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)
1326 struct cls_rule target;
1329 cls_rule_from_flow(flow, wildcards, priority, &target);
1330 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1333 rule_remove(ofproto, rule);
1338 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1340 struct rule *rule = rule_from_cls_rule(rule_);
1341 struct ofproto *ofproto = ofproto_;
1343 /* Mark the flow as not installed, even though it might really be
1344 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1345 * There is no point in uninstalling it individually since we are about to
1346 * blow away all the flows with dpif_flow_flush(). */
1347 rule->installed = false;
1349 rule_remove(ofproto, rule);
1353 ofproto_flush_flows(struct ofproto *ofproto)
1355 COVERAGE_INC(ofproto_flush);
1356 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1357 dpif_flow_flush(ofproto->dpif);
1358 if (ofproto->in_band) {
1359 in_band_flushed(ofproto->in_band);
1361 if (ofproto->fail_open) {
1362 fail_open_flushed(ofproto->fail_open);
1367 reinit_ports(struct ofproto *p)
1369 struct svec devnames;
1370 struct ofport *ofport;
1371 struct odp_port *odp_ports;
1375 COVERAGE_INC(ofproto_reinit_ports);
1377 svec_init(&devnames);
1378 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1379 svec_add (&devnames, (char *) ofport->opp.name);
1381 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1382 for (i = 0; i < n_odp_ports; i++) {
1383 svec_add (&devnames, odp_ports[i].devname);
1387 svec_sort_unique(&devnames);
1388 for (i = 0; i < devnames.n; i++) {
1389 update_port(p, devnames.names[i]);
1391 svec_destroy(&devnames);
1394 static struct ofport *
1395 make_ofport(const struct odp_port *odp_port)
1397 struct netdev_options netdev_options;
1398 enum netdev_flags flags;
1399 struct ofport *ofport;
1400 struct netdev *netdev;
1403 memset(&netdev_options, 0, sizeof netdev_options);
1404 netdev_options.name = odp_port->devname;
1405 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1407 error = netdev_open(&netdev_options, &netdev);
1409 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1410 "cannot be opened (%s)",
1411 odp_port->devname, odp_port->port,
1412 odp_port->devname, strerror(error));
1416 ofport = xmalloc(sizeof *ofport);
1417 ofport->netdev = netdev;
1418 ofport->odp_port = odp_port->port;
1419 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1420 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1421 memcpy(ofport->opp.name, odp_port->devname,
1422 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1423 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1425 netdev_get_flags(netdev, &flags);
1426 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1428 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1430 netdev_get_features(netdev,
1431 &ofport->opp.curr, &ofport->opp.advertised,
1432 &ofport->opp.supported, &ofport->opp.peer);
1437 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1439 if (get_port(p, odp_port->port)) {
1440 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1443 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1444 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1453 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1455 const struct ofp_phy_port *a = &a_->opp;
1456 const struct ofp_phy_port *b = &b_->opp;
1458 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1459 return (a->port_no == b->port_no
1460 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1461 && !strcmp((char *) a->name, (char *) b->name)
1462 && a->state == b->state
1463 && a->config == b->config
1464 && a->curr == b->curr
1465 && a->advertised == b->advertised
1466 && a->supported == b->supported
1467 && a->peer == b->peer);
1471 send_port_status(struct ofproto *p, const struct ofport *ofport,
1474 /* XXX Should limit the number of queued port status change messages. */
1475 struct ofconn *ofconn;
1476 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1477 struct ofp_port_status *ops;
1480 if (!ofconn_receives_async_msgs(ofconn)) {
1484 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1485 ops->reason = reason;
1486 ops->desc = ofport->opp;
1487 hton_ofp_phy_port(&ops->desc);
1488 queue_tx(b, ofconn, NULL);
1493 ofport_install(struct ofproto *p, struct ofport *ofport)
1495 const char *netdev_name = (const char *) ofport->opp.name;
1497 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1498 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1499 shash_add(&p->port_by_name, netdev_name, ofport);
1501 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1506 ofport_remove(struct ofproto *p, struct ofport *ofport)
1508 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1509 hmap_remove(&p->ports, &ofport->hmap_node);
1510 shash_delete(&p->port_by_name,
1511 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1513 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1518 ofport_free(struct ofport *ofport)
1521 netdev_close(ofport->netdev);
1526 static struct ofport *
1527 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1529 struct ofport *port;
1531 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1532 hash_int(odp_port, 0), &ofproto->ports) {
1533 if (port->odp_port == odp_port) {
1541 update_port(struct ofproto *p, const char *devname)
1543 struct odp_port odp_port;
1544 struct ofport *old_ofport;
1545 struct ofport *new_ofport;
1548 COVERAGE_INC(ofproto_update_port);
1550 /* Query the datapath for port information. */
1551 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1553 /* Find the old ofport. */
1554 old_ofport = shash_find_data(&p->port_by_name, devname);
1557 /* There's no port named 'devname' but there might be a port with
1558 * the same port number. This could happen if a port is deleted
1559 * and then a new one added in its place very quickly, or if a port
1560 * is renamed. In the former case we want to send an OFPPR_DELETE
1561 * and an OFPPR_ADD, and in the latter case we want to send a
1562 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1563 * the old port's ifindex against the new port, or perhaps less
1564 * reliably but more portably by comparing the old port's MAC
1565 * against the new port's MAC. However, this code isn't that smart
1566 * and always sends an OFPPR_MODIFY (XXX). */
1567 old_ofport = get_port(p, odp_port.port);
1569 } else if (error != ENOENT && error != ENODEV) {
1570 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1571 "%s", strerror(error));
1575 /* Create a new ofport. */
1576 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1578 /* Eliminate a few pathological cases. */
1579 if (!old_ofport && !new_ofport) {
1581 } else if (old_ofport && new_ofport) {
1582 /* Most of the 'config' bits are OpenFlow soft state, but
1583 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1584 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1585 * leaves the other bits 0.) */
1586 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1588 if (ofport_equal(old_ofport, new_ofport)) {
1589 /* False alarm--no change. */
1590 ofport_free(new_ofport);
1595 /* Now deal with the normal cases. */
1597 ofport_remove(p, old_ofport);
1600 ofport_install(p, new_ofport);
1602 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1603 (!old_ofport ? OFPPR_ADD
1604 : !new_ofport ? OFPPR_DELETE
1606 ofport_free(old_ofport);
1610 init_ports(struct ofproto *p)
1612 struct odp_port *ports;
1617 error = dpif_port_list(p->dpif, &ports, &n_ports);
1622 for (i = 0; i < n_ports; i++) {
1623 const struct odp_port *odp_port = &ports[i];
1624 if (!ofport_conflicts(p, odp_port)) {
1625 struct ofport *ofport = make_ofport(odp_port);
1627 ofport_install(p, ofport);
1635 static struct ofconn *
1636 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1638 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1639 ofconn->ofproto = p;
1640 list_push_back(&p->all_conns, &ofconn->node);
1641 ofconn->rconn = rconn;
1642 ofconn->type = type;
1643 ofconn->role = NX_ROLE_OTHER;
1644 ofconn->packet_in_counter = rconn_packet_counter_create ();
1645 ofconn->pktbuf = NULL;
1646 ofconn->miss_send_len = 0;
1647 ofconn->reply_counter = rconn_packet_counter_create ();
1652 ofconn_destroy(struct ofconn *ofconn)
1654 if (ofconn->type == OFCONN_PRIMARY) {
1655 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1657 discovery_destroy(ofconn->discovery);
1659 list_remove(&ofconn->node);
1660 switch_status_unregister(ofconn->ss);
1661 rconn_destroy(ofconn->rconn);
1662 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1663 rconn_packet_counter_destroy(ofconn->reply_counter);
1664 pktbuf_destroy(ofconn->pktbuf);
1669 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1674 if (ofconn->discovery) {
1675 char *controller_name;
1676 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1677 discovery_question_connectivity(ofconn->discovery);
1679 if (discovery_run(ofconn->discovery, &controller_name)) {
1680 if (controller_name) {
1681 char *ofconn_name = ofconn_make_name(p, controller_name);
1682 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1685 rconn_disconnect(ofconn->rconn);
1690 for (i = 0; i < N_SCHEDULERS; i++) {
1691 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1694 rconn_run(ofconn->rconn);
1696 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1697 /* Limit the number of iterations to prevent other tasks from
1699 for (iteration = 0; iteration < 50; iteration++) {
1700 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1705 fail_open_maybe_recover(p->fail_open);
1707 handle_openflow(ofconn, p, of_msg);
1708 ofpbuf_delete(of_msg);
1712 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1713 ofconn_destroy(ofconn);
1718 ofconn_wait(struct ofconn *ofconn)
1722 if (ofconn->discovery) {
1723 discovery_wait(ofconn->discovery);
1725 for (i = 0; i < N_SCHEDULERS; i++) {
1726 pinsched_wait(ofconn->schedulers[i]);
1728 rconn_run_wait(ofconn->rconn);
1729 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1730 rconn_recv_wait(ofconn->rconn);
1732 COVERAGE_INC(ofproto_ofconn_stuck);
1736 /* Returns true if 'ofconn' should receive asynchronous messages. */
1738 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1740 if (ofconn->type == OFCONN_PRIMARY) {
1741 /* Primary controllers always get asynchronous messages unless they
1742 * have configured themselves as "slaves". */
1743 return ofconn->role != NX_ROLE_SLAVE;
1745 /* Service connections don't get asynchronous messages unless they have
1746 * explicitly asked for them by setting a nonzero miss send length. */
1747 return ofconn->miss_send_len > 0;
1751 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1752 * and 'target', suitable for use in log messages for identifying the
1755 * The name is dynamically allocated. The caller should free it (with free())
1756 * when it is no longer needed. */
1758 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1760 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1764 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1768 for (i = 0; i < N_SCHEDULERS; i++) {
1769 struct pinsched **s = &ofconn->schedulers[i];
1773 *s = pinsched_create(rate, burst,
1774 ofconn->ofproto->switch_status);
1776 pinsched_set_limits(*s, rate, burst);
1779 pinsched_destroy(*s);
1786 ofservice_reconfigure(struct ofservice *ofservice,
1787 const struct ofproto_controller *c)
1789 ofservice->probe_interval = c->probe_interval;
1790 ofservice->rate_limit = c->rate_limit;
1791 ofservice->burst_limit = c->burst_limit;
1794 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1795 * positive errno value. */
1797 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1799 struct ofservice *ofservice;
1800 struct pvconn *pvconn;
1803 error = pvconn_open(c->target, &pvconn);
1808 ofservice = xzalloc(sizeof *ofservice);
1809 hmap_insert(&ofproto->services, &ofservice->node,
1810 hash_string(c->target, 0));
1811 ofservice->pvconn = pvconn;
1813 ofservice_reconfigure(ofservice, c);
1819 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1821 hmap_remove(&ofproto->services, &ofservice->node);
1822 pvconn_close(ofservice->pvconn);
1826 /* Finds and returns the ofservice within 'ofproto' that has the given
1827 * 'target', or a null pointer if none exists. */
1828 static struct ofservice *
1829 ofservice_lookup(struct ofproto *ofproto, const char *target)
1831 struct ofservice *ofservice;
1833 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1834 &ofproto->services) {
1835 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1842 /* Caller is responsible for initializing the 'cr' member of the returned
1844 static struct rule *
1845 rule_create(struct ofproto *ofproto, struct rule *super,
1846 const union ofp_action *actions, size_t n_actions,
1847 uint16_t idle_timeout, uint16_t hard_timeout,
1848 ovs_be64 flow_cookie, bool send_flow_removed)
1850 struct rule *rule = xzalloc(sizeof *rule);
1851 rule->idle_timeout = idle_timeout;
1852 rule->hard_timeout = hard_timeout;
1853 rule->flow_cookie = flow_cookie;
1854 rule->used = rule->created = time_msec();
1855 rule->send_flow_removed = send_flow_removed;
1856 rule->super = super;
1858 list_push_back(&super->list, &rule->list);
1860 list_init(&rule->list);
1862 if (n_actions > 0) {
1863 rule->n_actions = n_actions;
1864 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1866 netflow_flow_clear(&rule->nf_flow);
1867 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1872 static struct rule *
1873 rule_from_cls_rule(const struct cls_rule *cls_rule)
1875 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1879 rule_free(struct rule *rule)
1881 free(rule->actions);
1882 free(rule->odp_actions);
1886 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1887 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1888 * through all of its subrules and revalidates them, destroying any that no
1889 * longer has a super-rule (which is probably all of them).
1891 * Before calling this function, the caller must make have removed 'rule' from
1892 * the classifier. If 'rule' is an exact-match rule, the caller is also
1893 * responsible for ensuring that it has been uninstalled from the datapath. */
1895 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1898 struct rule *subrule, *next;
1899 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
1900 revalidate_rule(ofproto, subrule);
1903 list_remove(&rule->list);
1909 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1911 const union ofp_action *oa;
1912 struct actions_iterator i;
1914 if (out_port == htons(OFPP_NONE)) {
1917 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1918 oa = actions_next(&i)) {
1919 if (action_outputs_to_port(oa, out_port)) {
1926 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1927 * 'packet', which arrived on 'in_port'.
1929 * Takes ownership of 'packet'. */
1931 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1932 const union odp_action *actions, size_t n_actions,
1933 struct ofpbuf *packet)
1935 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1936 /* As an optimization, avoid a round-trip from userspace to kernel to
1937 * userspace. This also avoids possibly filling up kernel packet
1938 * buffers along the way. */
1939 struct odp_msg *msg;
1941 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1942 msg->type = _ODPL_ACTION_NR;
1943 msg->length = sizeof(struct odp_msg) + packet->size;
1944 msg->port = in_port;
1946 msg->arg = actions[0].controller.arg;
1948 send_packet_in(ofproto, packet);
1954 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
1955 ofpbuf_delete(packet);
1960 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1961 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1962 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1964 * The flow that 'packet' actually contains does not need to actually match
1965 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1966 * the packet and byte counters for 'rule' will be credited for the packet sent
1967 * out whether or not the packet actually matches 'rule'.
1969 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1970 * the caller must already have accurately composed ODP actions for it given
1971 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1972 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1973 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1974 * actions and apply them to 'packet'.
1976 * Takes ownership of 'packet'. */
1978 rule_execute(struct ofproto *ofproto, struct rule *rule,
1979 struct ofpbuf *packet, const struct flow *flow)
1981 const union odp_action *actions;
1982 struct odp_flow_stats stats;
1984 struct odp_actions a;
1986 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1988 /* Grab or compose the ODP actions.
1990 * The special case for an exact-match 'rule' where 'flow' is not the
1991 * rule's flow is important to avoid, e.g., sending a packet out its input
1992 * port simply because the ODP actions were composed for the wrong
1994 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1995 struct rule *super = rule->super ? rule->super : rule;
1996 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1997 packet, &a, NULL, 0, NULL)) {
1998 ofpbuf_delete(packet);
2001 actions = a.actions;
2002 n_actions = a.n_actions;
2004 actions = rule->odp_actions;
2005 n_actions = rule->n_odp_actions;
2008 /* Execute the ODP actions. */
2009 flow_extract_stats(flow, packet, &stats);
2010 if (execute_odp_actions(ofproto, flow->in_port,
2011 actions, n_actions, packet)) {
2012 update_stats(ofproto, rule, &stats);
2013 rule->used = time_msec();
2014 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2018 /* Inserts 'rule' into 'p''s flow table.
2020 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2021 * actions on it and credits the statistics for sending the packet to 'rule'.
2022 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2025 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2028 struct rule *displaced_rule;
2030 /* Insert the rule in the classifier. */
2031 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2032 if (!rule->cr.wc.wildcards) {
2033 rule_make_actions(p, rule, packet);
2036 /* Send the packet and credit it to the rule. */
2039 flow_extract(packet, 0, in_port, &flow);
2040 rule_execute(p, rule, packet, &flow);
2043 /* Install the rule in the datapath only after sending the packet, to
2044 * avoid packet reordering. */
2045 if (rule->cr.wc.wildcards) {
2046 COVERAGE_INC(ofproto_add_wc_flow);
2047 p->need_revalidate = true;
2049 rule_install(p, rule, displaced_rule);
2052 /* Free the rule that was displaced, if any. */
2053 if (displaced_rule) {
2054 rule_destroy(p, displaced_rule);
2058 static struct rule *
2059 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2060 const struct flow *flow)
2062 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2063 rule->idle_timeout, rule->hard_timeout,
2065 COVERAGE_INC(ofproto_subrule_create);
2066 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2067 : rule->cr.priority), &subrule->cr);
2069 if (classifier_insert(&ofproto->cls, &subrule->cr)) {
2077 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2079 * - If 'rule' was installed in the datapath, uninstalls it and updates
2080 * 'rule''s statistics (or its super-rule's statistics, if it is a
2081 * subrule), via rule_uninstall().
2083 * - Removes 'rule' from the classifier.
2085 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2086 * uninstalls and destroys) its subrules, via rule_destroy().
2089 rule_remove(struct ofproto *ofproto, struct rule *rule)
2091 if (rule->cr.wc.wildcards) {
2092 COVERAGE_INC(ofproto_del_wc_flow);
2093 ofproto->need_revalidate = true;
2095 rule_uninstall(ofproto, rule);
2097 classifier_remove(&ofproto->cls, &rule->cr);
2098 rule_destroy(ofproto, rule);
2101 /* Returns true if the actions changed, false otherwise. */
2103 rule_make_actions(struct ofproto *p, struct rule *rule,
2104 const struct ofpbuf *packet)
2106 const struct rule *super;
2107 struct odp_actions a;
2110 assert(!rule->cr.wc.wildcards);
2112 super = rule->super ? rule->super : rule;
2114 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2115 packet, &a, &rule->tags, &rule->may_install,
2116 &rule->nf_flow.output_iface);
2118 actions_len = a.n_actions * sizeof *a.actions;
2119 if (rule->n_odp_actions != a.n_actions
2120 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2121 COVERAGE_INC(ofproto_odp_unchanged);
2122 free(rule->odp_actions);
2123 rule->n_odp_actions = a.n_actions;
2124 rule->odp_actions = xmemdup(a.actions, actions_len);
2132 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2133 struct odp_flow_put *put)
2135 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2136 odp_flow_key_from_flow(&put->flow.key, &rule->cr.flow);
2137 put->flow.actions = rule->odp_actions;
2138 put->flow.n_actions = rule->n_odp_actions;
2139 put->flow.flags = 0;
2141 return dpif_flow_put(ofproto->dpif, put);
2145 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2147 assert(!rule->cr.wc.wildcards);
2149 if (rule->may_install) {
2150 struct odp_flow_put put;
2151 if (!do_put_flow(p, rule,
2152 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2154 rule->installed = true;
2155 if (displaced_rule) {
2156 update_stats(p, displaced_rule, &put.flow.stats);
2157 rule_post_uninstall(p, displaced_rule);
2160 } else if (displaced_rule) {
2161 rule_uninstall(p, displaced_rule);
2166 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2168 if (rule->installed) {
2169 struct odp_flow_put put;
2170 COVERAGE_INC(ofproto_dp_missed);
2171 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2173 rule_install(ofproto, rule, NULL);
2178 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2180 bool actions_changed;
2181 uint16_t new_out_iface, old_out_iface;
2183 old_out_iface = rule->nf_flow.output_iface;
2184 actions_changed = rule_make_actions(ofproto, rule, NULL);
2186 if (rule->may_install) {
2187 if (rule->installed) {
2188 if (actions_changed) {
2189 struct odp_flow_put put;
2190 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2191 | ODPPF_ZERO_STATS, &put);
2192 update_stats(ofproto, rule, &put.flow.stats);
2194 /* Temporarily set the old output iface so that NetFlow
2195 * messages have the correct output interface for the old
2197 new_out_iface = rule->nf_flow.output_iface;
2198 rule->nf_flow.output_iface = old_out_iface;
2199 rule_post_uninstall(ofproto, rule);
2200 rule->nf_flow.output_iface = new_out_iface;
2203 rule_install(ofproto, rule, NULL);
2206 rule_uninstall(ofproto, rule);
2211 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2213 uint64_t total_bytes = rule->byte_count + extra_bytes;
2215 if (ofproto->ofhooks->account_flow_cb
2216 && total_bytes > rule->accounted_bytes)
2218 ofproto->ofhooks->account_flow_cb(
2219 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2220 total_bytes - rule->accounted_bytes, ofproto->aux);
2221 rule->accounted_bytes = total_bytes;
2225 /* 'rule' must be an exact-match rule in 'p'.
2227 * If 'rule' is installed in the datapath, uninstalls it and updates's
2228 * statistics. If 'rule' is a subrule, the statistics that are updated are
2229 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2232 * If 'rule' is not installed, this function has no effect. */
2234 rule_uninstall(struct ofproto *p, struct rule *rule)
2236 assert(!rule->cr.wc.wildcards);
2237 if (rule->installed) {
2238 struct odp_flow odp_flow;
2240 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
2241 odp_flow.actions = NULL;
2242 odp_flow.n_actions = 0;
2244 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2245 update_stats(p, rule, &odp_flow.stats);
2247 rule->installed = false;
2249 rule_post_uninstall(p, rule);
2254 is_controller_rule(struct rule *rule)
2256 /* If the only action is send to the controller then don't report
2257 * NetFlow expiration messages since it is just part of the control
2258 * logic for the network and not real traffic. */
2262 && rule->super->n_actions == 1
2263 && action_outputs_to_port(&rule->super->actions[0],
2264 htons(OFPP_CONTROLLER)));
2268 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2270 struct rule *super = rule->super;
2272 rule_account(ofproto, rule, 0);
2274 if (ofproto->netflow && !is_controller_rule(rule)) {
2275 struct ofexpired expired;
2276 expired.flow = rule->cr.flow;
2277 expired.packet_count = rule->packet_count;
2278 expired.byte_count = rule->byte_count;
2279 expired.used = rule->used;
2280 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2283 super->packet_count += rule->packet_count;
2284 super->byte_count += rule->byte_count;
2286 /* Reset counters to prevent double counting if the rule ever gets
2288 rule->packet_count = 0;
2289 rule->byte_count = 0;
2290 rule->accounted_bytes = 0;
2292 netflow_flow_clear(&rule->nf_flow);
2297 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2298 struct rconn_packet_counter *counter)
2300 update_openflow_length(msg);
2301 if (rconn_send(ofconn->rconn, msg, counter)) {
2307 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2308 int error, const void *data, size_t len)
2311 struct ofp_error_msg *oem;
2313 if (!(error >> 16)) {
2314 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2319 COVERAGE_INC(ofproto_error);
2320 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2321 oh ? oh->xid : 0, &buf);
2322 oem->type = htons((unsigned int) error >> 16);
2323 oem->code = htons(error & 0xffff);
2324 memcpy(oem->data, data, len);
2325 queue_tx(buf, ofconn, ofconn->reply_counter);
2329 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2332 size_t oh_length = ntohs(oh->length);
2333 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2337 hton_ofp_phy_port(struct ofp_phy_port *opp)
2339 opp->port_no = htons(opp->port_no);
2340 opp->config = htonl(opp->config);
2341 opp->state = htonl(opp->state);
2342 opp->curr = htonl(opp->curr);
2343 opp->advertised = htonl(opp->advertised);
2344 opp->supported = htonl(opp->supported);
2345 opp->peer = htonl(opp->peer);
2349 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2351 struct ofp_header *rq = oh;
2352 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2357 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2358 struct ofp_header *oh)
2360 struct ofp_switch_features *osf;
2362 struct ofport *port;
2364 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2365 osf->datapath_id = htonll(p->datapath_id);
2366 osf->n_buffers = htonl(pktbuf_capacity());
2368 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2369 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2370 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2371 (1u << OFPAT_SET_VLAN_VID) |
2372 (1u << OFPAT_SET_VLAN_PCP) |
2373 (1u << OFPAT_STRIP_VLAN) |
2374 (1u << OFPAT_SET_DL_SRC) |
2375 (1u << OFPAT_SET_DL_DST) |
2376 (1u << OFPAT_SET_NW_SRC) |
2377 (1u << OFPAT_SET_NW_DST) |
2378 (1u << OFPAT_SET_NW_TOS) |
2379 (1u << OFPAT_SET_TP_SRC) |
2380 (1u << OFPAT_SET_TP_DST) |
2381 (1u << OFPAT_ENQUEUE));
2383 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2384 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2387 queue_tx(buf, ofconn, ofconn->reply_counter);
2392 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2393 struct ofp_header *oh)
2396 struct ofp_switch_config *osc;
2400 /* Figure out flags. */
2401 dpif_get_drop_frags(p->dpif, &drop_frags);
2402 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2405 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2406 osc->flags = htons(flags);
2407 osc->miss_send_len = htons(ofconn->miss_send_len);
2408 queue_tx(buf, ofconn, ofconn->reply_counter);
2414 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2415 struct ofp_switch_config *osc)
2420 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2424 flags = ntohs(osc->flags);
2426 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2427 switch (flags & OFPC_FRAG_MASK) {
2428 case OFPC_FRAG_NORMAL:
2429 dpif_set_drop_frags(p->dpif, false);
2431 case OFPC_FRAG_DROP:
2432 dpif_set_drop_frags(p->dpif, true);
2435 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2441 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2447 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2449 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2450 a->controller.arg = max_len;
2453 struct action_xlate_ctx {
2455 struct flow flow; /* Flow to which these actions correspond. */
2456 int recurse; /* Recursion level, via xlate_table_action. */
2457 struct ofproto *ofproto;
2458 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2459 * null pointer if we are revalidating
2460 * without a packet to refer to. */
2463 struct odp_actions *out; /* Datapath actions. */
2464 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2465 bool may_set_up_flow; /* True ordinarily; false if the actions must
2466 * be reassessed for every packet. */
2467 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2470 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2471 * flow translation. */
2472 #define MAX_RESUBMIT_RECURSION 8
2474 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2475 struct action_xlate_ctx *ctx);
2478 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2480 const struct ofport *ofport = get_port(ctx->ofproto, port);
2483 if (ofport->opp.config & OFPPC_NO_FWD) {
2484 /* Forwarding disabled on port. */
2489 * We don't have an ofport record for this port, but it doesn't hurt to
2490 * allow forwarding to it anyhow. Maybe such a port will appear later
2491 * and we're pre-populating the flow table.
2495 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2496 ctx->nf_output_iface = port;
2499 static struct rule *
2500 lookup_valid_rule(struct ofproto *ofproto, const struct flow *flow)
2503 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow,
2506 /* The rule we found might not be valid, since we could be in need of
2507 * revalidation. If it is not valid, don't return it. */
2510 && ofproto->need_revalidate
2511 && !revalidate_rule(ofproto, rule)) {
2512 COVERAGE_INC(ofproto_invalidated);
2520 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2522 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2523 uint16_t old_in_port;
2526 /* Look up a flow with 'in_port' as the input port. Then restore the
2527 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2528 * have surprising behavior). */
2529 old_in_port = ctx->flow.in_port;
2530 ctx->flow.in_port = in_port;
2531 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2532 ctx->flow.in_port = old_in_port;
2540 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2544 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2546 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2547 MAX_RESUBMIT_RECURSION);
2552 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2553 uint16_t *nf_output_iface, struct odp_actions *actions)
2555 struct ofport *ofport;
2557 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2558 uint16_t odp_port = ofport->odp_port;
2559 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2560 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2563 *nf_output_iface = NF_OUT_FLOOD;
2567 xlate_output_action__(struct action_xlate_ctx *ctx,
2568 uint16_t port, uint16_t max_len)
2571 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2573 ctx->nf_output_iface = NF_OUT_DROP;
2577 add_output_action(ctx, ctx->flow.in_port);
2580 xlate_table_action(ctx, ctx->flow.in_port);
2583 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2584 ctx->out, ctx->tags,
2585 &ctx->nf_output_iface,
2586 ctx->ofproto->aux)) {
2587 COVERAGE_INC(ofproto_uninstallable);
2588 ctx->may_set_up_flow = false;
2592 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2593 &ctx->nf_output_iface, ctx->out);
2596 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2597 &ctx->nf_output_iface, ctx->out);
2599 case OFPP_CONTROLLER:
2600 add_controller_action(ctx->out, max_len);
2603 add_output_action(ctx, ODPP_LOCAL);
2606 odp_port = ofp_port_to_odp_port(port);
2607 if (odp_port != ctx->flow.in_port) {
2608 add_output_action(ctx, odp_port);
2613 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2614 ctx->nf_output_iface = NF_OUT_FLOOD;
2615 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2616 ctx->nf_output_iface = prev_nf_output_iface;
2617 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2618 ctx->nf_output_iface != NF_OUT_FLOOD) {
2619 ctx->nf_output_iface = NF_OUT_MULTI;
2624 xlate_output_action(struct action_xlate_ctx *ctx,
2625 const struct ofp_action_output *oao)
2627 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2630 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2631 * optimization, because we're going to add another action that sets the
2632 * priority immediately after, or because there are no actions following the
2635 remove_pop_action(struct action_xlate_ctx *ctx)
2637 size_t n = ctx->out->n_actions;
2638 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2639 ctx->out->n_actions--;
2644 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2645 const struct ofp_action_enqueue *oae)
2647 uint16_t ofp_port, odp_port;
2651 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2654 /* Fall back to ordinary output action. */
2655 xlate_output_action__(ctx, ntohs(oae->port), 0);
2659 /* Figure out ODP output port. */
2660 ofp_port = ntohs(oae->port);
2661 if (ofp_port != OFPP_IN_PORT) {
2662 odp_port = ofp_port_to_odp_port(ofp_port);
2664 odp_port = ctx->flow.in_port;
2667 /* Add ODP actions. */
2668 remove_pop_action(ctx);
2669 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2671 add_output_action(ctx, odp_port);
2672 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2674 /* Update NetFlow output port. */
2675 if (ctx->nf_output_iface == NF_OUT_DROP) {
2676 ctx->nf_output_iface = odp_port;
2677 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2678 ctx->nf_output_iface = NF_OUT_MULTI;
2683 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2684 const struct nx_action_set_queue *nasq)
2689 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2692 /* Couldn't translate queue to a priority, so ignore. A warning
2693 * has already been logged. */
2697 remove_pop_action(ctx);
2698 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2703 xlate_nicira_action(struct action_xlate_ctx *ctx,
2704 const struct nx_action_header *nah)
2706 const struct nx_action_resubmit *nar;
2707 const struct nx_action_set_tunnel *nast;
2708 const struct nx_action_set_queue *nasq;
2709 union odp_action *oa;
2710 int subtype = ntohs(nah->subtype);
2712 assert(nah->vendor == htonl(NX_VENDOR_ID));
2714 case NXAST_RESUBMIT:
2715 nar = (const struct nx_action_resubmit *) nah;
2716 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2719 case NXAST_SET_TUNNEL:
2720 nast = (const struct nx_action_set_tunnel *) nah;
2721 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2722 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2725 case NXAST_DROP_SPOOFED_ARP:
2726 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2727 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2731 case NXAST_SET_QUEUE:
2732 nasq = (const struct nx_action_set_queue *) nah;
2733 xlate_set_queue_action(ctx, nasq);
2736 case NXAST_POP_QUEUE:
2737 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2740 /* If you add a new action here that modifies flow data, don't forget to
2741 * update the flow key in ctx->flow at the same time. */
2744 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2750 do_xlate_actions(const union ofp_action *in, size_t n_in,
2751 struct action_xlate_ctx *ctx)
2753 struct actions_iterator iter;
2754 const union ofp_action *ia;
2755 const struct ofport *port;
2757 port = get_port(ctx->ofproto, ctx->flow.in_port);
2758 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2759 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2760 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2761 /* Drop this flow. */
2765 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2766 uint16_t type = ntohs(ia->type);
2767 union odp_action *oa;
2771 xlate_output_action(ctx, &ia->output);
2774 case OFPAT_SET_VLAN_VID:
2775 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2776 oa->dl_tci.tci = ia->vlan_vid.vlan_vid;
2777 oa->dl_tci.tci |= htons(ctx->flow.dl_vlan_pcp << VLAN_PCP_SHIFT);
2778 ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
2781 case OFPAT_SET_VLAN_PCP:
2782 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2783 oa->dl_tci.tci = htons(ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT);
2784 oa->dl_tci.tci |= ctx->flow.dl_vlan;
2785 ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
2788 case OFPAT_STRIP_VLAN:
2789 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2790 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2791 ctx->flow.dl_vlan_pcp = 0;
2794 case OFPAT_SET_DL_SRC:
2795 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2796 memcpy(oa->dl_addr.dl_addr,
2797 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2798 memcpy(ctx->flow.dl_src,
2799 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2802 case OFPAT_SET_DL_DST:
2803 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2804 memcpy(oa->dl_addr.dl_addr,
2805 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2806 memcpy(ctx->flow.dl_dst,
2807 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2810 case OFPAT_SET_NW_SRC:
2811 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2812 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2815 case OFPAT_SET_NW_DST:
2816 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2817 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2820 case OFPAT_SET_NW_TOS:
2821 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2822 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2825 case OFPAT_SET_TP_SRC:
2826 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2827 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2830 case OFPAT_SET_TP_DST:
2831 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2832 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2836 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2840 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2844 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2851 xlate_actions(const union ofp_action *in, size_t n_in,
2852 const struct flow *flow, struct ofproto *ofproto,
2853 const struct ofpbuf *packet,
2854 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2855 uint16_t *nf_output_iface)
2857 tag_type no_tags = 0;
2858 struct action_xlate_ctx ctx;
2859 COVERAGE_INC(ofproto_ofp2odp);
2860 odp_actions_init(out);
2863 ctx.ofproto = ofproto;
2864 ctx.packet = packet;
2866 ctx.tags = tags ? tags : &no_tags;
2867 ctx.may_set_up_flow = true;
2868 ctx.nf_output_iface = NF_OUT_DROP;
2869 do_xlate_actions(in, n_in, &ctx);
2870 remove_pop_action(&ctx);
2872 /* Check with in-band control to see if we're allowed to set up this
2874 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2875 ctx.may_set_up_flow = false;
2878 if (may_set_up_flow) {
2879 *may_set_up_flow = ctx.may_set_up_flow;
2881 if (nf_output_iface) {
2882 *nf_output_iface = ctx.nf_output_iface;
2884 if (odp_actions_overflow(out)) {
2885 COVERAGE_INC(odp_overflow);
2886 odp_actions_init(out);
2887 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2892 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2893 * error message code (composed with ofp_mkerr()) for the caller to propagate
2894 * upward. Otherwise, returns 0.
2896 * 'oh' is used to make log messages more informative. */
2898 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2900 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2901 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2904 type_name = ofp_message_type_to_string(oh->type);
2905 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2909 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2916 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2917 struct ofp_header *oh)
2919 struct ofp_packet_out *opo;
2920 struct ofpbuf payload, *buffer;
2921 struct odp_actions actions;
2927 error = reject_slave_controller(ofconn, oh);
2932 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2936 opo = (struct ofp_packet_out *) oh;
2938 COVERAGE_INC(ofproto_packet_out);
2939 if (opo->buffer_id != htonl(UINT32_MAX)) {
2940 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2942 if (error || !buffer) {
2950 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2951 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2952 &flow, p, &payload, &actions, NULL, NULL, NULL);
2954 dpif_execute(p->dpif, actions.actions, actions.n_actions, &payload);
2956 ofpbuf_delete(buffer);
2962 update_port_config(struct ofproto *p, struct ofport *port,
2963 uint32_t config, uint32_t mask)
2965 mask &= config ^ port->opp.config;
2966 if (mask & OFPPC_PORT_DOWN) {
2967 if (config & OFPPC_PORT_DOWN) {
2968 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2970 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2973 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2974 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2975 if (mask & REVALIDATE_BITS) {
2976 COVERAGE_INC(ofproto_costly_flags);
2977 port->opp.config ^= mask & REVALIDATE_BITS;
2978 p->need_revalidate = true;
2980 #undef REVALIDATE_BITS
2981 if (mask & OFPPC_NO_PACKET_IN) {
2982 port->opp.config ^= OFPPC_NO_PACKET_IN;
2987 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2988 struct ofp_header *oh)
2990 const struct ofp_port_mod *opm;
2991 struct ofport *port;
2994 error = reject_slave_controller(ofconn, oh);
2998 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
3002 opm = (struct ofp_port_mod *) oh;
3004 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3006 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3007 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3008 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3010 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3011 if (opm->advertise) {
3012 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3018 static struct ofpbuf *
3019 make_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3021 struct ofp_stats_reply *osr;
3024 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3025 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3027 osr->flags = htons(0);
3031 static struct ofpbuf *
3032 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3034 return make_stats_reply(request->header.xid, request->type, body_len);
3038 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
3040 struct ofpbuf *msg = *msgp;
3041 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3042 if (nbytes + msg->size > UINT16_MAX) {
3043 struct ofp_stats_reply *reply = msg->data;
3044 reply->flags = htons(OFPSF_REPLY_MORE);
3045 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3046 queue_tx(msg, ofconn, ofconn->reply_counter);
3048 return ofpbuf_put_uninit(*msgp, nbytes);
3052 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3053 struct ofp_stats_request *request)
3055 struct ofp_desc_stats *ods;
3058 msg = start_stats_reply(request, sizeof *ods);
3059 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3060 memset(ods, 0, sizeof *ods);
3061 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3062 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3063 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3064 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3065 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3066 queue_tx(msg, ofconn, ofconn->reply_counter);
3072 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3073 struct ofp_stats_request *request)
3075 struct ofp_table_stats *ots;
3077 struct odp_stats dpstats;
3078 int n_exact, n_subrules, n_wild;
3081 msg = start_stats_reply(request, sizeof *ots * 2);
3083 /* Count rules of various kinds. */
3085 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) {
3090 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3091 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3094 dpif_get_dp_stats(p->dpif, &dpstats);
3095 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3096 memset(ots, 0, sizeof *ots);
3097 ots->table_id = TABLEID_HASH;
3098 strcpy(ots->name, "hash");
3099 ots->wildcards = htonl(0);
3100 ots->max_entries = htonl(dpstats.max_capacity);
3101 ots->active_count = htonl(n_exact);
3102 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3104 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3106 /* Classifier table. */
3107 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3108 memset(ots, 0, sizeof *ots);
3109 ots->table_id = TABLEID_CLASSIFIER;
3110 strcpy(ots->name, "classifier");
3111 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3113 ots->max_entries = htonl(65536);
3114 ots->active_count = htonl(n_wild);
3115 ots->lookup_count = htonll(0); /* XXX */
3116 ots->matched_count = htonll(0); /* XXX */
3118 queue_tx(msg, ofconn, ofconn->reply_counter);
3123 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3124 struct ofpbuf **msgp)
3126 struct netdev_stats stats;
3127 struct ofp_port_stats *ops;
3129 /* Intentionally ignore return value, since errors will set
3130 * 'stats' to all-1s, which is correct for OpenFlow, and
3131 * netdev_get_stats() will log errors. */
3132 netdev_get_stats(port->netdev, &stats);
3134 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3135 ops->port_no = htons(port->opp.port_no);
3136 memset(ops->pad, 0, sizeof ops->pad);
3137 ops->rx_packets = htonll(stats.rx_packets);
3138 ops->tx_packets = htonll(stats.tx_packets);
3139 ops->rx_bytes = htonll(stats.rx_bytes);
3140 ops->tx_bytes = htonll(stats.tx_bytes);
3141 ops->rx_dropped = htonll(stats.rx_dropped);
3142 ops->tx_dropped = htonll(stats.tx_dropped);
3143 ops->rx_errors = htonll(stats.rx_errors);
3144 ops->tx_errors = htonll(stats.tx_errors);
3145 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3146 ops->rx_over_err = htonll(stats.rx_over_errors);
3147 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3148 ops->collisions = htonll(stats.collisions);
3152 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3153 struct ofp_stats_request *osr,
3156 struct ofp_port_stats_request *psr;
3157 struct ofp_port_stats *ops;
3159 struct ofport *port;
3161 if (arg_size != sizeof *psr) {
3162 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3164 psr = (struct ofp_port_stats_request *) osr->body;
3166 msg = start_stats_reply(osr, sizeof *ops * 16);
3167 if (psr->port_no != htons(OFPP_NONE)) {
3168 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3170 append_port_stat(port, ofconn, &msg);
3173 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3174 append_port_stat(port, ofconn, &msg);
3178 queue_tx(msg, ofconn, ofconn->reply_counter);
3182 struct flow_stats_cbdata {
3183 struct ofproto *ofproto;
3184 struct ofconn *ofconn;
3189 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3190 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3191 * returned statistic include statistics for all of 'rule''s subrules. */
3193 query_stats(struct ofproto *p, struct rule *rule,
3194 uint64_t *packet_countp, uint64_t *byte_countp)
3196 uint64_t packet_count, byte_count;
3197 struct rule *subrule;
3198 struct odp_flow *odp_flows;
3201 /* Start from historical data for 'rule' itself that are no longer tracked
3202 * by the datapath. This counts, for example, subrules that have
3204 packet_count = rule->packet_count;
3205 byte_count = rule->byte_count;
3207 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3208 * wildcarded then on all of its subrules.
3210 * Also, add any statistics that are not tracked by the datapath for each
3211 * subrule. This includes, for example, statistics for packets that were
3212 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3214 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3215 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3216 if (rule->cr.wc.wildcards) {
3218 LIST_FOR_EACH (subrule, list, &rule->list) {
3219 odp_flow_key_from_flow(&odp_flows[i++].key, &subrule->cr.flow);
3220 packet_count += subrule->packet_count;
3221 byte_count += subrule->byte_count;
3224 odp_flow_key_from_flow(&odp_flows[0].key, &rule->cr.flow);
3227 /* Fetch up-to-date statistics from the datapath and add them in. */
3228 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3230 for (i = 0; i < n_odp_flows; i++) {
3231 struct odp_flow *odp_flow = &odp_flows[i];
3232 packet_count += odp_flow->stats.n_packets;
3233 byte_count += odp_flow->stats.n_bytes;
3238 /* Return the stats to the caller. */
3239 *packet_countp = packet_count;
3240 *byte_countp = byte_count;
3244 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3246 struct rule *rule = rule_from_cls_rule(rule_);
3247 struct flow_stats_cbdata *cbdata = cbdata_;
3248 struct ofp_flow_stats *ofs;
3249 uint64_t packet_count, byte_count;
3250 size_t act_len, len;
3251 long long int tdiff = time_msec() - rule->created;
3252 uint32_t sec = tdiff / 1000;
3253 uint32_t msec = tdiff - (sec * 1000);
3255 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3259 act_len = sizeof *rule->actions * rule->n_actions;
3260 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3262 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3264 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3265 ofs->length = htons(len);
3266 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3268 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3269 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3270 ofs->duration_sec = htonl(sec);
3271 ofs->duration_nsec = htonl(msec * 1000000);
3272 ofs->cookie = rule->flow_cookie;
3273 ofs->priority = htons(rule->cr.priority);
3274 ofs->idle_timeout = htons(rule->idle_timeout);
3275 ofs->hard_timeout = htons(rule->hard_timeout);
3276 memset(ofs->pad2, 0, sizeof ofs->pad2);
3277 ofs->packet_count = htonll(packet_count);
3278 ofs->byte_count = htonll(byte_count);
3279 if (rule->n_actions > 0) {
3280 memcpy(ofs->actions, rule->actions, act_len);
3285 table_id_to_include(uint8_t table_id)
3287 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3288 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3289 : table_id == 0xff ? CLS_INC_ALL
3294 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3295 const struct ofp_stats_request *osr,
3298 struct ofp_flow_stats_request *fsr;
3299 struct flow_stats_cbdata cbdata;
3300 struct cls_rule target;
3302 if (arg_size != sizeof *fsr) {
3303 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3305 fsr = (struct ofp_flow_stats_request *) osr->body;
3307 COVERAGE_INC(ofproto_flows_req);
3309 cbdata.ofconn = ofconn;
3310 cbdata.out_port = fsr->out_port;
3311 cbdata.msg = start_stats_reply(osr, 1024);
3312 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3313 classifier_for_each_match(&p->cls, &target,
3314 table_id_to_include(fsr->table_id),
3315 flow_stats_cb, &cbdata);
3316 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3320 struct flow_stats_ds_cbdata {
3321 struct ofproto *ofproto;
3326 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3328 struct rule *rule = rule_from_cls_rule(rule_);
3329 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3330 struct ds *results = cbdata->results;
3331 struct ofp_match match;
3332 uint64_t packet_count, byte_count;
3333 size_t act_len = sizeof *rule->actions * rule->n_actions;
3335 /* Don't report on subrules. */
3336 if (rule->super != NULL) {
3340 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3341 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3342 cbdata->ofproto->tun_id_from_cookie, &match);
3344 ds_put_format(results, "duration=%llds, ",
3345 (time_msec() - rule->created) / 1000);
3346 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3347 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3348 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3349 ofp_print_match(results, &match, true);
3351 ofp_print_actions(results, &rule->actions->header, act_len);
3353 ds_put_cstr(results, "drop");
3355 ds_put_cstr(results, "\n");
3358 /* Adds a pretty-printed description of all flows to 'results', including
3359 * those marked hidden by secchan (e.g., by in-band control). */
3361 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3363 struct ofp_match match;
3364 struct cls_rule target;
3365 struct flow_stats_ds_cbdata cbdata;
3367 memset(&match, 0, sizeof match);
3368 match.wildcards = htonl(OVSFW_ALL);
3371 cbdata.results = results;
3373 cls_rule_from_match(&match, 0, false, 0, &target);
3374 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3375 flow_stats_ds_cb, &cbdata);
3378 struct aggregate_stats_cbdata {
3379 struct ofproto *ofproto;
3381 uint64_t packet_count;
3382 uint64_t byte_count;
3387 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3389 struct rule *rule = rule_from_cls_rule(rule_);
3390 struct aggregate_stats_cbdata *cbdata = cbdata_;
3391 uint64_t packet_count, byte_count;
3393 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3397 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3399 cbdata->packet_count += packet_count;
3400 cbdata->byte_count += byte_count;
3405 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3406 const struct ofp_stats_request *osr,
3409 struct ofp_aggregate_stats_request *asr;
3410 struct ofp_aggregate_stats_reply *reply;
3411 struct aggregate_stats_cbdata cbdata;
3412 struct cls_rule target;
3415 if (arg_size != sizeof *asr) {
3416 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3418 asr = (struct ofp_aggregate_stats_request *) osr->body;
3420 COVERAGE_INC(ofproto_agg_request);
3422 cbdata.out_port = asr->out_port;
3423 cbdata.packet_count = 0;
3424 cbdata.byte_count = 0;
3426 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3427 classifier_for_each_match(&p->cls, &target,
3428 table_id_to_include(asr->table_id),
3429 aggregate_stats_cb, &cbdata);
3431 msg = start_stats_reply(osr, sizeof *reply);
3432 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3433 reply->flow_count = htonl(cbdata.n_flows);
3434 reply->packet_count = htonll(cbdata.packet_count);
3435 reply->byte_count = htonll(cbdata.byte_count);
3436 queue_tx(msg, ofconn, ofconn->reply_counter);
3440 struct queue_stats_cbdata {
3441 struct ofconn *ofconn;
3442 struct ofport *ofport;
3447 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3448 const struct netdev_queue_stats *stats)
3450 struct ofp_queue_stats *reply;
3452 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3453 reply->port_no = htons(cbdata->ofport->opp.port_no);
3454 memset(reply->pad, 0, sizeof reply->pad);
3455 reply->queue_id = htonl(queue_id);
3456 reply->tx_bytes = htonll(stats->tx_bytes);
3457 reply->tx_packets = htonll(stats->tx_packets);
3458 reply->tx_errors = htonll(stats->tx_errors);
3462 handle_queue_stats_dump_cb(uint32_t queue_id,
3463 struct netdev_queue_stats *stats,
3466 struct queue_stats_cbdata *cbdata = cbdata_;
3468 put_queue_stats(cbdata, queue_id, stats);
3472 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3473 struct queue_stats_cbdata *cbdata)
3475 cbdata->ofport = port;
3476 if (queue_id == OFPQ_ALL) {
3477 netdev_dump_queue_stats(port->netdev,
3478 handle_queue_stats_dump_cb, cbdata);
3480 struct netdev_queue_stats stats;
3482 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3483 put_queue_stats(cbdata, queue_id, &stats);
3489 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3490 const struct ofp_stats_request *osr,
3493 struct ofp_queue_stats_request *qsr;
3494 struct queue_stats_cbdata cbdata;
3495 struct ofport *port;
3496 unsigned int port_no;
3499 if (arg_size != sizeof *qsr) {
3500 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3502 qsr = (struct ofp_queue_stats_request *) osr->body;
3504 COVERAGE_INC(ofproto_queue_req);
3506 cbdata.ofconn = ofconn;
3507 cbdata.msg = start_stats_reply(osr, 128);
3509 port_no = ntohs(qsr->port_no);
3510 queue_id = ntohl(qsr->queue_id);
3511 if (port_no == OFPP_ALL) {
3512 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3513 handle_queue_stats_for_port(port, queue_id, &cbdata);
3515 } else if (port_no < ofproto->max_ports) {
3516 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3518 handle_queue_stats_for_port(port, queue_id, &cbdata);
3521 ofpbuf_delete(cbdata.msg);
3522 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3524 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3530 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3531 struct ofp_header *oh)
3533 struct ofp_stats_request *osr;
3537 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3542 osr = (struct ofp_stats_request *) oh;
3544 switch (ntohs(osr->type)) {
3546 return handle_desc_stats_request(p, ofconn, osr);
3549 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3551 case OFPST_AGGREGATE:
3552 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3555 return handle_table_stats_request(p, ofconn, osr);
3558 return handle_port_stats_request(p, ofconn, osr, arg_size);
3561 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3564 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3567 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3571 static long long int
3572 msec_from_nsec(uint64_t sec, uint32_t nsec)
3574 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3578 update_time(struct ofproto *ofproto, struct rule *rule,
3579 const struct odp_flow_stats *stats)
3581 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3582 if (used > rule->used) {
3584 if (rule->super && used > rule->super->used) {
3585 rule->super->used = used;
3587 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3592 update_stats(struct ofproto *ofproto, struct rule *rule,
3593 const struct odp_flow_stats *stats)
3595 if (stats->n_packets) {
3596 update_time(ofproto, rule, stats);
3597 rule->packet_count += stats->n_packets;
3598 rule->byte_count += stats->n_bytes;
3599 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3603 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3604 * in which no matching flow already exists in the flow table.
3606 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3607 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3608 * code as encoded by ofp_mkerr() on failure.
3610 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3613 add_flow(struct ofproto *p, struct ofconn *ofconn,
3614 const struct ofp_flow_mod *ofm, size_t n_actions)
3616 struct ofpbuf *packet;
3621 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3625 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3627 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3628 ntohs(ofm->priority))) {
3629 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3633 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3634 n_actions, ntohs(ofm->idle_timeout),
3635 ntohs(ofm->hard_timeout), ofm->cookie,
3636 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3637 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3638 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3641 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3642 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3646 in_port = UINT16_MAX;
3649 rule_insert(p, rule, packet, in_port);
3653 static struct rule *
3654 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3656 struct cls_rule target;
3658 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3659 p->tun_id_from_cookie, ofm->cookie, &target);
3660 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &target));
3664 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3665 struct rule *rule, const struct ofp_flow_mod *ofm)
3667 struct ofpbuf *packet;
3672 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3676 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3682 flow_extract(packet, 0, in_port, &flow);
3683 rule_execute(ofproto, rule, packet, &flow);
3688 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3690 struct modify_flows_cbdata {
3691 struct ofproto *ofproto;
3692 const struct ofp_flow_mod *ofm;
3697 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3698 size_t n_actions, struct rule *);
3699 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3701 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3702 * encoded by ofp_mkerr() on failure.
3704 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3707 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3708 const struct ofp_flow_mod *ofm, size_t n_actions)
3710 struct modify_flows_cbdata cbdata;
3711 struct cls_rule target;
3715 cbdata.n_actions = n_actions;
3716 cbdata.match = NULL;
3718 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3721 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3722 modify_flows_cb, &cbdata);
3724 /* This credits the packet to whichever flow happened to happened to
3725 * match last. That's weird. Maybe we should do a lookup for the
3726 * flow that actually matches the packet? Who knows. */
3727 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3730 return add_flow(p, ofconn, ofm, n_actions);
3734 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3735 * code as encoded by ofp_mkerr() on failure.
3737 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3740 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3741 struct ofp_flow_mod *ofm, size_t n_actions)
3743 struct rule *rule = find_flow_strict(p, ofm);
3744 if (rule && !rule_is_hidden(rule)) {
3745 modify_flow(p, ofm, n_actions, rule);
3746 return send_buffered_packet(p, ofconn, rule, ofm);
3748 return add_flow(p, ofconn, ofm, n_actions);
3752 /* Callback for modify_flows_loose(). */
3754 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3756 struct rule *rule = rule_from_cls_rule(rule_);
3757 struct modify_flows_cbdata *cbdata = cbdata_;
3759 if (!rule_is_hidden(rule)) {
3760 cbdata->match = rule;
3761 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3765 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3766 * been identified as a flow in 'p''s flow table to be modified, by changing
3767 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3768 * ofp_action[] structures). */
3770 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3771 size_t n_actions, struct rule *rule)
3773 size_t actions_len = n_actions * sizeof *rule->actions;
3775 rule->flow_cookie = ofm->cookie;
3777 /* If the actions are the same, do nothing. */
3778 if (n_actions == rule->n_actions
3779 && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len)))
3784 /* Replace actions. */
3785 free(rule->actions);
3786 rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL;
3787 rule->n_actions = n_actions;
3789 /* Make sure that the datapath gets updated properly. */
3790 if (rule->cr.wc.wildcards) {
3791 COVERAGE_INC(ofproto_mod_wc_flow);
3792 p->need_revalidate = true;
3794 rule_update_actions(p, rule);
3800 /* OFPFC_DELETE implementation. */
3802 struct delete_flows_cbdata {
3803 struct ofproto *ofproto;
3807 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3808 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3810 /* Implements OFPFC_DELETE. */
3812 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3814 struct delete_flows_cbdata cbdata;
3815 struct cls_rule target;
3818 cbdata.out_port = ofm->out_port;
3820 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3823 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3824 delete_flows_cb, &cbdata);
3827 /* Implements OFPFC_DELETE_STRICT. */
3829 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3831 struct rule *rule = find_flow_strict(p, ofm);
3833 delete_flow(p, rule, ofm->out_port);
3837 /* Callback for delete_flows_loose(). */
3839 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3841 struct rule *rule = rule_from_cls_rule(rule_);
3842 struct delete_flows_cbdata *cbdata = cbdata_;
3844 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3847 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3848 * been identified as a flow to delete from 'p''s flow table, by deleting the
3849 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3852 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3853 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3854 * specified 'out_port'. */
3856 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3858 if (rule_is_hidden(rule)) {
3862 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3866 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3867 rule_remove(p, rule);
3871 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3872 struct ofp_flow_mod *ofm)
3874 struct ofp_match orig_match;
3878 error = reject_slave_controller(ofconn, &ofm->header);
3882 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3883 sizeof *ofm->actions, &n_actions);
3888 /* We do not support the emergency flow cache. It will hopefully
3889 * get dropped from OpenFlow in the near future. */
3890 if (ofm->flags & htons(OFPFF_EMERG)) {
3891 /* There isn't a good fit for an error code, so just state that the
3892 * flow table is full. */
3893 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3896 /* Normalize ofp->match. If normalization actually changes anything, then
3897 * log the differences. */
3898 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3899 orig_match = ofm->match;
3900 normalize_match(&ofm->match);
3901 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3902 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3903 if (!VLOG_DROP_INFO(&normal_rl)) {
3904 char *old = ofp_match_to_literal_string(&orig_match);
3905 char *new = ofp_match_to_literal_string(&ofm->match);
3906 VLOG_INFO("%s: normalization changed ofp_match, details:",
3907 rconn_get_name(ofconn->rconn));
3908 VLOG_INFO(" pre: %s", old);
3909 VLOG_INFO("post: %s", new);
3915 if (!ofm->match.wildcards) {
3916 ofm->priority = htons(UINT16_MAX);
3919 error = validate_actions((const union ofp_action *) ofm->actions,
3920 n_actions, p->max_ports);
3925 switch (ntohs(ofm->command)) {
3927 return add_flow(p, ofconn, ofm, n_actions);
3930 return modify_flows_loose(p, ofconn, ofm, n_actions);
3932 case OFPFC_MODIFY_STRICT:
3933 return modify_flow_strict(p, ofconn, ofm, n_actions);
3936 delete_flows_loose(p, ofm);
3939 case OFPFC_DELETE_STRICT:
3940 delete_flow_strict(p, ofm);
3944 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3949 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3953 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3958 p->tun_id_from_cookie = !!msg->set;
3963 handle_role_request(struct ofproto *ofproto,
3964 struct ofconn *ofconn, struct nicira_header *msg)
3966 struct nx_role_request *nrr;
3967 struct nx_role_request *reply;
3971 if (ntohs(msg->header.length) != sizeof *nrr) {
3972 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3973 ntohs(msg->header.length), sizeof *nrr);
3974 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3976 nrr = (struct nx_role_request *) msg;
3978 if (ofconn->type != OFCONN_PRIMARY) {
3979 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3981 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3984 role = ntohl(nrr->role);
3985 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3986 && role != NX_ROLE_SLAVE) {
3987 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3989 /* There's no good error code for this. */
3990 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3993 if (role == NX_ROLE_MASTER) {
3994 struct ofconn *other;
3996 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
3997 if (other->role == NX_ROLE_MASTER) {
3998 other->role = NX_ROLE_SLAVE;
4002 ofconn->role = role;
4004 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
4006 reply->nxh.vendor = htonl(NX_VENDOR_ID);
4007 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
4008 reply->role = htonl(role);
4009 queue_tx(buf, ofconn, ofconn->reply_counter);
4015 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
4017 struct ofp_vendor_header *ovh = msg;
4018 struct nicira_header *nh;
4020 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4021 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4022 "(expected at least %zu)",
4023 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4024 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4026 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4027 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4029 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4030 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4031 "(expected at least %zu)",
4032 ntohs(ovh->header.length), sizeof(struct nicira_header));
4033 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4037 switch (ntohl(nh->subtype)) {
4038 case NXT_STATUS_REQUEST:
4039 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4042 case NXT_TUN_ID_FROM_COOKIE:
4043 return handle_tun_id_from_cookie(p, msg);
4045 case NXT_ROLE_REQUEST:
4046 return handle_role_request(p, ofconn, msg);
4049 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4053 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4055 struct ofp_header *ob;
4058 /* Currently, everything executes synchronously, so we can just
4059 * immediately send the barrier reply. */
4060 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4061 queue_tx(buf, ofconn, ofconn->reply_counter);
4066 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4067 struct ofpbuf *ofp_msg)
4069 struct ofp_header *oh = ofp_msg->data;
4072 COVERAGE_INC(ofproto_recv_openflow);
4074 case OFPT_ECHO_REQUEST:
4075 error = handle_echo_request(ofconn, oh);
4078 case OFPT_ECHO_REPLY:
4082 case OFPT_FEATURES_REQUEST:
4083 error = handle_features_request(p, ofconn, oh);
4086 case OFPT_GET_CONFIG_REQUEST:
4087 error = handle_get_config_request(p, ofconn, oh);
4090 case OFPT_SET_CONFIG:
4091 error = handle_set_config(p, ofconn, ofp_msg->data);
4094 case OFPT_PACKET_OUT:
4095 error = handle_packet_out(p, ofconn, ofp_msg->data);
4099 error = handle_port_mod(p, ofconn, oh);
4103 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4106 case OFPT_STATS_REQUEST:
4107 error = handle_stats_request(p, ofconn, oh);
4111 error = handle_vendor(p, ofconn, ofp_msg->data);
4114 case OFPT_BARRIER_REQUEST:
4115 error = handle_barrier_request(ofconn, oh);
4119 if (VLOG_IS_WARN_ENABLED()) {
4120 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4121 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4124 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4129 send_error_oh(ofconn, ofp_msg->data, error);
4134 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4136 struct odp_msg *msg = packet->data;
4138 struct ofpbuf payload;
4141 payload.data = msg + 1;
4142 payload.size = msg->length - sizeof *msg;
4143 flow_extract(&payload, msg->arg, msg->port, &flow);
4145 /* Check with in-band control to see if this packet should be sent
4146 * to the local port regardless of the flow table. */
4147 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4148 union odp_action action;
4150 memset(&action, 0, sizeof(action));
4151 action.output.type = ODPAT_OUTPUT;
4152 action.output.port = ODPP_LOCAL;
4153 dpif_execute(p->dpif, &action, 1, &payload);
4156 rule = lookup_valid_rule(p, &flow);
4158 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4159 struct ofport *port = get_port(p, msg->port);
4161 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4162 COVERAGE_INC(ofproto_no_packet_in);
4163 /* XXX install 'drop' flow entry */
4164 ofpbuf_delete(packet);
4168 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4171 COVERAGE_INC(ofproto_packet_in);
4172 send_packet_in(p, packet);
4176 if (rule->cr.wc.wildcards) {
4177 rule = rule_create_subrule(p, rule, &flow);
4178 rule_make_actions(p, rule, packet);
4180 if (!rule->may_install) {
4181 /* The rule is not installable, that is, we need to process every
4182 * packet, so process the current packet and set its actions into
4184 rule_make_actions(p, rule, packet);
4186 /* XXX revalidate rule if it needs it */
4190 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4192 * Extra-special case for fail-open mode.
4194 * We are in fail-open mode and the packet matched the fail-open rule,
4195 * but we are connected to a controller too. We should send the packet
4196 * up to the controller in the hope that it will try to set up a flow
4197 * and thereby allow us to exit fail-open.
4199 * See the top-level comment in fail-open.c for more information.
4201 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4202 DPIF_RECV_MSG_PADDING));
4205 ofpbuf_pull(packet, sizeof *msg);
4206 rule_execute(p, rule, packet, &flow);
4207 rule_reinstall(p, rule);
4211 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4213 struct odp_msg *msg = packet->data;
4215 switch (msg->type) {
4216 case _ODPL_ACTION_NR:
4217 COVERAGE_INC(ofproto_ctlr_action);
4218 send_packet_in(p, packet);
4221 case _ODPL_SFLOW_NR:
4223 ofproto_sflow_received(p->sflow, msg);
4225 ofpbuf_delete(packet);
4229 handle_odp_miss_msg(p, packet);
4233 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4239 /* Flow expiration. */
4241 struct expire_cbdata {
4242 struct ofproto *ofproto;
4246 static int ofproto_dp_max_idle(const struct ofproto *);
4247 static void ofproto_update_used(struct ofproto *);
4248 static void rule_expire(struct cls_rule *, void *cbdata);
4250 /* This function is called periodically by ofproto_run(). Its job is to
4251 * collect updates for the flows that have been installed into the datapath,
4252 * most importantly when they last were used, and then use that information to
4253 * expire flows that have not been used recently.
4255 * Returns the number of milliseconds after which it should be called again. */
4257 ofproto_expire(struct ofproto *ofproto)
4259 struct expire_cbdata cbdata;
4261 /* Update 'used' for each flow in the datapath. */
4262 ofproto_update_used(ofproto);
4264 /* Expire idle flows.
4266 * A wildcarded flow is idle only when all of its subrules have expired due
4267 * to becoming idle, so iterate through the exact-match flows first. */
4268 cbdata.ofproto = ofproto;
4269 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4270 classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
4271 classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
4273 /* Let the hook know that we're at a stable point: all outstanding data
4274 * in existing flows has been accounted to the account_cb. Thus, the
4275 * hook can now reasonably do operations that depend on having accurate
4276 * flow volume accounting (currently, that's just bond rebalancing). */
4277 if (ofproto->ofhooks->account_checkpoint_cb) {
4278 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4281 return MIN(cbdata.dp_max_idle, 1000);
4284 /* Update 'used' member of each flow currently installed into the datapath. */
4286 ofproto_update_used(struct ofproto *p)
4288 struct odp_flow *flows;
4293 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4298 for (i = 0; i < n_flows; i++) {
4299 struct odp_flow *f = &flows[i];
4300 struct cls_rule target;
4304 odp_flow_key_to_flow(&f->key, &flow);
4305 cls_rule_from_flow(&flow, 0, UINT16_MAX, &target);
4307 rule = rule_from_cls_rule(classifier_find_rule_exactly(&p->cls,
4310 if (rule && rule->installed) {
4311 update_time(p, rule, &f->stats);
4312 rule_account(p, rule, f->stats.n_bytes);
4314 /* There's a flow in the datapath that we know nothing about.
4316 COVERAGE_INC(ofproto_unexpected_rule);
4317 dpif_flow_del(p->dpif, f);
4324 /* Calculates and returns the number of milliseconds of idle time after which
4325 * flows should expire from the datapath and we should fold their statistics
4326 * into their parent rules in userspace. */
4328 ofproto_dp_max_idle(const struct ofproto *ofproto)
4331 * Idle time histogram.
4333 * Most of the time a switch has a relatively small number of flows. When
4334 * this is the case we might as well keep statistics for all of them in
4335 * userspace and to cache them in the kernel datapath for performance as
4338 * As the number of flows increases, the memory required to maintain
4339 * statistics about them in userspace and in the kernel becomes
4340 * significant. However, with a large number of flows it is likely that
4341 * only a few of them are "heavy hitters" that consume a large amount of
4342 * bandwidth. At this point, only heavy hitters are worth caching in the
4343 * kernel and maintaining in userspaces; other flows we can discard.
4345 * The technique used to compute the idle time is to build a histogram with
4346 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4347 * is installed in the kernel gets dropped in the appropriate bucket.
4348 * After the histogram has been built, we compute the cutoff so that only
4349 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4350 * cached. At least the most-recently-used bucket of flows is kept, so
4351 * actually an arbitrary number of flows can be kept in any given
4352 * expiration run (though the next run will delete most of those unless
4353 * they receive additional data).
4355 * This requires a second pass through the exact-match flows, in addition
4356 * to the pass made by ofproto_update_used(), because the former function
4357 * never looks at uninstallable flows.
4359 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4360 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4361 int buckets[N_BUCKETS] = { 0 };
4367 total = classifier_count_exact(&ofproto->cls);
4368 if (total <= 1000) {
4369 return N_BUCKETS * BUCKET_WIDTH;
4372 /* Build histogram. */
4374 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) {
4375 long long int idle = now - rule->used;
4376 int bucket = (idle <= 0 ? 0
4377 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4378 : (unsigned int) idle / BUCKET_WIDTH);
4382 /* Find the first bucket whose flows should be expired. */
4383 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4384 if (buckets[bucket]) {
4387 subtotal += buckets[bucket++];
4388 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4393 if (VLOG_IS_DBG_ENABLED()) {
4397 ds_put_cstr(&s, "keep");
4398 for (i = 0; i < N_BUCKETS; i++) {
4400 ds_put_cstr(&s, ", drop");
4403 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4406 VLOG_INFO("%s: %s (msec:count)",
4407 dpif_name(ofproto->dpif), ds_cstr(&s));
4411 return bucket * BUCKET_WIDTH;
4415 rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
4417 if (ofproto->netflow && !is_controller_rule(rule) &&
4418 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4419 struct ofexpired expired;
4420 struct odp_flow odp_flow;
4422 /* Get updated flow stats.
4424 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4425 * updated TCP flags and (2) the dpif_flow_list_all() in
4426 * ofproto_update_used() zeroed TCP flags. */
4427 memset(&odp_flow, 0, sizeof odp_flow);
4428 if (rule->installed) {
4429 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
4430 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4431 dpif_flow_get(ofproto->dpif, &odp_flow);
4433 if (odp_flow.stats.n_packets) {
4434 update_time(ofproto, rule, &odp_flow.stats);
4435 netflow_flow_update_flags(&rule->nf_flow,
4436 odp_flow.stats.tcp_flags);
4440 expired.flow = rule->cr.flow;
4441 expired.packet_count = rule->packet_count +
4442 odp_flow.stats.n_packets;
4443 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4444 expired.used = rule->used;
4446 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4450 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4451 * rules, then delete it entirely.
4453 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4454 * the datapath and fold its statistics back into its super-rule.
4456 * (This is a callback function for classifier_for_each().) */
4458 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4460 struct expire_cbdata *cbdata = cbdata_;
4461 struct ofproto *ofproto = cbdata->ofproto;
4462 struct rule *rule = rule_from_cls_rule(cls_rule);
4463 long long int hard_expire, idle_expire, expire, now;
4465 /* Calculate OpenFlow expiration times for 'rule'. */
4466 hard_expire = (rule->hard_timeout
4467 ? rule->created + rule->hard_timeout * 1000
4469 idle_expire = (rule->idle_timeout
4470 && (rule->super || list_is_empty(&rule->list))
4471 ? rule->used + rule->idle_timeout * 1000
4473 expire = MIN(hard_expire, idle_expire);
4477 /* 'rule' has not expired according to OpenFlow rules. */
4478 if (!rule->cr.wc.wildcards) {
4479 if (now >= rule->used + cbdata->dp_max_idle) {
4480 /* This rule is idle, so drop it to free up resources. */
4482 /* It's not part of the OpenFlow flow table, so we can
4483 * delete it entirely and fold its statistics into its
4485 rule_remove(ofproto, rule);
4487 /* It is part of the OpenFlow flow table, so we have to
4488 * keep the rule but we can at least uninstall it from the
4490 rule_uninstall(ofproto, rule);
4493 /* Send NetFlow active timeout if appropriate. */
4494 rule_active_timeout(cbdata->ofproto, rule);
4498 /* 'rule' has expired according to OpenFlow rules. */
4499 COVERAGE_INC(ofproto_expired);
4501 /* Update stats. (This is a no-op if the rule expired due to an idle
4502 * timeout, because that only happens when the rule has no subrules
4504 if (rule->cr.wc.wildcards) {
4505 struct rule *subrule, *next;
4506 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
4507 rule_remove(cbdata->ofproto, subrule);
4510 rule_uninstall(cbdata->ofproto, rule);
4513 /* Get rid of the rule. */
4514 if (!rule_is_hidden(rule)) {
4515 send_flow_removed(cbdata->ofproto, rule, now,
4517 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4519 rule_remove(cbdata->ofproto, rule);
4524 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4526 struct rule *sub = rule_from_cls_rule(sub_);
4527 struct revalidate_cbdata *cbdata = cbdata_;
4529 if (cbdata->revalidate_all
4530 || (cbdata->revalidate_subrules && sub->super)
4531 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4532 revalidate_rule(cbdata->ofproto, sub);
4537 revalidate_rule(struct ofproto *p, struct rule *rule)
4539 const struct flow *flow = &rule->cr.flow;
4541 COVERAGE_INC(ofproto_revalidate_rule);
4544 super = rule_from_cls_rule(classifier_lookup(&p->cls, flow,
4547 rule_remove(p, rule);
4549 } else if (super != rule->super) {
4550 COVERAGE_INC(ofproto_revalidate_moved);
4551 list_remove(&rule->list);
4552 list_push_back(&super->list, &rule->list);
4553 rule->super = super;
4554 rule->hard_timeout = super->hard_timeout;
4555 rule->idle_timeout = super->idle_timeout;
4556 rule->created = super->created;
4561 rule_update_actions(p, rule);
4565 static struct ofpbuf *
4566 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4567 long long int now, uint8_t reason)
4569 struct ofp_flow_removed *ofr;
4571 long long int tdiff = now - rule->created;
4572 uint32_t sec = tdiff / 1000;
4573 uint32_t msec = tdiff - (sec * 1000);
4575 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4576 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4578 ofr->cookie = rule->flow_cookie;
4579 ofr->priority = htons(rule->cr.priority);
4580 ofr->reason = reason;
4581 ofr->duration_sec = htonl(sec);
4582 ofr->duration_nsec = htonl(msec * 1000000);
4583 ofr->idle_timeout = htons(rule->idle_timeout);
4584 ofr->packet_count = htonll(rule->packet_count);
4585 ofr->byte_count = htonll(rule->byte_count);
4591 send_flow_removed(struct ofproto *p, struct rule *rule,
4592 long long int now, uint8_t reason)
4594 struct ofconn *ofconn;
4595 struct ofconn *prev;
4596 struct ofpbuf *buf = NULL;
4598 if (!rule->send_flow_removed) {
4602 /* We limit the maximum number of queued flow expirations it by accounting
4603 * them under the counter for replies. That works because preventing
4604 * OpenFlow requests from being processed also prevents new flows from
4605 * being added (and expiring). (It also prevents processing OpenFlow
4606 * requests that would not add new flows, so it is imperfect.) */
4609 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4610 if (rconn_is_connected(ofconn->rconn)
4611 && ofconn_receives_async_msgs(ofconn)) {
4613 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4615 buf = compose_flow_removed(p, rule, now, reason);
4621 queue_tx(buf, prev, prev->reply_counter);
4625 /* pinsched callback for sending 'packet' on 'ofconn'. */
4627 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4629 struct ofconn *ofconn = ofconn_;
4631 rconn_send_with_limit(ofconn->rconn, packet,
4632 ofconn->packet_in_counter, 100);
4635 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4636 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4637 * packet scheduler for sending.
4639 * 'max_len' specifies the maximum number of bytes of the packet to send on
4640 * 'ofconn' (INT_MAX specifies no limit).
4642 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4643 * ownership is transferred to this function. */
4645 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4648 struct ofproto *ofproto = ofconn->ofproto;
4649 struct ofp_packet_in *opi = packet->data;
4650 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4651 int send_len, trim_size;
4655 if (opi->reason == OFPR_ACTION) {
4656 buffer_id = UINT32_MAX;
4657 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4658 buffer_id = pktbuf_get_null();
4659 } else if (!ofconn->pktbuf) {
4660 buffer_id = UINT32_MAX;
4662 struct ofpbuf payload;
4663 payload.data = opi->data;
4664 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4665 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4668 /* Figure out how much of the packet to send. */
4669 send_len = ntohs(opi->total_len);
4670 if (buffer_id != UINT32_MAX) {
4671 send_len = MIN(send_len, ofconn->miss_send_len);
4673 send_len = MIN(send_len, max_len);
4675 /* Adjust packet length and clone if necessary. */
4676 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4678 packet = ofpbuf_clone_data(packet->data, trim_size);
4681 packet->size = trim_size;
4684 /* Update packet headers. */
4685 opi->buffer_id = htonl(buffer_id);
4686 update_openflow_length(packet);
4688 /* Hand over to packet scheduler. It might immediately call into
4689 * do_send_packet_in() or it might buffer it for a while (until a later
4690 * call to pinsched_run()). */
4691 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4692 packet, do_send_packet_in, ofconn);
4695 /* Replace struct odp_msg header in 'packet' by equivalent struct
4696 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4697 * returned by dpif_recv()).
4699 * The conversion is not complete: the caller still needs to trim any unneeded
4700 * payload off the end of the buffer, set the length in the OpenFlow header,
4701 * and set buffer_id. Those require us to know the controller settings and so
4702 * must be done on a per-controller basis.
4704 * Returns the maximum number of bytes of the packet that should be sent to
4705 * the controller (INT_MAX if no limit). */
4707 do_convert_to_packet_in(struct ofpbuf *packet)
4709 struct odp_msg *msg = packet->data;
4710 struct ofp_packet_in *opi;
4716 /* Extract relevant header fields */
4717 if (msg->type == _ODPL_ACTION_NR) {
4718 reason = OFPR_ACTION;
4721 reason = OFPR_NO_MATCH;
4724 total_len = msg->length - sizeof *msg;
4725 in_port = odp_port_to_ofp_port(msg->port);
4727 /* Repurpose packet buffer by overwriting header. */
4728 ofpbuf_pull(packet, sizeof(struct odp_msg));
4729 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4730 opi->header.version = OFP_VERSION;
4731 opi->header.type = OFPT_PACKET_IN;
4732 opi->total_len = htons(total_len);
4733 opi->in_port = htons(in_port);
4734 opi->reason = reason;
4739 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4740 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4741 * as necessary according to their individual configurations.
4743 * 'packet' must have sufficient headroom to convert it into a struct
4744 * ofp_packet_in (e.g. as returned by dpif_recv()).
4746 * Takes ownership of 'packet'. */
4748 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4750 struct ofconn *ofconn, *prev;
4753 max_len = do_convert_to_packet_in(packet);
4756 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4757 if (ofconn_receives_async_msgs(ofconn)) {
4759 schedule_packet_in(prev, packet, max_len, true);
4765 schedule_packet_in(prev, packet, max_len, false);
4767 ofpbuf_delete(packet);
4772 pick_datapath_id(const struct ofproto *ofproto)
4774 const struct ofport *port;
4776 port = get_port(ofproto, ODPP_LOCAL);
4778 uint8_t ea[ETH_ADDR_LEN];
4781 error = netdev_get_etheraddr(port->netdev, ea);
4783 return eth_addr_to_uint64(ea);
4785 VLOG_WARN("could not get MAC address for %s (%s)",
4786 netdev_get_name(port->netdev), strerror(error));
4788 return ofproto->fallback_dpid;
4792 pick_fallback_dpid(void)
4794 uint8_t ea[ETH_ADDR_LEN];
4795 eth_addr_nicira_random(ea);
4796 return eth_addr_to_uint64(ea);
4800 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4801 struct odp_actions *actions, tag_type *tags,
4802 uint16_t *nf_output_iface, void *ofproto_)
4804 struct ofproto *ofproto = ofproto_;
4807 /* Drop frames for reserved multicast addresses. */
4808 if (eth_addr_is_reserved(flow->dl_dst)) {
4812 /* Learn source MAC (but don't try to learn from revalidation). */
4813 if (packet != NULL) {
4814 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4816 GRAT_ARP_LOCK_NONE);
4818 /* The log messages here could actually be useful in debugging,
4819 * so keep the rate limit relatively high. */
4820 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4821 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4822 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4823 ofproto_revalidate(ofproto, rev_tag);
4827 /* Determine output port. */
4828 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4831 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4832 nf_output_iface, actions);
4833 } else if (out_port != flow->in_port) {
4834 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4835 *nf_output_iface = out_port;
4843 static const struct ofhooks default_ofhooks = {
4844 default_normal_ofhook_cb,