_clear(&rule->nf_flow);
2296 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2297 struct rconn_packet_counter *counter)
2299 update_openflow_length(msg);
2300 if (rconn_send(ofconn->rconn, msg, counter)) {
2306 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2307 int error, const void *data, size_t len)
2310 struct ofp_error_msg *oem;
2312 if (!(error >> 16)) {
2313 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2318 COVERAGE_INC(ofproto_error);
2319 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2320 oh ? oh->xid : 0, &buf);
2321 oem->type = htons((unsigned int) error >> 16);
2322 oem->code = htons(error & 0xffff);
2323 memcpy(oem->data, data, len);
2324 queue_tx(buf, ofconn, ofconn->reply_counter);
2328 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2331 size_t oh_length = ntohs(oh->length);
2332 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2336 hton_ofp_phy_port(struct ofp_phy_port *opp)
2338 opp->port_no = htons(opp->port_no);
2339 opp->config = htonl(opp->config);
2340 opp->state = htonl(opp->state);
2341 opp->curr = htonl(opp->curr);
2342 opp->advertised = htonl(opp->advertised);
2343 opp->supported = htonl(opp->supported);
2344 opp->peer = htonl(opp->peer);
2348 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2350 struct ofp_header *rq = oh;
2351 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2356 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2357 struct ofp_header *oh)
2359 struct ofp_switch_features *osf;
2361 struct ofport *port;
2363 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2364 osf->datapath_id = htonll(p->datapath_id);
2365 osf->n_buffers = htonl(pktbuf_capacity());
2367 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2368 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2369 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2370 (1u << OFPAT_SET_VLAN_VID) |
2371 (1u << OFPAT_SET_VLAN_PCP) |
2372 (1u << OFPAT_STRIP_VLAN) |
2373 (1u << OFPAT_SET_DL_SRC) |
2374 (1u << OFPAT_SET_DL_DST) |
2375 (1u << OFPAT_SET_NW_SRC) |
2376 (1u << OFPAT_SET_NW_DST) |
2377 (1u << OFPAT_SET_NW_TOS) |
2378 (1u << OFPAT_SET_TP_SRC) |
2379 (1u << OFPAT_SET_TP_DST) |
2380 (1u << OFPAT_ENQUEUE));
2382 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2383 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2386 queue_tx(buf, ofconn, ofconn->reply_counter);
2391 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2392 struct ofp_header *oh)
2395 struct ofp_switch_config *osc;
2399 /* Figure out flags. */
2400 dpif_get_drop_frags(p->dpif, &drop_frags);
2401 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2404 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2405 osc->flags = htons(flags);
2406 osc->miss_send_len = htons(ofconn->miss_send_len);
2407 queue_tx(buf, ofconn, ofconn->reply_counter);
2413 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2414 struct ofp_switch_config *osc)
2419 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2423 flags = ntohs(osc->flags);
2425 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2426 switch (flags & OFPC_FRAG_MASK) {
2427 case OFPC_FRAG_NORMAL:
2428 dpif_set_drop_frags(p->dpif, false);
2430 case OFPC_FRAG_DROP:
2431 dpif_set_drop_frags(p->dpif, true);
2434 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2440 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2446 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2448 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2449 a->controller.arg = max_len;
2452 struct action_xlate_ctx {
2454 struct flow flow; /* Flow to which these actions correspond. */
2455 int recurse; /* Recursion level, via xlate_table_action. */
2456 struct ofproto *ofproto;
2457 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2458 * null pointer if we are revalidating
2459 * without a packet to refer to. */
2462 struct odp_actions *out; /* Datapath actions. */
2463 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2464 bool may_set_up_flow; /* True ordinarily; false if the actions must
2465 * be reassessed for every packet. */
2466 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2469 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2470 * flow translation. */
2471 #define MAX_RESUBMIT_RECURSION 8
2473 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2474 struct action_xlate_ctx *ctx);
2477 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2479 const struct ofport *ofport = get_port(ctx->ofproto, port);
2482 if (ofport->opp.config & OFPPC_NO_FWD) {
2483 /* Forwarding disabled on port. */
2488 * We don't have an ofport record for this port, but it doesn't hurt to
2489 * allow forwarding to it anyhow. Maybe such a port will appear later
2490 * and we're pre-populating the flow table.
2494 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2495 ctx->nf_output_iface = port;
2498 static struct rule *
2499 lookup_valid_rule(struct ofproto *ofproto, const struct flow *flow)
2502 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2504 /* The rule we found might not be valid, since we could be in need of
2505 * revalidation. If it is not valid, don't return it. */
2508 && ofproto->need_revalidate
2509 && !revalidate_rule(ofproto, rule)) {
2510 COVERAGE_INC(ofproto_invalidated);
2518 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2520 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2521 uint16_t old_in_port;
2524 /* Look up a flow with 'in_port' as the input port. Then restore the
2525 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2526 * have surprising behavior). */
2527 old_in_port = ctx->flow.in_port;
2528 ctx->flow.in_port = in_port;
2529 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2530 ctx->flow.in_port = old_in_port;
2538 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2542 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2544 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2545 MAX_RESUBMIT_RECURSION);
2550 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2551 uint16_t *nf_output_iface, struct odp_actions *actions)
2553 struct ofport *ofport;
2555 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2556 uint16_t odp_port = ofport->odp_port;
2557 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2558 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2561 *nf_output_iface = NF_OUT_FLOOD;
2565 xlate_output_action__(struct action_xlate_ctx *ctx,
2566 uint16_t port, uint16_t max_len)
2569 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2571 ctx->nf_output_iface = NF_OUT_DROP;
2575 add_output_action(ctx, ctx->flow.in_port);
2578 xlate_table_action(ctx, ctx->flow.in_port);
2581 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2582 ctx->out, ctx->tags,
2583 &ctx->nf_output_iface,
2584 ctx->ofproto->aux)) {
2585 COVERAGE_INC(ofproto_uninstallable);
2586 ctx->may_set_up_flow = false;
2590 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2591 &ctx->nf_output_iface, ctx->out);
2594 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2595 &ctx->nf_output_iface, ctx->out);
2597 case OFPP_CONTROLLER:
2598 add_controller_action(ctx->out, max_len);
2601 add_output_action(ctx, ODPP_LOCAL);
2604 odp_port = ofp_port_to_odp_port(port);
2605 if (odp_port != ctx->flow.in_port) {
2606 add_output_action(ctx, odp_port);
2611 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2612 ctx->nf_output_iface = NF_OUT_FLOOD;
2613 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2614 ctx->nf_output_iface = prev_nf_output_iface;
2615 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2616 ctx->nf_output_iface != NF_OUT_FLOOD) {
2617 ctx->nf_output_iface = NF_OUT_MULTI;
2622 xlate_output_action(struct action_xlate_ctx *ctx,
2623 const struct ofp_action_output *oao)
2625 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2628 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2629 * optimization, because we're going to add another action that sets the
2630 * priority immediately after, or because there are no actions following the
2633 remove_pop_action(struct action_xlate_ctx *ctx)
2635 size_t n = ctx->out->n_actions;
2636 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2637 ctx->out->n_actions--;
2642 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2643 const struct ofp_action_enqueue *oae)
2645 uint16_t ofp_port, odp_port;
2649 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2652 /* Fall back to ordinary output action. */
2653 xlate_output_action__(ctx, ntohs(oae->port), 0);
2657 /* Figure out ODP output port. */
2658 ofp_port = ntohs(oae->port);
2659 if (ofp_port != OFPP_IN_PORT) {
2660 odp_port = ofp_port_to_odp_port(ofp_port);
2662 odp_port = ctx->flow.in_port;
2665 /* Add ODP actions. */
2666 remove_pop_action(ctx);
2667 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2669 add_output_action(ctx, odp_port);
2670 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2672 /* Update NetFlow output port. */
2673 if (ctx->nf_output_iface == NF_OUT_DROP) {
2674 ctx->nf_output_iface = odp_port;
2675 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2676 ctx->nf_output_iface = NF_OUT_MULTI;
2681 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2682 const struct nx_action_set_queue *nasq)
2687 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2690 /* Couldn't translate queue to a priority, so ignore. A warning
2691 * has already been logged. */
2695 remove_pop_action(ctx);
2696 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2701 xlate_nicira_action(struct action_xlate_ctx *ctx,
2702 const struct nx_action_header *nah)
2704 const struct nx_action_resubmit *nar;
2705 const struct nx_action_set_tunnel *nast;
2706 const struct nx_action_set_queue *nasq;
2707 union odp_action *oa;
2708 int subtype = ntohs(nah->subtype);
2710 assert(nah->vendor == htonl(NX_VENDOR_ID));
2712 case NXAST_RESUBMIT:
2713 nar = (const struct nx_action_resubmit *) nah;
2714 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2717 case NXAST_SET_TUNNEL:
2718 nast = (const struct nx_action_set_tunnel *) nah;
2719 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2720 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2723 case NXAST_DROP_SPOOFED_ARP:
2724 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2725 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2729 case NXAST_SET_QUEUE:
2730 nasq = (const struct nx_action_set_queue *) nah;
2731 xlate_set_queue_action(ctx, nasq);
2734 case NXAST_POP_QUEUE:
2735 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2738 /* If you add a new action here that modifies flow data, don't forget to
2739 * update the flow key in ctx->flow at the same time. */
2742 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2748 do_xlate_actions(const union ofp_action *in, size_t n_in,
2749 struct action_xlate_ctx *ctx)
2751 struct actions_iterator iter;
2752 const union ofp_action *ia;
2753 const struct ofport *port;
2755 port = get_port(ctx->ofproto, ctx->flow.in_port);
2756 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2757 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2758 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2759 /* Drop this flow. */
2763 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2764 uint16_t type = ntohs(ia->type);
2765 union odp_action *oa;
2769 xlate_output_action(ctx, &ia->output);
2772 case OFPAT_SET_VLAN_VID:
2773 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2774 oa->dl_tci.tci = ia->vlan_vid.vlan_vid;
2775 oa->dl_tci.tci |= htons(ctx->flow.dl_vlan_pcp << VLAN_PCP_SHIFT);
2776 ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
2779 case OFPAT_SET_VLAN_PCP:
2780 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2781 oa->dl_tci.tci = htons(ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT);
2782 oa->dl_tci.tci |= ctx->flow.dl_vlan;
2783 ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
2786 case OFPAT_STRIP_VLAN:
2787 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2788 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2789 ctx->flow.dl_vlan_pcp = 0;
2792 case OFPAT_SET_DL_SRC:
2793 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2794 memcpy(oa->dl_addr.dl_addr,
2795 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2796 memcpy(ctx->flow.dl_src,
2797 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2800 case OFPAT_SET_DL_DST:
2801 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2802 memcpy(oa->dl_addr.dl_addr,
2803 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2804 memcpy(ctx->flow.dl_dst,
2805 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2808 case OFPAT_SET_NW_SRC:
2809 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2810 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2813 case OFPAT_SET_NW_DST:
2814 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2815 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2818 case OFPAT_SET_NW_TOS:
2819 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2820 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2823 case OFPAT_SET_TP_SRC:
2824 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2825 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2828 case OFPAT_SET_TP_DST:
2829 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2830 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2834 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2838 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2842 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2849 xlate_actions(const union ofp_action *in, size_t n_in,
2850 const struct flow *flow, struct ofproto *ofproto,
2851 const struct ofpbuf *packet,
2852 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2853 uint16_t *nf_output_iface)
2855 tag_type no_tags = 0;
2856 struct action_xlate_ctx ctx;
2857 COVERAGE_INC(ofproto_ofp2odp);
2858 odp_actions_init(out);
2861 ctx.ofproto = ofproto;
2862 ctx.packet = packet;
2864 ctx.tags = tags ? tags : &no_tags;
2865 ctx.may_set_up_flow = true;
2866 ctx.nf_output_iface = NF_OUT_DROP;
2867 do_xlate_actions(in, n_in, &ctx);
2868 remove_pop_action(&ctx);
2870 /* Check with in-band control to see if we're allowed to set up this
2872 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2873 ctx.may_set_up_flow = false;
2876 if (may_set_up_flow) {
2877 *may_set_up_flow = ctx.may_set_up_flow;
2879 if (nf_output_iface) {
2880 *nf_output_iface = ctx.nf_output_iface;
2882 if (odp_actions_overflow(out)) {
2883 COVERAGE_INC(odp_overflow);
2884 odp_actions_init(out);
2885 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2890 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2891 * error message code (composed with ofp_mkerr()) for the caller to propagate
2892 * upward. Otherwise, returns 0.
2894 * 'oh' is used to make log messages more informative. */
2896 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2898 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2899 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2902 type_name = ofp_message_type_to_string(oh->type);
2903 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2907 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2914 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2915 struct ofp_header *oh)
2917 struct ofp_packet_out *opo;
2918 struct ofpbuf payload, *buffer;
2919 struct odp_actions actions;
2925 error = reject_slave_controller(ofconn, oh);
2930 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2934 opo = (struct ofp_packet_out *) oh;
2936 COVERAGE_INC(ofproto_packet_out);
2937 if (opo->buffer_id != htonl(UINT32_MAX)) {
2938 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2940 if (error || !buffer) {
2948 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2949 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2950 &flow, p, &payload, &actions, NULL, NULL, NULL);
2952 dpif_execute(p->dpif, actions.actions, actions.n_actions, &payload);
2954 ofpbuf_delete(buffer);
2960 update_port_config(struct ofproto *p, struct ofport *port,
2961 uint32_t config, uint32_t mask)
2963 mask &= config ^ port->opp.config;
2964 if (mask & OFPPC_PORT_DOWN) {
2965 if (config & OFPPC_PORT_DOWN) {
2966 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2968 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2971 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2972 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2973 if (mask & REVALIDATE_BITS) {
2974 COVERAGE_INC(ofproto_costly_flags);
2975 port->opp.config ^= mask & REVALIDATE_BITS;
2976 p->need_revalidate = true;
2978 #undef REVALIDATE_BITS
2979 if (mask & OFPPC_NO_PACKET_IN) {
2980 port->opp.config ^= OFPPC_NO_PACKET_IN;
2985 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2986 struct ofp_header *oh)
2988 const struct ofp_port_mod *opm;
2989 struct ofport *port;
2992 error = reject_slave_controller(ofconn, oh);
2996 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
3000 opm = (struct ofp_port_mod *) oh;
3002 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3004 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3005 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3006 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3008 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3009 if (opm->advertise) {
3010 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3016 static struct ofpbuf *
3017 make_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3019 struct ofp_stats_reply *osr;
3022 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3023 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3025 osr->flags = htons(0);
3029 static struct ofpbuf *
3030 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3032 return make_stats_reply(request->header.xid, request->type, body_len);
3036 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
3038 struct ofpbuf *msg = *msgp;
3039 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3040 if (nbytes + msg->size > UINT16_MAX) {
3041 struct ofp_stats_reply *reply = msg->data;
3042 reply->flags = htons(OFPSF_REPLY_MORE);
3043 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3044 queue_tx(msg, ofconn, ofconn->reply_counter);
3046 return ofpbuf_put_uninit(*msgp, nbytes);
3050 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3051 struct ofp_stats_request *request)
3053 struct ofp_desc_stats *ods;
3056 msg = start_stats_reply(request, sizeof *ods);
3057 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3058 memset(ods, 0, sizeof *ods);
3059 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3060 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3061 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3062 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3063 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3064 queue_tx(msg, ofconn, ofconn->reply_counter);
3070 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3071 struct ofp_stats_request *request)
3073 struct ofp_table_stats *ots;
3075 struct odp_stats dpstats;
3076 int n_exact, n_subrules, n_wild;
3079 msg = start_stats_reply(request, sizeof *ots * 2);
3081 /* Count rules of various kinds. */
3083 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) {
3088 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3089 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3092 dpif_get_dp_stats(p->dpif, &dpstats);
3093 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3094 memset(ots, 0, sizeof *ots);
3095 ots->table_id = TABLEID_HASH;
3096 strcpy(ots->name, "hash");
3097 ots->wildcards = htonl(0);
3098 ots->max_entries = htonl(dpstats.max_capacity);
3099 ots->active_count = htonl(n_exact);
3100 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3102 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3104 /* Classifier table. */
3105 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3106 memset(ots, 0, sizeof *ots);
3107 ots->table_id = TABLEID_CLASSIFIER;
3108 strcpy(ots->name, "classifier");
3109 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3111 ots->max_entries = htonl(65536);
3112 ots->active_count = htonl(n_wild);
3113 ots->lookup_count = htonll(0); /* XXX */
3114 ots->matched_count = htonll(0); /* XXX */
3116 queue_tx(msg, ofconn, ofconn->reply_counter);
3121 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3122 struct ofpbuf **msgp)
3124 struct netdev_stats stats;
3125 struct ofp_port_stats *ops;
3127 /* Intentionally ignore return value, since errors will set
3128 * 'stats' to all-1s, which is correct for OpenFlow, and
3129 * netdev_get_stats() will log errors. */
3130 netdev_get_stats(port->netdev, &stats);
3132 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3133 ops->port_no = htons(port->opp.port_no);
3134 memset(ops->pad, 0, sizeof ops->pad);
3135 ops->rx_packets = htonll(stats.rx_packets);
3136 ops->tx_packets = htonll(stats.tx_packets);
3137 ops->rx_bytes = htonll(stats.rx_bytes);
3138 ops->tx_bytes = htonll(stats.tx_bytes);
3139 ops->rx_dropped = htonll(stats.rx_dropped);
3140 ops->tx_dropped = htonll(stats.tx_dropped);
3141 ops->rx_errors = htonll(stats.rx_errors);
3142 ops->tx_errors = htonll(stats.tx_errors);
3143 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3144 ops->rx_over_err = htonll(stats.rx_over_errors);
3145 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3146 ops->collisions = htonll(stats.collisions);
3150 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3151 struct ofp_stats_request *osr,
3154 struct ofp_port_stats_request *psr;
3155 struct ofp_port_stats *ops;
3157 struct ofport *port;
3159 if (arg_size != sizeof *psr) {
3160 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3162 psr = (struct ofp_port_stats_request *) osr->body;
3164 msg = start_stats_reply(osr, sizeof *ops * 16);
3165 if (psr->port_no != htons(OFPP_NONE)) {
3166 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3168 append_port_stat(port, ofconn, &msg);
3171 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3172 append_port_stat(port, ofconn, &msg);
3176 queue_tx(msg, ofconn, ofconn->reply_counter);
3180 struct flow_stats_cbdata {
3181 struct ofproto *ofproto;
3182 struct ofconn *ofconn;
3187 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3188 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3189 * returned statistic include statistics for all of 'rule''s subrules. */
3191 query_stats(struct ofproto *p, struct rule *rule,
3192 uint64_t *packet_countp, uint64_t *byte_countp)
3194 uint64_t packet_count, byte_count;
3195 struct rule *subrule;
3196 struct odp_flow *odp_flows;
3199 /* Start from historical data for 'rule' itself that are no longer tracked
3200 * by the datapath. This counts, for example, subrules that have
3202 packet_count = rule->packet_count;
3203 byte_count = rule->byte_count;
3205 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3206 * wildcarded then on all of its subrules.
3208 * Also, add any statistics that are not tracked by the datapath for each
3209 * subrule. This includes, for example, statistics for packets that were
3210 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3212 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3213 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3214 if (rule->cr.wc.wildcards) {
3216 LIST_FOR_EACH (subrule, list, &rule->list) {
3217 odp_flow_key_from_flow(&odp_flows[i++].key, &subrule->cr.flow);
3218 packet_count += subrule->packet_count;
3219 byte_count += subrule->byte_count;
3222 odp_flow_key_from_flow(&odp_flows[0].key, &rule->cr.flow);
3225 /* Fetch up-to-date statistics from the datapath and add them in. */
3226 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3228 for (i = 0; i < n_odp_flows; i++) {
3229 struct odp_flow *odp_flow = &odp_flows[i];
3230 packet_count += odp_flow->stats.n_packets;
3231 byte_count += odp_flow->stats.n_bytes;
3236 /* Return the stats to the caller. */
3237 *packet_countp = packet_count;
3238 *byte_countp = byte_count;
3242 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3244 struct rule *rule = rule_from_cls_rule(rule_);
3245 struct flow_stats_cbdata *cbdata = cbdata_;
3246 struct ofp_flow_stats *ofs;
3247 uint64_t packet_count, byte_count;
3248 size_t act_len, len;
3249 long long int tdiff = time_msec() - rule->created;
3250 uint32_t sec = tdiff / 1000;
3251 uint32_t msec = tdiff - (sec * 1000);
3253 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3257 act_len = sizeof *rule->actions * rule->n_actions;
3258 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3260 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3262 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3263 ofs->length = htons(len);
3264 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3266 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3267 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3268 ofs->duration_sec = htonl(sec);
3269 ofs->duration_nsec = htonl(msec * 1000000);
3270 ofs->cookie = rule->flow_cookie;
3271 ofs->priority = htons(rule->cr.priority);
3272 ofs->idle_timeout = htons(rule->idle_timeout);
3273 ofs->hard_timeout = htons(rule->hard_timeout);
3274 memset(ofs->pad2, 0, sizeof ofs->pad2);
3275 ofs->packet_count = htonll(packet_count);
3276 ofs->byte_count = htonll(byte_count);
3277 if (rule->n_actions > 0) {
3278 memcpy(ofs->actions, rule->actions, act_len);
3283 table_id_to_include(uint8_t table_id)
3285 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3286 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3287 : table_id == 0xff ? CLS_INC_ALL
3292 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3293 const struct ofp_stats_request *osr,
3296 struct ofp_flow_stats_request *fsr;
3297 struct flow_stats_cbdata cbdata;
3298 struct cls_rule target;
3300 if (arg_size != sizeof *fsr) {
3301 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3303 fsr = (struct ofp_flow_stats_request *) osr->body;
3305 COVERAGE_INC(ofproto_flows_req);
3307 cbdata.ofconn = ofconn;
3308 cbdata.out_port = fsr->out_port;
3309 cbdata.msg = start_stats_reply(osr, 1024);
3310 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3311 classifier_for_each_match(&p->cls, &target,
3312 table_id_to_include(fsr->table_id),
3313 flow_stats_cb, &cbdata);
3314 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3318 struct flow_stats_ds_cbdata {
3319 struct ofproto *ofproto;
3324 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3326 struct rule *rule = rule_from_cls_rule(rule_);
3327 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3328 struct ds *results = cbdata->results;
3329 struct ofp_match match;
3330 uint64_t packet_count, byte_count;
3331 size_t act_len = sizeof *rule->actions * rule->n_actions;
3333 /* Don't report on subrules. */
3334 if (rule->super != NULL) {
3338 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3339 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3340 cbdata->ofproto->tun_id_from_cookie, &match);
3342 ds_put_format(results, "duration=%llds, ",
3343 (time_msec() - rule->created) / 1000);
3344 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3345 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3346 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3347 ofp_print_match(results, &match, true);
3349 ofp_print_actions(results, &rule->actions->header, act_len);
3351 ds_put_cstr(results, "drop");
3353 ds_put_cstr(results, "\n");
3356 /* Adds a pretty-printed description of all flows to 'results', including
3357 * those marked hidden by secchan (e.g., by in-band control). */
3359 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3361 struct ofp_match match;
3362 struct cls_rule target;
3363 struct flow_stats_ds_cbdata cbdata;
3365 memset(&match, 0, sizeof match);
3366 match.wildcards = htonl(OVSFW_ALL);
3369 cbdata.results = results;
3371 cls_rule_from_match(&match, 0, false, 0, &target);
3372 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3373 flow_stats_ds_cb, &cbdata);
3376 struct aggregate_stats_cbdata {
3377 struct ofproto *ofproto;
3379 uint64_t packet_count;
3380 uint64_t byte_count;
3385 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3387 struct rule *rule = rule_from_cls_rule(rule_);
3388 struct aggregate_stats_cbdata *cbdata = cbdata_;
3389 uint64_t packet_count, byte_count;
3391 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3395 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3397 cbdata->packet_count += packet_count;
3398 cbdata->byte_count += byte_count;
3403 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3404 const struct ofp_stats_request *osr,
3407 struct ofp_aggregate_stats_request *asr;
3408 struct ofp_aggregate_stats_reply *reply;
3409 struct aggregate_stats_cbdata cbdata;
3410 struct cls_rule target;
3413 if (arg_size != sizeof *asr) {
3414 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3416 asr = (struct ofp_aggregate_stats_request *) osr->body;
3418 COVERAGE_INC(ofproto_agg_request);
3420 cbdata.out_port = asr->out_port;
3421 cbdata.packet_count = 0;
3422 cbdata.byte_count = 0;
3424 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3425 classifier_for_each_match(&p->cls, &target,
3426 table_id_to_include(asr->table_id),
3427 aggregate_stats_cb, &cbdata);
3429 msg = start_stats_reply(osr, sizeof *reply);
3430 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3431 reply->flow_count = htonl(cbdata.n_flows);
3432 reply->packet_count = htonll(cbdata.packet_count);
3433 reply->byte_count = htonll(cbdata.byte_count);
3434 queue_tx(msg, ofconn, ofconn->reply_counter);
3438 struct queue_stats_cbdata {
3439 struct ofconn *ofconn;
3440 struct ofport *ofport;
3445 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3446 const struct netdev_queue_stats *stats)
3448 struct ofp_queue_stats *reply;
3450 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3451 reply->port_no = htons(cbdata->ofport->opp.port_no);
3452 memset(reply->pad, 0, sizeof reply->pad);
3453 reply->queue_id = htonl(queue_id);
3454 reply->tx_bytes = htonll(stats->tx_bytes);
3455 reply->tx_packets = htonll(stats->tx_packets);
3456 reply->tx_errors = htonll(stats->tx_errors);
3460 handle_queue_stats_dump_cb(uint32_t queue_id,
3461 struct netdev_queue_stats *stats,
3464 struct queue_stats_cbdata *cbdata = cbdata_;
3466 put_queue_stats(cbdata, queue_id, stats);
3470 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3471 struct queue_stats_cbdata *cbdata)
3473 cbdata->ofport = port;
3474 if (queue_id == OFPQ_ALL) {
3475 netdev_dump_queue_stats(port->netdev,
3476 handle_queue_stats_dump_cb, cbdata);
3478 struct netdev_queue_stats stats;
3480 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3481 put_queue_stats(cbdata, queue_id, &stats);
3487 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3488 const struct ofp_stats_request *osr,
3491 struct ofp_queue_stats_request *qsr;
3492 struct queue_stats_cbdata cbdata;
3493 struct ofport *port;
3494 unsigned int port_no;
3497 if (arg_size != sizeof *qsr) {
3498 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3500 qsr = (struct ofp_queue_stats_request *) osr->body;
3502 COVERAGE_INC(ofproto_queue_req);
3504 cbdata.ofconn = ofconn;
3505 cbdata.msg = start_stats_reply(osr, 128);
3507 port_no = ntohs(qsr->port_no);
3508 queue_id = ntohl(qsr->queue_id);
3509 if (port_no == OFPP_ALL) {
3510 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3511 handle_queue_stats_for_port(port, queue_id, &cbdata);
3513 } else if (port_no < ofproto->max_ports) {
3514 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3516 handle_queue_stats_for_port(port, queue_id, &cbdata);
3519 ofpbuf_delete(cbdata.msg);
3520 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3522 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3528 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3529 struct ofp_header *oh)
3531 struct ofp_stats_request *osr;
3535 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3540 osr = (struct ofp_stats_request *) oh;
3542 switch (ntohs(osr->type)) {
3544 return handle_desc_stats_request(p, ofconn, osr);
3547 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3549 case OFPST_AGGREGATE:
3550 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3553 return handle_table_stats_request(p, ofconn, osr);
3556 return handle_port_stats_request(p, ofconn, osr, arg_size);
3559 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3562 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3565 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3569 static long long int
3570 msec_from_nsec(uint64_t sec, uint32_t nsec)
3572 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3576 update_time(struct ofproto *ofproto, struct rule *rule,
3577 const struct odp_flow_stats *stats)
3579 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3580 if (used > rule->used) {
3582 if (rule->super && used > rule->super->used) {
3583 rule->super->used = used;
3585 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3590 update_stats(struct ofproto *ofproto, struct rule *rule,
3591 const struct odp_flow_stats *stats)
3593 if (stats->n_packets) {
3594 update_time(ofproto, rule, stats);
3595 rule->packet_count += stats->n_packets;
3596 rule->byte_count += stats->n_bytes;
3597 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3601 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3602 * in which no matching flow already exists in the flow table.
3604 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3605 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3606 * code as encoded by ofp_mkerr() on failure.
3608 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3611 add_flow(struct ofproto *p, struct ofconn *ofconn,
3612 const struct ofp_flow_mod *ofm, size_t n_actions)
3614 struct ofpbuf *packet;
3619 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3623 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3625 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3626 ntohs(ofm->priority))) {
3627 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3631 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3632 n_actions, ntohs(ofm->idle_timeout),
3633 ntohs(ofm->hard_timeout), ofm->cookie,
3634 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3635 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3636 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3639 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3640 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3644 in_port = UINT16_MAX;
3647 rule_insert(p, rule, packet, in_port);
3651 static struct rule *
3652 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3657 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3659 return rule_from_cls_rule(classifier_find_rule_exactly(
3660 &p->cls, &flow, wildcards,
3661 ntohs(ofm->priority)));
3665 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3666 struct rule *rule, const struct ofp_flow_mod *ofm)
3668 struct ofpbuf *packet;
3673 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3677 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3683 flow_extract(packet, 0, in_port, &flow);
3684 rule_execute(ofproto, rule, packet, &flow);
3689 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3691 struct modify_flows_cbdata {
3692 struct ofproto *ofproto;
3693 const struct ofp_flow_mod *ofm;
3698 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3699 size_t n_actions, struct rule *);
3700 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3702 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3703 * encoded by ofp_mkerr() on failure.
3705 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3708 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3709 const struct ofp_flow_mod *ofm, size_t n_actions)
3711 struct modify_flows_cbdata cbdata;
3712 struct cls_rule target;
3716 cbdata.n_actions = n_actions;
3717 cbdata.match = NULL;
3719 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3722 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3723 modify_flows_cb, &cbdata);
3725 /* This credits the packet to whichever flow happened to happened to
3726 * match last. That's weird. Maybe we should do a lookup for the
3727 * flow that actually matches the packet? Who knows. */
3728 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3731 return add_flow(p, ofconn, ofm, n_actions);
3735 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3736 * code as encoded by ofp_mkerr() on failure.
3738 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3741 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3742 struct ofp_flow_mod *ofm, size_t n_actions)
3744 struct rule *rule = find_flow_strict(p, ofm);
3745 if (rule && !rule_is_hidden(rule)) {
3746 modify_flow(p, ofm, n_actions, rule);
3747 return send_buffered_packet(p, ofconn, rule, ofm);
3749 return add_flow(p, ofconn, ofm, n_actions);
3753 /* Callback for modify_flows_loose(). */
3755 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3757 struct rule *rule = rule_from_cls_rule(rule_);
3758 struct modify_flows_cbdata *cbdata = cbdata_;
3760 if (!rule_is_hidden(rule)) {
3761 cbdata->match = rule;
3762 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3766 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3767 * been identified as a flow in 'p''s flow table to be modified, by changing
3768 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3769 * ofp_action[] structures). */
3771 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3772 size_t n_actions, struct rule *rule)
3774 size_t actions_len = n_actions * sizeof *rule->actions;
3776 rule->flow_cookie = ofm->cookie;
3778 /* If the actions are the same, do nothing. */
3779 if (n_actions == rule->n_actions
3780 && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len)))
3785 /* Replace actions. */
3786 free(rule->actions);
3787 rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL;
3788 rule->n_actions = n_actions;
3790 /* Make sure that the datapath gets updated properly. */
3791 if (rule->cr.wc.wildcards) {
3792 COVERAGE_INC(ofproto_mod_wc_flow);
3793 p->need_revalidate = true;
3795 rule_update_actions(p, rule);
3801 /* OFPFC_DELETE implementation. */
3803 struct delete_flows_cbdata {
3804 struct ofproto *ofproto;
3808 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3809 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3811 /* Implements OFPFC_DELETE. */
3813 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3815 struct delete_flows_cbdata cbdata;
3816 struct cls_rule target;
3819 cbdata.out_port = ofm->out_port;
3821 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3824 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3825 delete_flows_cb, &cbdata);
3828 /* Implements OFPFC_DELETE_STRICT. */
3830 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3832 struct rule *rule = find_flow_strict(p, ofm);
3834 delete_flow(p, rule, ofm->out_port);
3838 /* Callback for delete_flows_loose(). */
3840 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3842 struct rule *rule = rule_from_cls_rule(rule_);
3843 struct delete_flows_cbdata *cbdata = cbdata_;
3845 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3848 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3849 * been identified as a flow to delete from 'p''s flow table, by deleting the
3850 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3853 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3854 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3855 * specified 'out_port'. */
3857 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3859 if (rule_is_hidden(rule)) {
3863 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3867 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3868 rule_remove(p, rule);
3872 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3873 struct ofp_flow_mod *ofm)
3875 struct ofp_match orig_match;
3879 error = reject_slave_controller(ofconn, &ofm->header);
3883 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3884 sizeof *ofm->actions, &n_actions);
3889 /* We do not support the emergency flow cache. It will hopefully
3890 * get dropped from OpenFlow in the near future. */
3891 if (ofm->flags & htons(OFPFF_EMERG)) {
3892 /* There isn't a good fit for an error code, so just state that the
3893 * flow table is full. */
3894 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3897 /* Normalize ofp->match. If normalization actually changes anything, then
3898 * log the differences. */
3899 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3900 orig_match = ofm->match;
3901 normalize_match(&ofm->match);
3902 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3903 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3904 if (!VLOG_DROP_INFO(&normal_rl)) {
3905 char *old = ofp_match_to_literal_string(&orig_match);
3906 char *new = ofp_match_to_literal_string(&ofm->match);
3907 VLOG_INFO("%s: normalization changed ofp_match, details:",
3908 rconn_get_name(ofconn->rconn));
3909 VLOG_INFO(" pre: %s", old);
3910 VLOG_INFO("post: %s", new);
3916 if (!ofm->match.wildcards) {
3917 ofm->priority = htons(UINT16_MAX);
3920 error = validate_actions((const union ofp_action *) ofm->actions,
3921 n_actions, p->max_ports);
3926 switch (ntohs(ofm->command)) {
3928 return add_flow(p, ofconn, ofm, n_actions);
3931 return modify_flows_loose(p, ofconn, ofm, n_actions);
3933 case OFPFC_MODIFY_STRICT:
3934 return modify_flow_strict(p, ofconn, ofm, n_actions);
3937 delete_flows_loose(p, ofm);
3940 case OFPFC_DELETE_STRICT:
3941 delete_flow_strict(p, ofm);
3945 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3950 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3954 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3959 p->tun_id_from_cookie = !!msg->set;
3964 handle_role_request(struct ofproto *ofproto,
3965 struct ofconn *ofconn, struct nicira_header *msg)
3967 struct nx_role_request *nrr;
3968 struct nx_role_request *reply;
3972 if (ntohs(msg->header.length) != sizeof *nrr) {
3973 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3974 ntohs(msg->header.length), sizeof *nrr);
3975 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3977 nrr = (struct nx_role_request *) msg;
3979 if (ofconn->type != OFCONN_PRIMARY) {
3980 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3982 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3985 role = ntohl(nrr->role);
3986 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3987 && role != NX_ROLE_SLAVE) {
3988 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3990 /* There's no good error code for this. */
3991 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3994 if (role == NX_ROLE_MASTER) {
3995 struct ofconn *other;
3997 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
3998 if (other->role == NX_ROLE_MASTER) {
3999 other->role = NX_ROLE_SLAVE;
4003 ofconn->role = role;
4005 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
4007 reply->nxh.vendor = htonl(NX_VENDOR_ID);
4008 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
4009 reply->role = htonl(role);
4010 queue_tx(buf, ofconn, ofconn->reply_counter);
4016 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
4018 struct ofp_vendor_header *ovh = msg;
4019 struct nicira_header *nh;
4021 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4022 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4023 "(expected at least %zu)",
4024 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4025 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4027 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4028 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4030 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4031 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4032 "(expected at least %zu)",
4033 ntohs(ovh->header.length), sizeof(struct nicira_header));
4034 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4038 switch (ntohl(nh->subtype)) {
4039 case NXT_STATUS_REQUEST:
4040 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4043 case NXT_TUN_ID_FROM_COOKIE:
4044 return handle_tun_id_from_cookie(p, msg);
4046 case NXT_ROLE_REQUEST:
4047 return handle_role_request(p, ofconn, msg);
4050 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4054 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4056 struct ofp_header *ob;
4059 /* Currently, everything executes synchronously, so we can just
4060 * immediately send the barrier reply. */
4061 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4062 queue_tx(buf, ofconn, ofconn->reply_counter);
4067 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4068 struct ofpbuf *ofp_msg)
4070 struct ofp_header *oh = ofp_msg->data;
4073 COVERAGE_INC(ofproto_recv_openflow);
4075 case OFPT_ECHO_REQUEST:
4076 error = handle_echo_request(ofconn, oh);
4079 case OFPT_ECHO_REPLY:
4083 case OFPT_FEATURES_REQUEST:
4084 error = handle_features_request(p, ofconn, oh);
4087 case OFPT_GET_CONFIG_REQUEST:
4088 error = handle_get_config_request(p, ofconn, oh);
4091 case OFPT_SET_CONFIG:
4092 error = handle_set_config(p, ofconn, ofp_msg->data);
4095 case OFPT_PACKET_OUT:
4096 error = handle_packet_out(p, ofconn, ofp_msg->data);
4100 error = handle_port_mod(p, ofconn, oh);
4104 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4107 case OFPT_STATS_REQUEST:
4108 error = handle_stats_request(p, ofconn, oh);
4112 error = handle_vendor(p, ofconn, ofp_msg->data);
4115 case OFPT_BARRIER_REQUEST:
4116 error = handle_barrier_request(ofconn, oh);
4120 if (VLOG_IS_WARN_ENABLED()) {
4121 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4122 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4125 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4130 send_error_oh(ofconn, ofp_msg->data, error);
4135 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4137 struct odp_msg *msg = packet->data;
4139 struct ofpbuf payload;
4142 payload.data = msg + 1;
4143 payload.size = msg->length - sizeof *msg;
4144 flow_extract(&payload, msg->arg, msg->port, &flow);
4146 /* Check with in-band control to see if this packet should be sent
4147 * to the local port regardless of the flow table. */
4148 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4149 union odp_action action;
4151 memset(&action, 0, sizeof(action));
4152 action.output.type = ODPAT_OUTPUT;
4153 action.output.port = ODPP_LOCAL;
4154 dpif_execute(p->dpif, &action, 1, &payload);
4157 rule = lookup_valid_rule(p, &flow);
4159 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4160 struct ofport *port = get_port(p, msg->port);
4162 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4163 COVERAGE_INC(ofproto_no_packet_in);
4164 /* XXX install 'drop' flow entry */
4165 ofpbuf_delete(packet);
4169 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4172 COVERAGE_INC(ofproto_packet_in);
4173 send_packet_in(p, packet);
4177 if (rule->cr.wc.wildcards) {
4178 rule = rule_create_subrule(p, rule, &flow);
4179 rule_make_actions(p, rule, packet);
4181 if (!rule->may_install) {
4182 /* The rule is not installable, that is, we need to process every
4183 * packet, so process the current packet and set its actions into
4185 rule_make_actions(p, rule, packet);
4187 /* XXX revalidate rule if it needs it */
4191 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4193 * Extra-special case for fail-open mode.
4195 * We are in fail-open mode and the packet matched the fail-open rule,
4196 * but we are connected to a controller too. We should send the packet
4197 * up to the controller in the hope that it will try to set up a flow
4198 * and thereby allow us to exit fail-open.
4200 * See the top-level comment in fail-open.c for more information.
4202 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4203 DPIF_RECV_MSG_PADDING));
4206 ofpbuf_pull(packet, sizeof *msg);
4207 rule_execute(p, rule, packet, &flow);
4208 rule_reinstall(p, rule);
4212 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4214 struct odp_msg *msg = packet->data;
4216 switch (msg->type) {
4217 case _ODPL_ACTION_NR:
4218 COVERAGE_INC(ofproto_ctlr_action);
4219 send_packet_in(p, packet);
4222 case _ODPL_SFLOW_NR:
4224 ofproto_sflow_received(p->sflow, msg);
4226 ofpbuf_delete(packet);
4230 handle_odp_miss_msg(p, packet);
4234 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4240 /* Flow expiration. */
4242 struct expire_cbdata {
4243 struct ofproto *ofproto;
4247 static int ofproto_dp_max_idle(const struct ofproto *);
4248 static void ofproto_update_used(struct ofproto *);
4249 static void rule_expire(struct cls_rule *, void *cbdata);
4251 /* This function is called periodically by ofproto_run(). Its job is to
4252 * collect updates for the flows that have been installed into the datapath,
4253 * most importantly when they last were used, and then use that information to
4254 * expire flows that have not been used recently.
4256 * Returns the number of milliseconds after which it should be called again. */
4258 ofproto_expire(struct ofproto *ofproto)
4260 struct expire_cbdata cbdata;
4262 /* Update 'used' for each flow in the datapath. */
4263 ofproto_update_used(ofproto);
4265 /* Expire idle flows.
4267 * A wildcarded flow is idle only when all of its subrules have expired due
4268 * to becoming idle, so iterate through the exact-match flows first. */
4269 cbdata.ofproto = ofproto;
4270 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4271 classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
4272 classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
4274 /* Let the hook know that we're at a stable point: all outstanding data
4275 * in existing flows has been accounted to the account_cb. Thus, the
4276 * hook can now reasonably do operations that depend on having accurate
4277 * flow volume accounting (currently, that's just bond rebalancing). */
4278 if (ofproto->ofhooks->account_checkpoint_cb) {
4279 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4282 return MIN(cbdata.dp_max_idle, 1000);
4285 /* Update 'used' member of each flow currently installed into the datapath. */
4287 ofproto_update_used(struct ofproto *p)
4289 struct odp_flow *flows;
4294 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4299 for (i = 0; i < n_flows; i++) {
4300 struct odp_flow *f = &flows[i];
4304 odp_flow_key_to_flow(&f->key, &flow);
4306 rule = rule_from_cls_rule(
4307 classifier_find_rule_exactly(&p->cls, &flow, 0, UINT16_MAX));
4309 if (rule && rule->installed) {
4310 update_time(p, rule, &f->stats);
4311 rule_account(p, rule, f->stats.n_bytes);
4313 /* There's a flow in the datapath that we know nothing about.
4315 COVERAGE_INC(ofproto_unexpected_rule);
4316 dpif_flow_del(p->dpif, f);
4323 /* Calculates and returns the number of milliseconds of idle time after which
4324 * flows should expire from the datapath and we should fold their statistics
4325 * into their parent rules in userspace. */
4327 ofproto_dp_max_idle(const struct ofproto *ofproto)
4330 * Idle time histogram.
4332 * Most of the time a switch has a relatively small number of flows. When
4333 * this is the case we might as well keep statistics for all of them in
4334 * userspace and to cache them in the kernel datapath for performance as
4337 * As the number of flows increases, the memory required to maintain
4338 * statistics about them in userspace and in the kernel becomes
4339 * significant. However, with a large number of flows it is likely that
4340 * only a few of them are "heavy hitters" that consume a large amount of
4341 * bandwidth. At this point, only heavy hitters are worth caching in the
4342 * kernel and maintaining in userspaces; other flows we can discard.
4344 * The technique used to compute the idle time is to build a histogram with
4345 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4346 * is installed in the kernel gets dropped in the appropriate bucket.
4347 * After the histogram has been built, we compute the cutoff so that only
4348 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4349 * cached. At least the most-recently-used bucket of flows is kept, so
4350 * actually an arbitrary number of flows can be kept in any given
4351 * expiration run (though the next run will delete most of those unless
4352 * they receive additional data).
4354 * This requires a second pass through the exact-match flows, in addition
4355 * to the pass made by ofproto_update_used(), because the former function
4356 * never looks at uninstallable flows.
4358 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4359 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4360 int buckets[N_BUCKETS] = { 0 };
4366 total = classifier_count_exact(&ofproto->cls);
4367 if (total <= 1000) {
4368 return N_BUCKETS * BUCKET_WIDTH;
4371 /* Build histogram. */
4373 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) {
4374 long long int idle = now - rule->used;
4375 int bucket = (idle <= 0 ? 0
4376 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4377 : (unsigned int) idle / BUCKET_WIDTH);
4381 /* Find the first bucket whose flows should be expired. */
4382 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4383 if (buckets[bucket]) {
4386 subtotal += buckets[bucket++];
4387 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4392 if (VLOG_IS_DBG_ENABLED()) {
4396 ds_put_cstr(&s, "keep");
4397 for (i = 0; i < N_BUCKETS; i++) {
4399 ds_put_cstr(&s, ", drop");
4402 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4405 VLOG_INFO("%s: %s (msec:count)",
4406 dpif_name(ofproto->dpif), ds_cstr(&s));
4410 return bucket * BUCKET_WIDTH;
4414 rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
4416 if (ofproto->netflow && !is_controller_rule(rule) &&
4417 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4418 struct ofexpired expired;
4419 struct odp_flow odp_flow;
4421 /* Get updated flow stats.
4423 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4424 * updated TCP flags and (2) the dpif_flow_list_all() in
4425 * ofproto_update_used() zeroed TCP flags. */
4426 memset(&odp_flow, 0, sizeof odp_flow);
4427 if (rule->installed) {
4428 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
4429 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4430 dpif_flow_get(ofproto->dpif, &odp_flow);
4432 if (odp_flow.stats.n_packets) {
4433 update_time(ofproto, rule, &odp_flow.stats);
4434 netflow_flow_update_flags(&rule->nf_flow,
4435 odp_flow.stats.tcp_flags);
4439 expired.flow = rule->cr.flow;
4440 expired.packet_count = rule->packet_count +
4441 odp_flow.stats.n_packets;
4442 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4443 expired.used = rule->used;
4445 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4449 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4450 * rules, then delete it entirely.
4452 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4453 * the datapath and fold its statistics back into its super-rule.
4455 * (This is a callback function for classifier_for_each().) */
4457 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4459 struct expire_cbdata *cbdata = cbdata_;
4460 struct ofproto *ofproto = cbdata->ofproto;
4461 struct rule *rule = rule_from_cls_rule(cls_rule);
4462 long long int hard_expire, idle_expire, expire, now;
4464 /* Calculate OpenFlow expiration times for 'rule'. */
4465 hard_expire = (rule->hard_timeout
4466 ? rule->created + rule->hard_timeout * 1000
4468 idle_expire = (rule->idle_timeout
4469 && (rule->super || list_is_empty(&rule->list))
4470 ? rule->used + rule->idle_timeout * 1000
4472 expire = MIN(hard_expire, idle_expire);
4476 /* 'rule' has not expired according to OpenFlow rules. */
4477 if (!rule->cr.wc.wildcards) {
4478 if (now >= rule->used + cbdata->dp_max_idle) {
4479 /* This rule is idle, so drop it to free up resources. */
4481 /* It's not part of the OpenFlow flow table, so we can
4482 * delete it entirely and fold its statistics into its
4484 rule_remove(ofproto, rule);
4486 /* It is part of the OpenFlow flow table, so we have to
4487 * keep the rule but we can at least uninstall it from the
4489 rule_uninstall(ofproto, rule);
4492 /* Send NetFlow active timeout if appropriate. */
4493 rule_active_timeout(cbdata->ofproto, rule);
4497 /* 'rule' has expired according to OpenFlow rules. */
4498 COVERAGE_INC(ofproto_expired);
4500 /* Update stats. (This is a no-op if the rule expired due to an idle
4501 * timeout, because that only happens when the rule has no subrules
4503 if (rule->cr.wc.wildcards) {
4504 struct rule *subrule, *next;
4505 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
4506 rule_remove(cbdata->ofproto, subrule);
4509 rule_uninstall(cbdata->ofproto, rule);
4512 /* Get rid of the rule. */
4513 if (!rule_is_hidden(rule)) {
4514 send_flow_removed(cbdata->ofproto, rule, now,
4516 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4518 rule_remove(cbdata->ofproto, rule);
4523 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4525 struct rule *sub = rule_from_cls_rule(sub_);
4526 struct revalidate_cbdata *cbdata = cbdata_;
4528 if (cbdata->revalidate_all
4529 || (cbdata->revalidate_subrules && sub->super)
4530 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4531 revalidate_rule(cbdata->ofproto, sub);
4536 revalidate_rule(struct ofproto *p, struct rule *rule)
4538 const struct flow *flow = &rule->cr.flow;
4540 COVERAGE_INC(ofproto_revalidate_rule);
4543 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4545 rule_remove(p, rule);
4547 } else if (super != rule->super) {
4548 COVERAGE_INC(ofproto_revalidate_moved);
4549 list_remove(&rule->list);
4550 list_push_back(&super->list, &rule->list);
4551 rule->super = super;
4552 rule->hard_timeout = super->hard_timeout;
4553 rule->idle_timeout = super->idle_timeout;
4554 rule->created = super->created;
4559 rule_update_actions(p, rule);
4563 static struct ofpbuf *
4564 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4565 long long int now, uint8_t reason)
4567 struct ofp_flow_removed *ofr;
4569 long long int tdiff = now - rule->created;
4570 uint32_t sec = tdiff / 1000;
4571 uint32_t msec = tdiff - (sec * 1000);
4573 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4574 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4576 ofr->cookie = rule->flow_cookie;
4577 ofr->priority = htons(rule->cr.priority);
4578 ofr->reason = reason;
4579 ofr->duration_sec = htonl(sec);
4580 ofr->duration_nsec = htonl(msec * 1000000);
4581 ofr->idle_timeout = htons(rule->idle_timeout);
4582 ofr->packet_count = htonll(rule->packet_count);
4583 ofr->byte_count = htonll(rule->byte_count);
4589 send_flow_removed(struct ofproto *p, struct rule *rule,
4590 long long int now, uint8_t reason)
4592 struct ofconn *ofconn;
4593 struct ofconn *prev;
4594 struct ofpbuf *buf = NULL;
4596 if (!rule->send_flow_removed) {
4600 /* We limit the maximum number of queued flow expirations it by accounting
4601 * them under the counter for replies. That works because preventing
4602 * OpenFlow requests from being processed also prevents new flows from
4603 * being added (and expiring). (It also prevents processing OpenFlow
4604 * requests that would not add new flows, so it is imperfect.) */
4607 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4608 if (rconn_is_connected(ofconn->rconn)
4609 && ofconn_receives_async_msgs(ofconn)) {
4611 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4613 buf = compose_flow_removed(p, rule, now, reason);
4619 queue_tx(buf, prev, prev->reply_counter);
4623 /* pinsched callback for sending 'packet' on 'ofconn'. */
4625 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4627 struct ofconn *ofconn = ofconn_;
4629 rconn_send_with_limit(ofconn->rconn, packet,
4630 ofconn->packet_in_counter, 100);
4633 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4634 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4635 * packet scheduler for sending.
4637 * 'max_len' specifies the maximum number of bytes of the packet to send on
4638 * 'ofconn' (INT_MAX specifies no limit).
4640 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4641 * ownership is transferred to this function. */
4643 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4646 struct ofproto *ofproto = ofconn->ofproto;
4647 struct ofp_packet_in *opi = packet->data;
4648 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4649 int send_len, trim_size;
4653 if (opi->reason == OFPR_ACTION) {
4654 buffer_id = UINT32_MAX;
4655 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4656 buffer_id = pktbuf_get_null();
4657 } else if (!ofconn->pktbuf) {
4658 buffer_id = UINT32_MAX;
4660 struct ofpbuf payload;
4661 payload.data = opi->data;
4662 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4663 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4666 /* Figure out how much of the packet to send. */
4667 send_len = ntohs(opi->total_len);
4668 if (buffer_id != UINT32_MAX) {
4669 send_len = MIN(send_len, ofconn->miss_send_len);
4671 send_len = MIN(send_len, max_len);
4673 /* Adjust packet length and clone if necessary. */
4674 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4676 packet = ofpbuf_clone_data(packet->data, trim_size);
4679 packet->size = trim_size;
4682 /* Update packet headers. */
4683 opi->buffer_id = htonl(buffer_id);
4684 update_openflow_length(packet);
4686 /* Hand over to packet scheduler. It might immediately call into
4687 * do_send_packet_in() or it might buffer it for a while (until a later
4688 * call to pinsched_run()). */
4689 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4690 packet, do_send_packet_in, ofconn);
4693 /* Replace struct odp_msg header in 'packet' by equivalent struct
4694 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4695 * returned by dpif_recv()).
4697 * The conversion is not complete: the caller still needs to trim any unneeded
4698 * payload off the end of the buffer, set the length in the OpenFlow header,
4699 * and set buffer_id. Those require us to know the controller settings and so
4700 * must be done on a per-controller basis.
4702 * Returns the maximum number of bytes of the packet that should be sent to
4703 * the controller (INT_MAX if no limit). */
4705 do_convert_to_packet_in(struct ofpbuf *packet)
4707 struct odp_msg *msg = packet->data;
4708 struct ofp_packet_in *opi;
4714 /* Extract relevant header fields */
4715 if (msg->type == _ODPL_ACTION_NR) {
4716 reason = OFPR_ACTION;
4719 reason = OFPR_NO_MATCH;
4722 total_len = msg->length - sizeof *msg;
4723 in_port = odp_port_to_ofp_port(msg->port);
4725 /* Repurpose packet buffer by overwriting header. */
4726 ofpbuf_pull(packet, sizeof(struct odp_msg));
4727 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4728 opi->header.version = OFP_VERSION;
4729 opi->header.type = OFPT_PACKET_IN;
4730 opi->total_len = htons(total_len);
4731 opi->in_port = htons(in_port);
4732 opi->reason = reason;
4737 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4738 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4739 * as necessary according to their individual configurations.
4741 * 'packet' must have sufficient headroom to convert it into a struct
4742 * ofp_packet_in (e.g. as returned by dpif_recv()).
4744 * Takes ownership of 'packet'. */
4746 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4748 struct ofconn *ofconn, *prev;
4751 max_len = do_convert_to_packet_in(packet);
4754 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4755 if (ofconn_receives_async_msgs(ofconn)) {
4757 schedule_packet_in(prev, packet, max_len, true);
4763 schedule_packet_in(prev, packet, max_len, false);
4765 ofpbuf_delete(packet);
4770 pick_datapath_id(const struct ofproto *ofproto)
4772 const struct ofport *port;
4774 port = get_port(ofproto, ODPP_LOCAL);
4776 uint8_t ea[ETH_ADDR_LEN];
4779 error = netdev_get_etheraddr(port->netdev, ea);
4781 return eth_addr_to_uint64(ea);
4783 VLOG_WARN("could not get MAC address for %s (%s)",
4784 netdev_get_name(port->netdev), strerror(error));
4786 return ofproto->fallback_dpid;
4790 pick_fallback_dpid(void)
4792 uint8_t ea[ETH_ADDR_LEN];
4793 eth_addr_nicira_random(ea);
4794 return eth_addr_to_uint64(ea);
4798 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4799 struct odp_actions *actions, tag_type *tags,
4800 uint16_t *nf_output_iface, void *ofproto_)
4802 struct ofproto *ofproto = ofproto_;
4805 /* Drop frames for reserved multicast addresses. */
4806 if (eth_addr_is_reserved(flow->dl_dst)) {
4810 /* Learn source MAC (but don't try to learn from revalidation). */
4811 if (packet != NULL) {
4812 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4814 GRAT_ARP_LOCK_NONE);
4816 /* The log messages here could actually be useful in debugging,
4817 * so keep the rate limit relatively high. */
4818 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4819 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4820 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4821 ofproto_revalidate(ofproto, rev_tag);
4825 /* Determine output port. */
4826 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4829 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4830 nf_output_iface, actions);
4831 } else if (out_port != flow->in_port) {
4832 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4833 *nf_output_iface = out_port;
4841 static const struct ofhooks default_ofhooks = {
4842 default_normal_ofhook_cb,