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"
68 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
69 struct netdev *netdev;
70 struct ofp_phy_port opp; /* In host byte order. */
74 static void ofport_free(struct ofport *);
75 static void hton_ofp_phy_port(struct ofp_phy_port *);
77 static int xlate_actions(const union ofp_action *in, size_t n_in,
78 const struct flow *, struct ofproto *,
79 const struct ofpbuf *packet,
80 struct odp_actions *out, tag_type *tags,
81 bool *may_set_up_flow, uint16_t *nf_output_iface);
86 ovs_be64 flow_cookie; /* Controller-issued identifier. */
87 uint16_t idle_timeout; /* In seconds from time of last use. */
88 uint16_t hard_timeout; /* In seconds from time of creation. */
89 bool send_flow_removed; /* Send a flow removed message? */
90 long long int used; /* Last-used time (0 if never used). */
91 long long int created; /* Creation time. */
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
94 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
95 tag_type tags; /* Tags (set only by hooks). */
96 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
98 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
99 * exact-match rule (having cr.wc.wildcards of 0) generated from the
100 * wildcard rule 'super'. In this case, 'list' is an element of the
103 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
104 * a list of subrules. A super-rule with no wildcards (where
105 * cr.wc.wildcards is 0) will never have any subrules. */
111 * 'n_actions' is the number of elements in the 'actions' array. A single
112 * action may take up more more than one element's worth of space.
114 * A subrule has no actions (it uses the super-rule's actions). */
116 union ofp_action *actions;
120 * A super-rule with wildcard fields never has ODP actions (since the
121 * datapath only supports exact-match flows). */
122 bool installed; /* Installed in datapath? */
123 bool may_install; /* True ordinarily; false if actions must
124 * be reassessed for every packet. */
126 union odp_action *odp_actions;
130 rule_is_hidden(const struct rule *rule)
132 /* Subrules are merely an implementation detail, so hide them from the
134 if (rule->super != NULL) {
138 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
139 * (e.g. by in-band control) and are intentionally hidden from the
141 if (rule->cr.priority > UINT16_MAX) {
148 static struct rule *rule_create(struct ofproto *, struct rule *super,
149 const union ofp_action *, size_t n_actions,
150 uint16_t idle_timeout, uint16_t hard_timeout,
151 ovs_be64 flow_cookie, bool send_flow_removed);
152 static void rule_free(struct rule *);
153 static void rule_destroy(struct ofproto *, struct rule *);
154 static struct rule *rule_from_cls_rule(const struct cls_rule *);
155 static void rule_insert(struct ofproto *, struct rule *,
156 struct ofpbuf *packet, uint16_t in_port);
157 static void rule_remove(struct ofproto *, struct rule *);
158 static bool rule_make_actions(struct ofproto *, struct rule *,
159 const struct ofpbuf *packet);
160 static void rule_install(struct ofproto *, struct rule *,
161 struct rule *displaced_rule);
162 static void rule_uninstall(struct ofproto *, struct rule *);
163 static void rule_post_uninstall(struct ofproto *, struct rule *);
164 static void send_flow_removed(struct ofproto *p, struct rule *rule,
165 long long int now, uint8_t reason);
167 /* ofproto supports two kinds of OpenFlow connections:
169 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
170 * maintains persistent connections to these controllers and by default
171 * sends them asynchronous messages such as packet-ins.
173 * - "Service" connections, e.g. from ovs-ofctl. When these connections
174 * drop, it is the other side's responsibility to reconnect them if
175 * necessary. ofproto does not send them asynchronous messages by default.
177 * Currently, active (tcp, ssl, unix) connections are always "primary"
178 * connections and passive (ptcp, pssl, punix) connections are always "service"
179 * connections. There is no inherent reason for this, but it reflects the
183 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
184 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
187 /* A listener for incoming OpenFlow "service" connections. */
189 struct hmap_node node; /* In struct ofproto's "services" hmap. */
190 struct pvconn *pvconn; /* OpenFlow connection listener. */
192 /* These are not used by ofservice directly. They are settings for
193 * accepted "struct ofconn"s from the pvconn. */
194 int probe_interval; /* Max idle time before probing, in seconds. */
195 int rate_limit; /* Max packet-in rate in packets per second. */
196 int burst_limit; /* Limit on accumulating packet credits. */
199 static struct ofservice *ofservice_lookup(struct ofproto *,
201 static int ofservice_create(struct ofproto *,
202 const struct ofproto_controller *);
203 static void ofservice_reconfigure(struct ofservice *,
204 const struct ofproto_controller *);
205 static void ofservice_destroy(struct ofproto *, struct ofservice *);
207 /* An OpenFlow connection. */
209 struct ofproto *ofproto; /* The ofproto that owns this connection. */
210 struct list node; /* In struct ofproto's "all_conns" list. */
211 struct rconn *rconn; /* OpenFlow connection. */
212 enum ofconn_type type; /* Type. */
214 /* OFPT_PACKET_IN related data. */
215 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
216 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
217 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
218 int miss_send_len; /* Bytes to send of buffered packets. */
220 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
221 * requests, and the maximum number before we stop reading OpenFlow
223 #define OFCONN_REPLY_MAX 100
224 struct rconn_packet_counter *reply_counter;
226 /* type == OFCONN_PRIMARY only. */
227 enum nx_role role; /* Role. */
228 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
229 struct discovery *discovery; /* Controller discovery object, if enabled. */
230 struct status_category *ss; /* Switch status category. */
231 enum ofproto_band band; /* In-band or out-of-band? */
234 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
235 * "schedulers" array. Their values are 0 and 1, and their meanings and values
236 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
237 * case anything ever changes, check their values here. */
238 #define N_SCHEDULERS 2
239 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
240 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
241 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
242 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
244 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
246 static void ofconn_destroy(struct ofconn *);
247 static void ofconn_run(struct ofconn *, struct ofproto *);
248 static void ofconn_wait(struct ofconn *);
249 static bool ofconn_receives_async_msgs(const struct ofconn *);
250 static char *ofconn_make_name(const struct ofproto *, const char *target);
251 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
253 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
254 struct rconn_packet_counter *counter);
256 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
257 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
261 uint64_t datapath_id; /* Datapath ID. */
262 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
263 char *mfr_desc; /* Manufacturer. */
264 char *hw_desc; /* Hardware. */
265 char *sw_desc; /* Software version. */
266 char *serial_desc; /* Serial number. */
267 char *dp_desc; /* Datapath description. */
271 struct netdev_monitor *netdev_monitor;
272 struct hmap ports; /* Contains "struct ofport"s. */
273 struct shash port_by_name;
277 struct switch_status *switch_status;
278 struct fail_open *fail_open;
279 struct netflow *netflow;
280 struct ofproto_sflow *sflow;
282 /* In-band control. */
283 struct in_band *in_band;
284 long long int next_in_band_update;
285 struct sockaddr_in *extra_in_band_remotes;
286 size_t n_extra_remotes;
289 struct classifier cls;
290 bool need_revalidate;
291 long long int next_expiration;
292 struct tag_set revalidate_set;
293 bool tun_id_from_cookie;
295 /* OpenFlow connections. */
296 struct hmap controllers; /* Controller "struct ofconn"s. */
297 struct list all_conns; /* Contains "struct ofconn"s. */
298 enum ofproto_fail_mode fail_mode;
300 /* OpenFlow listeners. */
301 struct hmap services; /* Contains "struct ofservice"s. */
302 struct pvconn **snoops;
305 /* Hooks for ovs-vswitchd. */
306 const struct ofhooks *ofhooks;
309 /* Used by default ofhooks. */
310 struct mac_learning *ml;
313 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
315 static const struct ofhooks default_ofhooks;
317 static uint64_t pick_datapath_id(const struct ofproto *);
318 static uint64_t pick_fallback_dpid(void);
320 static int ofproto_expire(struct ofproto *);
322 static void update_stats(struct ofproto *, struct rule *,
323 const struct odp_flow_stats *);
324 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
325 static void revalidate_cb(struct cls_rule *rule_, void *p_);
327 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
329 static void handle_openflow(struct ofconn *, struct ofproto *,
332 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
333 static void update_port(struct ofproto *, const char *devname);
334 static int init_ports(struct ofproto *);
335 static void reinit_ports(struct ofproto *);
338 ofproto_create(const char *datapath, const char *datapath_type,
339 const struct ofhooks *ofhooks, void *aux,
340 struct ofproto **ofprotop)
342 struct odp_stats stats;
349 /* Connect to datapath and start listening for messages. */
350 error = dpif_open(datapath, datapath_type, &dpif);
352 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
355 error = dpif_get_dp_stats(dpif, &stats);
357 VLOG_ERR("failed to obtain stats for datapath %s: %s",
358 datapath, strerror(error));
362 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
364 VLOG_ERR("failed to listen on datapath %s: %s",
365 datapath, strerror(error));
369 dpif_flow_flush(dpif);
370 dpif_recv_purge(dpif);
372 /* Initialize settings. */
373 p = xzalloc(sizeof *p);
374 p->fallback_dpid = pick_fallback_dpid();
375 p->datapath_id = p->fallback_dpid;
376 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
377 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
378 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
379 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
380 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
382 /* Initialize datapath. */
384 p->netdev_monitor = netdev_monitor_create();
385 hmap_init(&p->ports);
386 shash_init(&p->port_by_name);
387 p->max_ports = stats.max_ports;
389 /* Initialize submodules. */
390 p->switch_status = switch_status_create(p);
396 /* Initialize flow table. */
397 classifier_init(&p->cls);
398 p->need_revalidate = false;
399 p->next_expiration = time_msec() + 1000;
400 tag_set_init(&p->revalidate_set);
402 /* Initialize OpenFlow connections. */
403 list_init(&p->all_conns);
404 hmap_init(&p->controllers);
405 hmap_init(&p->services);
409 /* Initialize hooks. */
411 p->ofhooks = ofhooks;
415 p->ofhooks = &default_ofhooks;
417 p->ml = mac_learning_create();
420 /* Pick final datapath ID. */
421 p->datapath_id = pick_datapath_id(p);
422 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
429 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
431 uint64_t old_dpid = p->datapath_id;
432 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
433 if (p->datapath_id != old_dpid) {
434 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
436 /* Force all active connections to reconnect, since there is no way to
437 * notify a controller that the datapath ID has changed. */
438 ofproto_reconnect_controllers(p);
443 is_discovery_controller(const struct ofproto_controller *c)
445 return !strcmp(c->target, "discover");
449 is_in_band_controller(const struct ofproto_controller *c)
451 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
454 /* Creates a new controller in 'ofproto'. Some of the settings are initially
455 * drawn from 'c', but update_controller() needs to be called later to finish
456 * the new ofconn's configuration. */
458 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
460 struct discovery *discovery;
461 struct ofconn *ofconn;
463 if (is_discovery_controller(c)) {
464 int error = discovery_create(c->accept_re, c->update_resolv_conf,
465 ofproto->dpif, ofproto->switch_status,
474 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
475 ofconn->pktbuf = pktbuf_create();
476 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
478 ofconn->discovery = discovery;
480 char *name = ofconn_make_name(ofproto, c->target);
481 rconn_connect(ofconn->rconn, c->target, name);
484 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
485 hash_string(c->target, 0));
488 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
489 * target or turn discovery on or off (these are done by creating new ofconns
490 * and deleting old ones), but it can update the rest of an ofconn's
493 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
497 ofconn->band = (is_in_band_controller(c)
498 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
500 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
502 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
503 rconn_set_probe_interval(ofconn->rconn, probe_interval);
505 if (ofconn->discovery) {
506 discovery_set_update_resolv_conf(ofconn->discovery,
507 c->update_resolv_conf);
508 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
511 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
515 ofconn_get_target(const struct ofconn *ofconn)
517 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
520 static struct ofconn *
521 find_controller_by_target(struct ofproto *ofproto, const char *target)
523 struct ofconn *ofconn;
525 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
526 hash_string(target, 0), &ofproto->controllers) {
527 if (!strcmp(ofconn_get_target(ofconn), target)) {
535 update_in_band_remotes(struct ofproto *ofproto)
537 const struct ofconn *ofconn;
538 struct sockaddr_in *addrs;
539 size_t max_addrs, n_addrs;
543 /* Allocate enough memory for as many remotes as we could possibly have. */
544 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
545 addrs = xmalloc(max_addrs * sizeof *addrs);
548 /* Add all the remotes. */
550 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
551 struct sockaddr_in *sin = &addrs[n_addrs];
553 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
557 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
558 if (sin->sin_addr.s_addr) {
559 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
562 if (ofconn->discovery) {
566 for (i = 0; i < ofproto->n_extra_remotes; i++) {
567 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
570 /* Create or update or destroy in-band.
572 * Ordinarily we only enable in-band if there's at least one remote
573 * address, but discovery needs the in-band rules for DHCP to be installed
574 * even before we know any remote addresses. */
575 if (n_addrs || discovery) {
576 if (!ofproto->in_band) {
577 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
580 if (ofproto->in_band) {
581 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
583 ofproto->next_in_band_update = time_msec() + 1000;
585 in_band_destroy(ofproto->in_band);
586 ofproto->in_band = NULL;
594 update_fail_open(struct ofproto *p)
596 struct ofconn *ofconn;
598 if (!hmap_is_empty(&p->controllers)
599 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
600 struct rconn **rconns;
604 p->fail_open = fail_open_create(p, p->switch_status);
608 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
609 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
610 rconns[n++] = ofconn->rconn;
613 fail_open_set_controllers(p->fail_open, rconns, n);
614 /* p->fail_open takes ownership of 'rconns'. */
616 fail_open_destroy(p->fail_open);
622 ofproto_set_controllers(struct ofproto *p,
623 const struct ofproto_controller *controllers,
624 size_t n_controllers)
626 struct shash new_controllers;
627 struct ofconn *ofconn, *next_ofconn;
628 struct ofservice *ofservice, *next_ofservice;
632 /* Create newly configured controllers and services.
633 * Create a name to ofproto_controller mapping in 'new_controllers'. */
634 shash_init(&new_controllers);
635 for (i = 0; i < n_controllers; i++) {
636 const struct ofproto_controller *c = &controllers[i];
638 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
639 if (!find_controller_by_target(p, c->target)) {
640 add_controller(p, c);
642 } else if (!pvconn_verify_name(c->target)) {
643 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
647 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
648 dpif_name(p->dpif), c->target);
652 shash_add_once(&new_controllers, c->target, &controllers[i]);
655 /* Delete controllers that are no longer configured.
656 * Update configuration of all now-existing controllers. */
658 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
659 struct ofproto_controller *c;
661 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
663 ofconn_destroy(ofconn);
665 update_controller(ofconn, c);
672 /* Delete services that are no longer configured.
673 * Update configuration of all now-existing services. */
674 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
675 struct ofproto_controller *c;
677 c = shash_find_data(&new_controllers,
678 pvconn_get_name(ofservice->pvconn));
680 ofservice_destroy(p, ofservice);
682 ofservice_reconfigure(ofservice, c);
686 shash_destroy(&new_controllers);
688 update_in_band_remotes(p);
691 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
692 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
693 struct ofconn, hmap_node);
694 ofconn->ss = switch_status_register(p->switch_status, "remote",
695 rconn_status_cb, ofconn->rconn);
700 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
702 p->fail_mode = fail_mode;
706 /* Drops the connections between 'ofproto' and all of its controllers, forcing
707 * them to reconnect. */
709 ofproto_reconnect_controllers(struct ofproto *ofproto)
711 struct ofconn *ofconn;
713 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
714 rconn_reconnect(ofconn->rconn);
719 any_extras_changed(const struct ofproto *ofproto,
720 const struct sockaddr_in *extras, size_t n)
724 if (n != ofproto->n_extra_remotes) {
728 for (i = 0; i < n; i++) {
729 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
730 const struct sockaddr_in *new = &extras[i];
732 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
733 old->sin_port != new->sin_port) {
741 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
742 * in-band control should guarantee access, in the same way that in-band
743 * control guarantees access to OpenFlow controllers. */
745 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
746 const struct sockaddr_in *extras, size_t n)
748 if (!any_extras_changed(ofproto, extras, n)) {
752 free(ofproto->extra_in_band_remotes);
753 ofproto->n_extra_remotes = n;
754 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
756 update_in_band_remotes(ofproto);
760 ofproto_set_desc(struct ofproto *p,
761 const char *mfr_desc, const char *hw_desc,
762 const char *sw_desc, const char *serial_desc,
765 struct ofp_desc_stats *ods;
768 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
769 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
770 sizeof ods->mfr_desc);
773 p->mfr_desc = xstrdup(mfr_desc);
776 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
777 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
778 sizeof ods->hw_desc);
781 p->hw_desc = xstrdup(hw_desc);
784 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
785 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
786 sizeof ods->sw_desc);
789 p->sw_desc = xstrdup(sw_desc);
792 if (strlen(serial_desc) >= sizeof ods->serial_num) {
793 VLOG_WARN("truncating serial_desc, must be less than %zu "
795 sizeof ods->serial_num);
797 free(p->serial_desc);
798 p->serial_desc = xstrdup(serial_desc);
801 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
802 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
803 sizeof ods->dp_desc);
806 p->dp_desc = xstrdup(dp_desc);
811 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
812 const struct svec *svec)
814 struct pvconn **pvconns = *pvconnsp;
815 size_t n_pvconns = *n_pvconnsp;
819 for (i = 0; i < n_pvconns; i++) {
820 pvconn_close(pvconns[i]);
824 pvconns = xmalloc(svec->n * sizeof *pvconns);
826 for (i = 0; i < svec->n; i++) {
827 const char *name = svec->names[i];
828 struct pvconn *pvconn;
831 error = pvconn_open(name, &pvconn);
833 pvconns[n_pvconns++] = pvconn;
835 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
843 *n_pvconnsp = n_pvconns;
849 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
851 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
855 ofproto_set_netflow(struct ofproto *ofproto,
856 const struct netflow_options *nf_options)
858 if (nf_options && nf_options->collectors.n) {
859 if (!ofproto->netflow) {
860 ofproto->netflow = netflow_create();
862 return netflow_set_options(ofproto->netflow, nf_options);
864 netflow_destroy(ofproto->netflow);
865 ofproto->netflow = NULL;
871 ofproto_set_sflow(struct ofproto *ofproto,
872 const struct ofproto_sflow_options *oso)
874 struct ofproto_sflow *os = ofproto->sflow;
877 struct ofport *ofport;
879 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
880 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
881 ofproto_sflow_add_port(os, ofport->odp_port,
882 netdev_get_name(ofport->netdev));
885 ofproto_sflow_set_options(os, oso);
887 ofproto_sflow_destroy(os);
888 ofproto->sflow = NULL;
893 ofproto_get_datapath_id(const struct ofproto *ofproto)
895 return ofproto->datapath_id;
899 ofproto_has_primary_controller(const struct ofproto *ofproto)
901 return !hmap_is_empty(&ofproto->controllers);
904 enum ofproto_fail_mode
905 ofproto_get_fail_mode(const struct ofproto *p)
911 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
915 for (i = 0; i < ofproto->n_snoops; i++) {
916 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
921 ofproto_destroy(struct ofproto *p)
923 struct ofservice *ofservice, *next_ofservice;
924 struct ofconn *ofconn, *next_ofconn;
925 struct ofport *ofport, *next_ofport;
932 /* Destroy fail-open and in-band early, since they touch the classifier. */
933 fail_open_destroy(p->fail_open);
936 in_band_destroy(p->in_band);
938 free(p->extra_in_band_remotes);
940 ofproto_flush_flows(p);
941 classifier_destroy(&p->cls);
943 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
944 ofconn_destroy(ofconn);
946 hmap_destroy(&p->controllers);
949 netdev_monitor_destroy(p->netdev_monitor);
950 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
951 hmap_remove(&p->ports, &ofport->hmap_node);
954 shash_destroy(&p->port_by_name);
956 switch_status_destroy(p->switch_status);
957 netflow_destroy(p->netflow);
958 ofproto_sflow_destroy(p->sflow);
960 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
961 ofservice_destroy(p, ofservice);
963 hmap_destroy(&p->services);
965 for (i = 0; i < p->n_snoops; i++) {
966 pvconn_close(p->snoops[i]);
970 mac_learning_destroy(p->ml);
975 free(p->serial_desc);
978 hmap_destroy(&p->ports);
984 ofproto_run(struct ofproto *p)
986 int error = ofproto_run1(p);
988 error = ofproto_run2(p, false);
994 process_port_change(struct ofproto *ofproto, int error, char *devname)
996 if (error == ENOBUFS) {
997 reinit_ports(ofproto);
999 update_port(ofproto, devname);
1004 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1005 * means that 'ofconn' is more interesting for monitoring than a lower return
1008 snoop_preference(const struct ofconn *ofconn)
1010 switch (ofconn->role) {
1011 case NX_ROLE_MASTER:
1018 /* Shouldn't happen. */
1023 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1024 * Connects this vconn to a controller. */
1026 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1028 struct ofconn *ofconn, *best;
1030 /* Pick a controller for monitoring. */
1032 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1033 if (ofconn->type == OFCONN_PRIMARY
1034 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1040 rconn_add_monitor(best->rconn, vconn);
1042 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1048 ofproto_run1(struct ofproto *p)
1050 struct ofconn *ofconn, *next_ofconn;
1051 struct ofservice *ofservice;
1056 if (shash_is_empty(&p->port_by_name)) {
1060 for (i = 0; i < 50; i++) {
1063 error = dpif_recv(p->dpif, &buf);
1065 if (error == ENODEV) {
1066 /* Someone destroyed the datapath behind our back. The caller
1067 * better destroy us and give up, because we're just going to
1068 * spin from here on out. */
1069 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1070 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1071 dpif_name(p->dpif));
1077 handle_odp_msg(p, buf);
1080 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1081 process_port_change(p, error, devname);
1083 while ((error = netdev_monitor_poll(p->netdev_monitor,
1084 &devname)) != EAGAIN) {
1085 process_port_change(p, error, devname);
1089 if (time_msec() >= p->next_in_band_update) {
1090 update_in_band_remotes(p);
1092 in_band_run(p->in_band);
1095 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1096 ofconn_run(ofconn, p);
1099 /* Fail-open maintenance. Do this after processing the ofconns since
1100 * fail-open checks the status of the controller rconn. */
1102 fail_open_run(p->fail_open);
1105 HMAP_FOR_EACH (ofservice, node, &p->services) {
1106 struct vconn *vconn;
1109 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1111 struct rconn *rconn;
1114 rconn = rconn_create(ofservice->probe_interval, 0);
1115 name = ofconn_make_name(p, vconn_get_name(vconn));
1116 rconn_connect_unreliably(rconn, vconn, name);
1119 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1120 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1121 ofservice->burst_limit);
1122 } else if (retval != EAGAIN) {
1123 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1127 for (i = 0; i < p->n_snoops; i++) {
1128 struct vconn *vconn;
1131 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1133 add_snooper(p, vconn);
1134 } else if (retval != EAGAIN) {
1135 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1139 if (time_msec() >= p->next_expiration) {
1140 int delay = ofproto_expire(p);
1141 p->next_expiration = time_msec() + delay;
1142 COVERAGE_INC(ofproto_expiration);
1146 netflow_run(p->netflow);
1149 ofproto_sflow_run(p->sflow);
1155 struct revalidate_cbdata {
1156 struct ofproto *ofproto;
1157 bool revalidate_all; /* Revalidate all exact-match rules? */
1158 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1159 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1163 ofproto_run2(struct ofproto *p, bool revalidate_all)
1165 if (p->need_revalidate || revalidate_all
1166 || !tag_set_is_empty(&p->revalidate_set)) {
1167 struct revalidate_cbdata cbdata;
1169 cbdata.revalidate_all = revalidate_all;
1170 cbdata.revalidate_subrules = p->need_revalidate;
1171 cbdata.revalidate_set = p->revalidate_set;
1172 tag_set_init(&p->revalidate_set);
1173 COVERAGE_INC(ofproto_revalidate);
1174 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1175 p->need_revalidate = false;
1182 ofproto_wait(struct ofproto *p)
1184 struct ofservice *ofservice;
1185 struct ofconn *ofconn;
1188 dpif_recv_wait(p->dpif);
1189 dpif_port_poll_wait(p->dpif);
1190 netdev_monitor_poll_wait(p->netdev_monitor);
1191 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1192 ofconn_wait(ofconn);
1195 poll_timer_wait_until(p->next_in_band_update);
1196 in_band_wait(p->in_band);
1199 fail_open_wait(p->fail_open);
1202 ofproto_sflow_wait(p->sflow);
1204 if (!tag_set_is_empty(&p->revalidate_set)) {
1205 poll_immediate_wake();
1207 if (p->need_revalidate) {
1208 /* Shouldn't happen, but if it does just go around again. */
1209 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1210 poll_immediate_wake();
1211 } else if (p->next_expiration != LLONG_MAX) {
1212 poll_timer_wait_until(p->next_expiration);
1214 HMAP_FOR_EACH (ofservice, node, &p->services) {
1215 pvconn_wait(ofservice->pvconn);
1217 for (i = 0; i < p->n_snoops; i++) {
1218 pvconn_wait(p->snoops[i]);
1223 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1225 tag_set_add(&ofproto->revalidate_set, tag);
1229 ofproto_get_revalidate_set(struct ofproto *ofproto)
1231 return &ofproto->revalidate_set;
1235 ofproto_is_alive(const struct ofproto *p)
1237 return !hmap_is_empty(&p->controllers);
1240 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1242 * This is almost the same as calling dpif_port_del() directly on the
1243 * datapath, but it also makes 'ofproto' close its open netdev for the port
1244 * (if any). This makes it possible to create a new netdev of a different
1245 * type under the same name, which otherwise the netdev library would refuse
1246 * to do because of the conflict. (The netdev would eventually get closed on
1247 * the next trip through ofproto_run(), but this interface is more direct.)
1249 * Returns 0 if successful, otherwise a positive errno. */
1251 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1253 struct ofport *ofport = get_port(ofproto, odp_port);
1254 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1257 error = dpif_port_del(ofproto->dpif, odp_port);
1259 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1260 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1261 } else if (ofport) {
1262 /* 'name' is ofport->opp.name and update_port() is going to destroy
1263 * 'ofport'. Just in case update_port() refers to 'name' after it
1264 * destroys 'ofport', make a copy of it around the update_port()
1266 char *devname = xstrdup(name);
1267 update_port(ofproto, devname);
1273 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1274 * true if 'odp_port' exists and should be included, false otherwise. */
1276 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1278 struct ofport *ofport = get_port(ofproto, odp_port);
1279 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1283 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1284 const union ofp_action *actions, size_t n_actions,
1285 const struct ofpbuf *packet)
1287 struct odp_actions odp_actions;
1290 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1296 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1298 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
1303 ofproto_add_flow(struct ofproto *p, const struct flow *flow,
1304 uint32_t wildcards, unsigned int priority,
1305 const union ofp_action *actions, size_t n_actions,
1309 rule = rule_create(p, NULL, actions, n_actions,
1310 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1312 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1313 rule_insert(p, rule, NULL, 0);
1317 ofproto_delete_flow(struct ofproto *ofproto, const struct flow *flow,
1318 uint32_t wildcards, unsigned int priority)
1320 struct cls_rule target;
1323 cls_rule_from_flow(flow, wildcards, priority, &target);
1324 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1327 rule_remove(ofproto, rule);
1332 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1334 struct rule *rule = rule_from_cls_rule(rule_);
1335 struct ofproto *ofproto = ofproto_;
1337 /* Mark the flow as not installed, even though it might really be
1338 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1339 * There is no point in uninstalling it individually since we are about to
1340 * blow away all the flows with dpif_flow_flush(). */
1341 rule->installed = false;
1343 rule_remove(ofproto, rule);
1347 ofproto_flush_flows(struct ofproto *ofproto)
1349 COVERAGE_INC(ofproto_flush);
1350 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1351 dpif_flow_flush(ofproto->dpif);
1352 if (ofproto->in_band) {
1353 in_band_flushed(ofproto->in_band);
1355 if (ofproto->fail_open) {
1356 fail_open_flushed(ofproto->fail_open);
1361 reinit_ports(struct ofproto *p)
1363 struct svec devnames;
1364 struct ofport *ofport;
1365 struct odp_port *odp_ports;
1369 COVERAGE_INC(ofproto_reinit_ports);
1371 svec_init(&devnames);
1372 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1373 svec_add (&devnames, (char *) ofport->opp.name);
1375 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1376 for (i = 0; i < n_odp_ports; i++) {
1377 svec_add (&devnames, odp_ports[i].devname);
1381 svec_sort_unique(&devnames);
1382 for (i = 0; i < devnames.n; i++) {
1383 update_port(p, devnames.names[i]);
1385 svec_destroy(&devnames);
1388 static struct ofport *
1389 make_ofport(const struct odp_port *odp_port)
1391 struct netdev_options netdev_options;
1392 enum netdev_flags flags;
1393 struct ofport *ofport;
1394 struct netdev *netdev;
1397 memset(&netdev_options, 0, sizeof netdev_options);
1398 netdev_options.name = odp_port->devname;
1399 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1401 error = netdev_open(&netdev_options, &netdev);
1403 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1404 "cannot be opened (%s)",
1405 odp_port->devname, odp_port->port,
1406 odp_port->devname, strerror(error));
1410 ofport = xmalloc(sizeof *ofport);
1411 ofport->netdev = netdev;
1412 ofport->odp_port = odp_port->port;
1413 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1414 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1415 memcpy(ofport->opp.name, odp_port->devname,
1416 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1417 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1419 netdev_get_flags(netdev, &flags);
1420 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1422 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1424 netdev_get_features(netdev,
1425 &ofport->opp.curr, &ofport->opp.advertised,
1426 &ofport->opp.supported, &ofport->opp.peer);
1431 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1433 if (get_port(p, odp_port->port)) {
1434 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1437 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1438 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1447 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1449 const struct ofp_phy_port *a = &a_->opp;
1450 const struct ofp_phy_port *b = &b_->opp;
1452 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1453 return (a->port_no == b->port_no
1454 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1455 && !strcmp((char *) a->name, (char *) b->name)
1456 && a->state == b->state
1457 && a->config == b->config
1458 && a->curr == b->curr
1459 && a->advertised == b->advertised
1460 && a->supported == b->supported
1461 && a->peer == b->peer);
1465 send_port_status(struct ofproto *p, const struct ofport *ofport,
1468 /* XXX Should limit the number of queued port status change messages. */
1469 struct ofconn *ofconn;
1470 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1471 struct ofp_port_status *ops;
1474 if (!ofconn_receives_async_msgs(ofconn)) {
1478 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1479 ops->reason = reason;
1480 ops->desc = ofport->opp;
1481 hton_ofp_phy_port(&ops->desc);
1482 queue_tx(b, ofconn, NULL);
1487 ofport_install(struct ofproto *p, struct ofport *ofport)
1489 const char *netdev_name = (const char *) ofport->opp.name;
1491 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1492 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1493 shash_add(&p->port_by_name, netdev_name, ofport);
1495 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1500 ofport_remove(struct ofproto *p, struct ofport *ofport)
1502 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1503 hmap_remove(&p->ports, &ofport->hmap_node);
1504 shash_delete(&p->port_by_name,
1505 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1507 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1512 ofport_free(struct ofport *ofport)
1515 netdev_close(ofport->netdev);
1520 static struct ofport *
1521 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1523 struct ofport *port;
1525 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1526 hash_int(odp_port, 0), &ofproto->ports) {
1527 if (port->odp_port == odp_port) {
1535 update_port(struct ofproto *p, const char *devname)
1537 struct odp_port odp_port;
1538 struct ofport *old_ofport;
1539 struct ofport *new_ofport;
1542 COVERAGE_INC(ofproto_update_port);
1544 /* Query the datapath for port information. */
1545 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1547 /* Find the old ofport. */
1548 old_ofport = shash_find_data(&p->port_by_name, devname);
1551 /* There's no port named 'devname' but there might be a port with
1552 * the same port number. This could happen if a port is deleted
1553 * and then a new one added in its place very quickly, or if a port
1554 * is renamed. In the former case we want to send an OFPPR_DELETE
1555 * and an OFPPR_ADD, and in the latter case we want to send a
1556 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1557 * the old port's ifindex against the new port, or perhaps less
1558 * reliably but more portably by comparing the old port's MAC
1559 * against the new port's MAC. However, this code isn't that smart
1560 * and always sends an OFPPR_MODIFY (XXX). */
1561 old_ofport = get_port(p, odp_port.port);
1563 } else if (error != ENOENT && error != ENODEV) {
1564 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1565 "%s", strerror(error));
1569 /* Create a new ofport. */
1570 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1572 /* Eliminate a few pathological cases. */
1573 if (!old_ofport && !new_ofport) {
1575 } else if (old_ofport && new_ofport) {
1576 /* Most of the 'config' bits are OpenFlow soft state, but
1577 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1578 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1579 * leaves the other bits 0.) */
1580 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1582 if (ofport_equal(old_ofport, new_ofport)) {
1583 /* False alarm--no change. */
1584 ofport_free(new_ofport);
1589 /* Now deal with the normal cases. */
1591 ofport_remove(p, old_ofport);
1594 ofport_install(p, new_ofport);
1596 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1597 (!old_ofport ? OFPPR_ADD
1598 : !new_ofport ? OFPPR_DELETE
1600 ofport_free(old_ofport);
1604 init_ports(struct ofproto *p)
1606 struct odp_port *ports;
1611 error = dpif_port_list(p->dpif, &ports, &n_ports);
1616 for (i = 0; i < n_ports; i++) {
1617 const struct odp_port *odp_port = &ports[i];
1618 if (!ofport_conflicts(p, odp_port)) {
1619 struct ofport *ofport = make_ofport(odp_port);
1621 ofport_install(p, ofport);
1629 static struct ofconn *
1630 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1632 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1633 ofconn->ofproto = p;
1634 list_push_back(&p->all_conns, &ofconn->node);
1635 ofconn->rconn = rconn;
1636 ofconn->type = type;
1637 ofconn->role = NX_ROLE_OTHER;
1638 ofconn->packet_in_counter = rconn_packet_counter_create ();
1639 ofconn->pktbuf = NULL;
1640 ofconn->miss_send_len = 0;
1641 ofconn->reply_counter = rconn_packet_counter_create ();
1646 ofconn_destroy(struct ofconn *ofconn)
1648 if (ofconn->type == OFCONN_PRIMARY) {
1649 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1651 discovery_destroy(ofconn->discovery);
1653 list_remove(&ofconn->node);
1654 switch_status_unregister(ofconn->ss);
1655 rconn_destroy(ofconn->rconn);
1656 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1657 rconn_packet_counter_destroy(ofconn->reply_counter);
1658 pktbuf_destroy(ofconn->pktbuf);
1663 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1668 if (ofconn->discovery) {
1669 char *controller_name;
1670 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1671 discovery_question_connectivity(ofconn->discovery);
1673 if (discovery_run(ofconn->discovery, &controller_name)) {
1674 if (controller_name) {
1675 char *ofconn_name = ofconn_make_name(p, controller_name);
1676 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1679 rconn_disconnect(ofconn->rconn);
1684 for (i = 0; i < N_SCHEDULERS; i++) {
1685 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1688 rconn_run(ofconn->rconn);
1690 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1691 /* Limit the number of iterations to prevent other tasks from
1693 for (iteration = 0; iteration < 50; iteration++) {
1694 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1699 fail_open_maybe_recover(p->fail_open);
1701 handle_openflow(ofconn, p, of_msg);
1702 ofpbuf_delete(of_msg);
1706 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1707 ofconn_destroy(ofconn);
1712 ofconn_wait(struct ofconn *ofconn)
1716 if (ofconn->discovery) {
1717 discovery_wait(ofconn->discovery);
1719 for (i = 0; i < N_SCHEDULERS; i++) {
1720 pinsched_wait(ofconn->schedulers[i]);
1722 rconn_run_wait(ofconn->rconn);
1723 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1724 rconn_recv_wait(ofconn->rconn);
1726 COVERAGE_INC(ofproto_ofconn_stuck);
1730 /* Returns true if 'ofconn' should receive asynchronous messages. */
1732 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1734 if (ofconn->type == OFCONN_PRIMARY) {
1735 /* Primary controllers always get asynchronous messages unless they
1736 * have configured themselves as "slaves". */
1737 return ofconn->role != NX_ROLE_SLAVE;
1739 /* Service connections don't get asynchronous messages unless they have
1740 * explicitly asked for them by setting a nonzero miss send length. */
1741 return ofconn->miss_send_len > 0;
1745 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1746 * and 'target', suitable for use in log messages for identifying the
1749 * The name is dynamically allocated. The caller should free it (with free())
1750 * when it is no longer needed. */
1752 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1754 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1758 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1762 for (i = 0; i < N_SCHEDULERS; i++) {
1763 struct pinsched **s = &ofconn->schedulers[i];
1767 *s = pinsched_create(rate, burst,
1768 ofconn->ofproto->switch_status);
1770 pinsched_set_limits(*s, rate, burst);
1773 pinsched_destroy(*s);
1780 ofservice_reconfigure(struct ofservice *ofservice,
1781 const struct ofproto_controller *c)
1783 ofservice->probe_interval = c->probe_interval;
1784 ofservice->rate_limit = c->rate_limit;
1785 ofservice->burst_limit = c->burst_limit;
1788 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1789 * positive errno value. */
1791 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1793 struct ofservice *ofservice;
1794 struct pvconn *pvconn;
1797 error = pvconn_open(c->target, &pvconn);
1802 ofservice = xzalloc(sizeof *ofservice);
1803 hmap_insert(&ofproto->services, &ofservice->node,
1804 hash_string(c->target, 0));
1805 ofservice->pvconn = pvconn;
1807 ofservice_reconfigure(ofservice, c);
1813 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1815 hmap_remove(&ofproto->services, &ofservice->node);
1816 pvconn_close(ofservice->pvconn);
1820 /* Finds and returns the ofservice within 'ofproto' that has the given
1821 * 'target', or a null pointer if none exists. */
1822 static struct ofservice *
1823 ofservice_lookup(struct ofproto *ofproto, const char *target)
1825 struct ofservice *ofservice;
1827 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1828 &ofproto->services) {
1829 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1836 /* Caller is responsible for initializing the 'cr' member of the returned
1838 static struct rule *
1839 rule_create(struct ofproto *ofproto, struct rule *super,
1840 const union ofp_action *actions, size_t n_actions,
1841 uint16_t idle_timeout, uint16_t hard_timeout,
1842 ovs_be64 flow_cookie, bool send_flow_removed)
1844 struct rule *rule = xzalloc(sizeof *rule);
1845 rule->idle_timeout = idle_timeout;
1846 rule->hard_timeout = hard_timeout;
1847 rule->flow_cookie = flow_cookie;
1848 rule->used = rule->created = time_msec();
1849 rule->send_flow_removed = send_flow_removed;
1850 rule->super = super;
1852 list_push_back(&super->list, &rule->list);
1854 list_init(&rule->list);
1856 if (n_actions > 0) {
1857 rule->n_actions = n_actions;
1858 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1860 netflow_flow_clear(&rule->nf_flow);
1861 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1866 static struct rule *
1867 rule_from_cls_rule(const struct cls_rule *cls_rule)
1869 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1873 rule_free(struct rule *rule)
1875 free(rule->actions);
1876 free(rule->odp_actions);
1880 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1881 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1882 * through all of its subrules and revalidates them, destroying any that no
1883 * longer has a super-rule (which is probably all of them).
1885 * Before calling this function, the caller must make have removed 'rule' from
1886 * the classifier. If 'rule' is an exact-match rule, the caller is also
1887 * responsible for ensuring that it has been uninstalled from the datapath. */
1889 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1892 struct rule *subrule, *next;
1893 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
1894 revalidate_rule(ofproto, subrule);
1897 list_remove(&rule->list);
1903 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1905 const union ofp_action *oa;
1906 struct actions_iterator i;
1908 if (out_port == htons(OFPP_NONE)) {
1911 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1912 oa = actions_next(&i)) {
1913 if (action_outputs_to_port(oa, out_port)) {
1920 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1921 * 'packet', which arrived on 'in_port'.
1923 * Takes ownership of 'packet'. */
1925 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1926 const union odp_action *actions, size_t n_actions,
1927 struct ofpbuf *packet)
1929 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1930 /* As an optimization, avoid a round-trip from userspace to kernel to
1931 * userspace. This also avoids possibly filling up kernel packet
1932 * buffers along the way. */
1933 struct odp_msg *msg;
1935 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1936 msg->type = _ODPL_ACTION_NR;
1937 msg->length = sizeof(struct odp_msg) + packet->size;
1938 msg->port = in_port;
1940 msg->arg = actions[0].controller.arg;
1942 send_packet_in(ofproto, packet);
1948 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
1949 ofpbuf_delete(packet);
1954 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1955 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1956 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1958 * The flow that 'packet' actually contains does not need to actually match
1959 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1960 * the packet and byte counters for 'rule' will be credited for the packet sent
1961 * out whether or not the packet actually matches 'rule'.
1963 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1964 * the caller must already have accurately composed ODP actions for it given
1965 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1966 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1967 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1968 * actions and apply them to 'packet'.
1970 * Takes ownership of 'packet'. */
1972 rule_execute(struct ofproto *ofproto, struct rule *rule,
1973 struct ofpbuf *packet, const struct flow *flow)
1975 const union odp_action *actions;
1976 struct odp_flow_stats stats;
1978 struct odp_actions a;
1980 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1982 /* Grab or compose the ODP actions.
1984 * The special case for an exact-match 'rule' where 'flow' is not the
1985 * rule's flow is important to avoid, e.g., sending a packet out its input
1986 * port simply because the ODP actions were composed for the wrong
1988 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1989 struct rule *super = rule->super ? rule->super : rule;
1990 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1991 packet, &a, NULL, 0, NULL)) {
1992 ofpbuf_delete(packet);
1995 actions = a.actions;
1996 n_actions = a.n_actions;
1998 actions = rule->odp_actions;
1999 n_actions = rule->n_odp_actions;
2002 /* Execute the ODP actions. */
2003 flow_extract_stats(flow, packet, &stats);
2004 if (execute_odp_actions(ofproto, flow->in_port,
2005 actions, n_actions, packet)) {
2006 update_stats(ofproto, rule, &stats);
2007 rule->used = time_msec();
2008 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2012 /* Inserts 'rule' into 'p''s flow table.
2014 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2015 * actions on it and credits the statistics for sending the packet to 'rule'.
2016 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2019 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2022 struct rule *displaced_rule;
2024 /* Insert the rule in the classifier. */
2025 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2026 if (!rule->cr.wc.wildcards) {
2027 rule_make_actions(p, rule, packet);
2030 /* Send the packet and credit it to the rule. */
2033 flow_extract(packet, 0, in_port, &flow);
2034 rule_execute(p, rule, packet, &flow);
2037 /* Install the rule in the datapath only after sending the packet, to
2038 * avoid packet reordering. */
2039 if (rule->cr.wc.wildcards) {
2040 COVERAGE_INC(ofproto_add_wc_flow);
2041 p->need_revalidate = true;
2043 rule_install(p, rule, displaced_rule);
2046 /* Free the rule that was displaced, if any. */
2047 if (displaced_rule) {
2048 rule_destroy(p, displaced_rule);
2052 static struct rule *
2053 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2054 const struct flow *flow)
2056 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2057 rule->idle_timeout, rule->hard_timeout,
2059 COVERAGE_INC(ofproto_subrule_create);
2060 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2061 : rule->cr.priority), &subrule->cr);
2063 if (classifier_insert(&ofproto->cls, &subrule->cr)) {
2071 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2073 * - If 'rule' was installed in the datapath, uninstalls it and updates
2074 * 'rule''s statistics (or its super-rule's statistics, if it is a
2075 * subrule), via rule_uninstall().
2077 * - Removes 'rule' from the classifier.
2079 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2080 * uninstalls and destroys) its subrules, via rule_destroy().
2083 rule_remove(struct ofproto *ofproto, struct rule *rule)
2085 if (rule->cr.wc.wildcards) {
2086 COVERAGE_INC(ofproto_del_wc_flow);
2087 ofproto->need_revalidate = true;
2089 rule_uninstall(ofproto, rule);
2091 classifier_remove(&ofproto->cls, &rule->cr);
2092 rule_destroy(ofproto, rule);
2095 /* Returns true if the actions changed, false otherwise. */
2097 rule_make_actions(struct ofproto *p, struct rule *rule,
2098 const struct ofpbuf *packet)
2100 const struct rule *super;
2101 struct odp_actions a;
2104 assert(!rule->cr.wc.wildcards);
2106 super = rule->super ? rule->super : rule;
2108 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2109 packet, &a, &rule->tags, &rule->may_install,
2110 &rule->nf_flow.output_iface);
2112 actions_len = a.n_actions * sizeof *a.actions;
2113 if (rule->n_odp_actions != a.n_actions
2114 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2115 COVERAGE_INC(ofproto_odp_unchanged);
2116 free(rule->odp_actions);
2117 rule->n_odp_actions = a.n_actions;
2118 rule->odp_actions = xmemdup(a.actions, actions_len);
2126 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2127 struct odp_flow_put *put)
2129 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2130 odp_flow_key_from_flow(&put->flow.key, &rule->cr.flow);
2131 put->flow.actions = rule->odp_actions;
2132 put->flow.n_actions = rule->n_odp_actions;
2133 put->flow.flags = 0;
2135 return dpif_flow_put(ofproto->dpif, put);
2139 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2141 assert(!rule->cr.wc.wildcards);
2143 if (rule->may_install) {
2144 struct odp_flow_put put;
2145 if (!do_put_flow(p, rule,
2146 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2148 rule->installed = true;
2149 if (displaced_rule) {
2150 update_stats(p, displaced_rule, &put.flow.stats);
2151 rule_post_uninstall(p, displaced_rule);
2154 } else if (displaced_rule) {
2155 rule_uninstall(p, displaced_rule);
2160 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2162 if (rule->installed) {
2163 struct odp_flow_put put;
2164 COVERAGE_INC(ofproto_dp_missed);
2165 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2167 rule_install(ofproto, rule, NULL);
2172 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2174 bool actions_changed;
2175 uint16_t new_out_iface, old_out_iface;
2177 old_out_iface = rule->nf_flow.output_iface;
2178 actions_changed = rule_make_actions(ofproto, rule, NULL);
2180 if (rule->may_install) {
2181 if (rule->installed) {
2182 if (actions_changed) {
2183 struct odp_flow_put put;
2184 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2185 | ODPPF_ZERO_STATS, &put);
2186 update_stats(ofproto, rule, &put.flow.stats);
2188 /* Temporarily set the old output iface so that NetFlow
2189 * messages have the correct output interface for the old
2191 new_out_iface = rule->nf_flow.output_iface;
2192 rule->nf_flow.output_iface = old_out_iface;
2193 rule_post_uninstall(ofproto, rule);
2194 rule->nf_flow.output_iface = new_out_iface;
2197 rule_install(ofproto, rule, NULL);
2200 rule_uninstall(ofproto, rule);
2205 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2207 uint64_t total_bytes = rule->byte_count + extra_bytes;
2209 if (ofproto->ofhooks->account_flow_cb
2210 && total_bytes > rule->accounted_bytes)
2212 ofproto->ofhooks->account_flow_cb(
2213 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2214 total_bytes - rule->accounted_bytes, ofproto->aux);
2215 rule->accounted_bytes = total_bytes;
2219 /* 'rule' must be an exact-match rule in 'p'.
2221 * If 'rule' is installed in the datapath, uninstalls it and updates's
2222 * statistics. If 'rule' is a subrule, the statistics that are updated are
2223 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2226 * If 'rule' is not installed, this function has no effect. */
2228 rule_uninstall(struct ofproto *p, struct rule *rule)
2230 assert(!rule->cr.wc.wildcards);
2231 if (rule->installed) {
2232 struct odp_flow odp_flow;
2234 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
2235 odp_flow.actions = NULL;
2236 odp_flow.n_actions = 0;
2238 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2239 update_stats(p, rule, &odp_flow.stats);
2241 rule->installed = false;
2243 rule_post_uninstall(p, rule);
2248 is_controller_rule(struct rule *rule)
2250 /* If the only action is send to the controller then don't report
2251 * NetFlow expiration messages since it is just part of the control
2252 * logic for the network and not real traffic. */
2256 && rule->super->n_actions == 1
2257 && action_outputs_to_port(&rule->super->actions[0],
2258 htons(OFPP_CONTROLLER)));
2262 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2264 struct rule *super = rule->super;
2266 rule_account(ofproto, rule, 0);
2268 if (ofproto->netflow && !is_controller_rule(rule)) {
2269 struct ofexpired expired;
2270 expired.flow = rule->cr.flow;
2271 expired.packet_count = rule->packet_count;
2272 expired.byte_count = rule->byte_count;
2273 expired.used = rule->used;
2274 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2277 super->packet_count += rule->packet_count;
2278 super->byte_count += rule->byte_count;
2280 /* Reset counters to prevent double counting if the rule ever gets
2282 rule->packet_count = 0;
2283 rule->byte_count = 0;
2284 rule->accounted_bytes = 0;
2286 netflow_flow_clear(&rule->nf_flow);
2291 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2292 struct rconn_packet_counter *counter)
2294 update_openflow_length(msg);
2295 if (rconn_send(ofconn->rconn, msg, counter)) {
2301 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2304 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2306 COVERAGE_INC(ofproto_error);
2307 queue_tx(buf, ofconn, ofconn->reply_counter);
2312 hton_ofp_phy_port(struct ofp_phy_port *opp)
2314 opp->port_no = htons(opp->port_no);
2315 opp->config = htonl(opp->config);
2316 opp->state = htonl(opp->state);
2317 opp->curr = htonl(opp->curr);
2318 opp->advertised = htonl(opp->advertised);
2319 opp->supported = htonl(opp->supported);
2320 opp->peer = htonl(opp->peer);
2324 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2326 struct ofp_header *rq = oh;
2327 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2332 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2333 struct ofp_header *oh)
2335 struct ofp_switch_features *osf;
2337 struct ofport *port;
2339 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2340 osf->datapath_id = htonll(p->datapath_id);
2341 osf->n_buffers = htonl(pktbuf_capacity());
2343 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2344 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2345 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2346 (1u << OFPAT_SET_VLAN_VID) |
2347 (1u << OFPAT_SET_VLAN_PCP) |
2348 (1u << OFPAT_STRIP_VLAN) |
2349 (1u << OFPAT_SET_DL_SRC) |
2350 (1u << OFPAT_SET_DL_DST) |
2351 (1u << OFPAT_SET_NW_SRC) |
2352 (1u << OFPAT_SET_NW_DST) |
2353 (1u << OFPAT_SET_NW_TOS) |
2354 (1u << OFPAT_SET_TP_SRC) |
2355 (1u << OFPAT_SET_TP_DST) |
2356 (1u << OFPAT_ENQUEUE));
2358 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2359 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2362 queue_tx(buf, ofconn, ofconn->reply_counter);
2367 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2368 struct ofp_header *oh)
2371 struct ofp_switch_config *osc;
2375 /* Figure out flags. */
2376 dpif_get_drop_frags(p->dpif, &drop_frags);
2377 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2380 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2381 osc->flags = htons(flags);
2382 osc->miss_send_len = htons(ofconn->miss_send_len);
2383 queue_tx(buf, ofconn, ofconn->reply_counter);
2389 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2390 struct ofp_switch_config *osc)
2395 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2399 flags = ntohs(osc->flags);
2401 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2402 switch (flags & OFPC_FRAG_MASK) {
2403 case OFPC_FRAG_NORMAL:
2404 dpif_set_drop_frags(p->dpif, false);
2406 case OFPC_FRAG_DROP:
2407 dpif_set_drop_frags(p->dpif, true);
2410 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2416 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2422 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2424 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2425 a->controller.arg = max_len;
2428 struct action_xlate_ctx {
2430 struct flow flow; /* Flow to which these actions correspond. */
2431 int recurse; /* Recursion level, via xlate_table_action. */
2432 struct ofproto *ofproto;
2433 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2434 * null pointer if we are revalidating
2435 * without a packet to refer to. */
2438 struct odp_actions *out; /* Datapath actions. */
2439 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2440 bool may_set_up_flow; /* True ordinarily; false if the actions must
2441 * be reassessed for every packet. */
2442 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2445 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2446 * flow translation. */
2447 #define MAX_RESUBMIT_RECURSION 8
2449 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2450 struct action_xlate_ctx *ctx);
2453 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2455 const struct ofport *ofport = get_port(ctx->ofproto, port);
2458 if (ofport->opp.config & OFPPC_NO_FWD) {
2459 /* Forwarding disabled on port. */
2464 * We don't have an ofport record for this port, but it doesn't hurt to
2465 * allow forwarding to it anyhow. Maybe such a port will appear later
2466 * and we're pre-populating the flow table.
2470 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2471 ctx->nf_output_iface = port;
2474 static struct rule *
2475 lookup_valid_rule(struct ofproto *ofproto, const struct flow *flow)
2478 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow,
2481 /* The rule we found might not be valid, since we could be in need of
2482 * revalidation. If it is not valid, don't return it. */
2485 && ofproto->need_revalidate
2486 && !revalidate_rule(ofproto, rule)) {
2487 COVERAGE_INC(ofproto_invalidated);
2495 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2497 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2498 uint16_t old_in_port;
2501 /* Look up a flow with 'in_port' as the input port. Then restore the
2502 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2503 * have surprising behavior). */
2504 old_in_port = ctx->flow.in_port;
2505 ctx->flow.in_port = in_port;
2506 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2507 ctx->flow.in_port = old_in_port;
2515 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2519 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2521 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2522 MAX_RESUBMIT_RECURSION);
2527 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2528 uint16_t *nf_output_iface, struct odp_actions *actions)
2530 struct ofport *ofport;
2532 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2533 uint16_t odp_port = ofport->odp_port;
2534 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2535 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2538 *nf_output_iface = NF_OUT_FLOOD;
2542 xlate_output_action__(struct action_xlate_ctx *ctx,
2543 uint16_t port, uint16_t max_len)
2546 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2548 ctx->nf_output_iface = NF_OUT_DROP;
2552 add_output_action(ctx, ctx->flow.in_port);
2555 xlate_table_action(ctx, ctx->flow.in_port);
2558 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2559 ctx->out, ctx->tags,
2560 &ctx->nf_output_iface,
2561 ctx->ofproto->aux)) {
2562 COVERAGE_INC(ofproto_uninstallable);
2563 ctx->may_set_up_flow = false;
2567 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2568 &ctx->nf_output_iface, ctx->out);
2571 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2572 &ctx->nf_output_iface, ctx->out);
2574 case OFPP_CONTROLLER:
2575 add_controller_action(ctx->out, max_len);
2578 add_output_action(ctx, ODPP_LOCAL);
2581 odp_port = ofp_port_to_odp_port(port);
2582 if (odp_port != ctx->flow.in_port) {
2583 add_output_action(ctx, odp_port);
2588 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2589 ctx->nf_output_iface = NF_OUT_FLOOD;
2590 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2591 ctx->nf_output_iface = prev_nf_output_iface;
2592 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2593 ctx->nf_output_iface != NF_OUT_FLOOD) {
2594 ctx->nf_output_iface = NF_OUT_MULTI;
2599 xlate_output_action(struct action_xlate_ctx *ctx,
2600 const struct ofp_action_output *oao)
2602 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2605 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2606 * optimization, because we're going to add another action that sets the
2607 * priority immediately after, or because there are no actions following the
2610 remove_pop_action(struct action_xlate_ctx *ctx)
2612 size_t n = ctx->out->n_actions;
2613 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2614 ctx->out->n_actions--;
2619 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2620 const struct ofp_action_enqueue *oae)
2622 uint16_t ofp_port, odp_port;
2626 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2629 /* Fall back to ordinary output action. */
2630 xlate_output_action__(ctx, ntohs(oae->port), 0);
2634 /* Figure out ODP output port. */
2635 ofp_port = ntohs(oae->port);
2636 if (ofp_port != OFPP_IN_PORT) {
2637 odp_port = ofp_port_to_odp_port(ofp_port);
2639 odp_port = ctx->flow.in_port;
2642 /* Add ODP actions. */
2643 remove_pop_action(ctx);
2644 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2646 add_output_action(ctx, odp_port);
2647 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2649 /* Update NetFlow output port. */
2650 if (ctx->nf_output_iface == NF_OUT_DROP) {
2651 ctx->nf_output_iface = odp_port;
2652 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2653 ctx->nf_output_iface = NF_OUT_MULTI;
2658 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2659 const struct nx_action_set_queue *nasq)
2664 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2667 /* Couldn't translate queue to a priority, so ignore. A warning
2668 * has already been logged. */
2672 remove_pop_action(ctx);
2673 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2678 xlate_nicira_action(struct action_xlate_ctx *ctx,
2679 const struct nx_action_header *nah)
2681 const struct nx_action_resubmit *nar;
2682 const struct nx_action_set_tunnel *nast;
2683 const struct nx_action_set_queue *nasq;
2684 union odp_action *oa;
2685 int subtype = ntohs(nah->subtype);
2687 assert(nah->vendor == htonl(NX_VENDOR_ID));
2689 case NXAST_RESUBMIT:
2690 nar = (const struct nx_action_resubmit *) nah;
2691 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2694 case NXAST_SET_TUNNEL:
2695 nast = (const struct nx_action_set_tunnel *) nah;
2696 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2697 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2700 case NXAST_DROP_SPOOFED_ARP:
2701 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2702 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2706 case NXAST_SET_QUEUE:
2707 nasq = (const struct nx_action_set_queue *) nah;
2708 xlate_set_queue_action(ctx, nasq);
2711 case NXAST_POP_QUEUE:
2712 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2715 /* If you add a new action here that modifies flow data, don't forget to
2716 * update the flow key in ctx->flow at the same time. */
2719 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2725 do_xlate_actions(const union ofp_action *in, size_t n_in,
2726 struct action_xlate_ctx *ctx)
2728 struct actions_iterator iter;
2729 const union ofp_action *ia;
2730 const struct ofport *port;
2732 port = get_port(ctx->ofproto, ctx->flow.in_port);
2733 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2734 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2735 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2736 /* Drop this flow. */
2740 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2741 uint16_t type = ntohs(ia->type);
2742 union odp_action *oa;
2746 xlate_output_action(ctx, &ia->output);
2749 case OFPAT_SET_VLAN_VID:
2750 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2751 oa->dl_tci.tci = ia->vlan_vid.vlan_vid;
2752 oa->dl_tci.tci |= htons(ctx->flow.dl_vlan_pcp << VLAN_PCP_SHIFT);
2753 ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
2756 case OFPAT_SET_VLAN_PCP:
2757 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2758 oa->dl_tci.tci = htons(ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT);
2759 oa->dl_tci.tci |= ctx->flow.dl_vlan;
2760 ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
2763 case OFPAT_STRIP_VLAN:
2764 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2765 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2766 ctx->flow.dl_vlan_pcp = 0;
2769 case OFPAT_SET_DL_SRC:
2770 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2771 memcpy(oa->dl_addr.dl_addr,
2772 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2773 memcpy(ctx->flow.dl_src,
2774 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2777 case OFPAT_SET_DL_DST:
2778 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2779 memcpy(oa->dl_addr.dl_addr,
2780 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2781 memcpy(ctx->flow.dl_dst,
2782 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2785 case OFPAT_SET_NW_SRC:
2786 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2787 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2790 case OFPAT_SET_NW_DST:
2791 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2792 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2795 case OFPAT_SET_NW_TOS:
2796 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2797 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2800 case OFPAT_SET_TP_SRC:
2801 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2802 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2805 case OFPAT_SET_TP_DST:
2806 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2807 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2811 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2815 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2819 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2826 xlate_actions(const union ofp_action *in, size_t n_in,
2827 const struct flow *flow, struct ofproto *ofproto,
2828 const struct ofpbuf *packet,
2829 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2830 uint16_t *nf_output_iface)
2832 tag_type no_tags = 0;
2833 struct action_xlate_ctx ctx;
2834 COVERAGE_INC(ofproto_ofp2odp);
2835 odp_actions_init(out);
2838 ctx.ofproto = ofproto;
2839 ctx.packet = packet;
2841 ctx.tags = tags ? tags : &no_tags;
2842 ctx.may_set_up_flow = true;
2843 ctx.nf_output_iface = NF_OUT_DROP;
2844 do_xlate_actions(in, n_in, &ctx);
2845 remove_pop_action(&ctx);
2847 /* Check with in-band control to see if we're allowed to set up this
2849 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2850 ctx.may_set_up_flow = false;
2853 if (may_set_up_flow) {
2854 *may_set_up_flow = ctx.may_set_up_flow;
2856 if (nf_output_iface) {
2857 *nf_output_iface = ctx.nf_output_iface;
2859 if (odp_actions_overflow(out)) {
2860 COVERAGE_INC(odp_overflow);
2861 odp_actions_init(out);
2862 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2867 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2868 * error message code (composed with ofp_mkerr()) for the caller to propagate
2869 * upward. Otherwise, returns 0.
2871 * 'oh' is used to make log messages more informative. */
2873 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2875 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2876 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2879 type_name = ofp_message_type_to_string(oh->type);
2880 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2884 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2891 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2892 struct ofp_header *oh)
2894 struct ofp_packet_out *opo;
2895 struct ofpbuf payload, *buffer;
2896 struct odp_actions actions;
2902 error = reject_slave_controller(ofconn, oh);
2907 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2911 opo = (struct ofp_packet_out *) oh;
2913 COVERAGE_INC(ofproto_packet_out);
2914 if (opo->buffer_id != htonl(UINT32_MAX)) {
2915 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2917 if (error || !buffer) {
2925 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2926 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2927 &flow, p, &payload, &actions, NULL, NULL, NULL);
2929 dpif_execute(p->dpif, actions.actions, actions.n_actions, &payload);
2931 ofpbuf_delete(buffer);
2937 update_port_config(struct ofproto *p, struct ofport *port,
2938 uint32_t config, uint32_t mask)
2940 mask &= config ^ port->opp.config;
2941 if (mask & OFPPC_PORT_DOWN) {
2942 if (config & OFPPC_PORT_DOWN) {
2943 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2945 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2948 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2949 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2950 if (mask & REVALIDATE_BITS) {
2951 COVERAGE_INC(ofproto_costly_flags);
2952 port->opp.config ^= mask & REVALIDATE_BITS;
2953 p->need_revalidate = true;
2955 #undef REVALIDATE_BITS
2956 if (mask & OFPPC_NO_PACKET_IN) {
2957 port->opp.config ^= OFPPC_NO_PACKET_IN;
2962 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2963 struct ofp_header *oh)
2965 const struct ofp_port_mod *opm;
2966 struct ofport *port;
2969 error = reject_slave_controller(ofconn, oh);
2973 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2977 opm = (struct ofp_port_mod *) oh;
2979 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2981 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2982 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2983 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2985 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2986 if (opm->advertise) {
2987 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2993 static struct ofpbuf *
2994 make_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2996 struct ofp_stats_reply *osr;
2999 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3000 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3002 osr->flags = htons(0);
3006 static struct ofpbuf *
3007 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3009 return make_stats_reply(request->header.xid, request->type, body_len);
3013 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
3015 struct ofpbuf *msg = *msgp;
3016 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3017 if (nbytes + msg->size > UINT16_MAX) {
3018 struct ofp_stats_reply *reply = msg->data;
3019 reply->flags = htons(OFPSF_REPLY_MORE);
3020 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3021 queue_tx(msg, ofconn, ofconn->reply_counter);
3023 return ofpbuf_put_uninit(*msgp, nbytes);
3027 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3028 struct ofp_stats_request *request)
3030 struct ofp_desc_stats *ods;
3033 msg = start_stats_reply(request, sizeof *ods);
3034 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3035 memset(ods, 0, sizeof *ods);
3036 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3037 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3038 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3039 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3040 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3041 queue_tx(msg, ofconn, ofconn->reply_counter);
3047 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3048 struct ofp_stats_request *request)
3050 struct ofp_table_stats *ots;
3055 msg = start_stats_reply(request, sizeof *ots * 2);
3057 /* Count rules other than subrules. */
3058 n_rules = classifier_count(&p->cls);
3059 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) {
3065 /* Classifier table. */
3066 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3067 memset(ots, 0, sizeof *ots);
3068 strcpy(ots->name, "classifier");
3069 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3071 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3072 ots->active_count = htonl(n_rules);
3073 ots->lookup_count = htonll(0); /* XXX */
3074 ots->matched_count = htonll(0); /* XXX */
3076 queue_tx(msg, ofconn, ofconn->reply_counter);
3081 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3082 struct ofpbuf **msgp)
3084 struct netdev_stats stats;
3085 struct ofp_port_stats *ops;
3087 /* Intentionally ignore return value, since errors will set
3088 * 'stats' to all-1s, which is correct for OpenFlow, and
3089 * netdev_get_stats() will log errors. */
3090 netdev_get_stats(port->netdev, &stats);
3092 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3093 ops->port_no = htons(port->opp.port_no);
3094 memset(ops->pad, 0, sizeof ops->pad);
3095 ops->rx_packets = htonll(stats.rx_packets);
3096 ops->tx_packets = htonll(stats.tx_packets);
3097 ops->rx_bytes = htonll(stats.rx_bytes);
3098 ops->tx_bytes = htonll(stats.tx_bytes);
3099 ops->rx_dropped = htonll(stats.rx_dropped);
3100 ops->tx_dropped = htonll(stats.tx_dropped);
3101 ops->rx_errors = htonll(stats.rx_errors);
3102 ops->tx_errors = htonll(stats.tx_errors);
3103 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3104 ops->rx_over_err = htonll(stats.rx_over_errors);
3105 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3106 ops->collisions = htonll(stats.collisions);
3110 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3111 struct ofp_stats_request *osr,
3114 struct ofp_port_stats_request *psr;
3115 struct ofp_port_stats *ops;
3117 struct ofport *port;
3119 if (arg_size != sizeof *psr) {
3120 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3122 psr = (struct ofp_port_stats_request *) osr->body;
3124 msg = start_stats_reply(osr, sizeof *ops * 16);
3125 if (psr->port_no != htons(OFPP_NONE)) {
3126 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3128 append_port_stat(port, ofconn, &msg);
3131 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3132 append_port_stat(port, ofconn, &msg);
3136 queue_tx(msg, ofconn, ofconn->reply_counter);
3140 struct flow_stats_cbdata {
3141 struct ofproto *ofproto;
3142 struct ofconn *ofconn;
3147 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3148 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3149 * returned statistic include statistics for all of 'rule''s subrules. */
3151 query_stats(struct ofproto *p, struct rule *rule,
3152 uint64_t *packet_countp, uint64_t *byte_countp)
3154 uint64_t packet_count, byte_count;
3155 struct rule *subrule;
3156 struct odp_flow *odp_flows;
3159 /* Start from historical data for 'rule' itself that are no longer tracked
3160 * by the datapath. This counts, for example, subrules that have
3162 packet_count = rule->packet_count;
3163 byte_count = rule->byte_count;
3165 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3166 * wildcarded then on all of its subrules.
3168 * Also, add any statistics that are not tracked by the datapath for each
3169 * subrule. This includes, for example, statistics for packets that were
3170 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3172 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3173 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3174 if (rule->cr.wc.wildcards) {
3176 LIST_FOR_EACH (subrule, list, &rule->list) {
3177 odp_flow_key_from_flow(&odp_flows[i++].key, &subrule->cr.flow);
3178 packet_count += subrule->packet_count;
3179 byte_count += subrule->byte_count;
3182 odp_flow_key_from_flow(&odp_flows[0].key, &rule->cr.flow);
3185 /* Fetch up-to-date statistics from the datapath and add them in. */
3186 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3188 for (i = 0; i < n_odp_flows; i++) {
3189 struct odp_flow *odp_flow = &odp_flows[i];
3190 packet_count += odp_flow->stats.n_packets;
3191 byte_count += odp_flow->stats.n_bytes;
3196 /* Return the stats to the caller. */
3197 *packet_countp = packet_count;
3198 *byte_countp = byte_count;
3202 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3204 struct rule *rule = rule_from_cls_rule(rule_);
3205 struct flow_stats_cbdata *cbdata = cbdata_;
3206 struct ofp_flow_stats *ofs;
3207 uint64_t packet_count, byte_count;
3208 size_t act_len, len;
3209 long long int tdiff = time_msec() - rule->created;
3210 uint32_t sec = tdiff / 1000;
3211 uint32_t msec = tdiff - (sec * 1000);
3213 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3217 act_len = sizeof *rule->actions * rule->n_actions;
3218 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3220 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3222 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3223 ofs->length = htons(len);
3226 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3227 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3228 ofs->duration_sec = htonl(sec);
3229 ofs->duration_nsec = htonl(msec * 1000000);
3230 ofs->cookie = rule->flow_cookie;
3231 ofs->priority = htons(rule->cr.priority);
3232 ofs->idle_timeout = htons(rule->idle_timeout);
3233 ofs->hard_timeout = htons(rule->hard_timeout);
3234 memset(ofs->pad2, 0, sizeof ofs->pad2);
3235 ofs->packet_count = htonll(packet_count);
3236 ofs->byte_count = htonll(byte_count);
3237 if (rule->n_actions > 0) {
3238 memcpy(ofs->actions, rule->actions, act_len);
3243 table_id_to_include(uint8_t table_id)
3245 return table_id == 0 || table_id == 0xff ? CLS_INC_ALL : 0;
3249 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3250 const struct ofp_stats_request *osr,
3253 struct ofp_flow_stats_request *fsr;
3254 struct flow_stats_cbdata cbdata;
3255 struct cls_rule target;
3257 if (arg_size != sizeof *fsr) {
3258 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3260 fsr = (struct ofp_flow_stats_request *) osr->body;
3262 COVERAGE_INC(ofproto_flows_req);
3264 cbdata.ofconn = ofconn;
3265 cbdata.out_port = fsr->out_port;
3266 cbdata.msg = start_stats_reply(osr, 1024);
3267 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3268 classifier_for_each_match(&p->cls, &target,
3269 table_id_to_include(fsr->table_id),
3270 flow_stats_cb, &cbdata);
3271 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3275 struct flow_stats_ds_cbdata {
3276 struct ofproto *ofproto;
3281 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3283 struct rule *rule = rule_from_cls_rule(rule_);
3284 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3285 struct ds *results = cbdata->results;
3286 struct ofp_match match;
3287 uint64_t packet_count, byte_count;
3288 size_t act_len = sizeof *rule->actions * rule->n_actions;
3290 /* Don't report on subrules. */
3291 if (rule->super != NULL) {
3295 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3296 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3297 cbdata->ofproto->tun_id_from_cookie, &match);
3299 ds_put_format(results, "duration=%llds, ",
3300 (time_msec() - rule->created) / 1000);
3301 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3302 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3303 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3304 ofp_print_match(results, &match, true);
3306 ofp_print_actions(results, &rule->actions->header, act_len);
3308 ds_put_cstr(results, "drop");
3310 ds_put_cstr(results, "\n");
3313 /* Adds a pretty-printed description of all flows to 'results', including
3314 * those marked hidden by secchan (e.g., by in-band control). */
3316 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3318 struct ofp_match match;
3319 struct cls_rule target;
3320 struct flow_stats_ds_cbdata cbdata;
3322 memset(&match, 0, sizeof match);
3323 match.wildcards = htonl(OVSFW_ALL);
3326 cbdata.results = results;
3328 cls_rule_from_match(&match, 0, false, 0, &target);
3329 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3330 flow_stats_ds_cb, &cbdata);
3333 struct aggregate_stats_cbdata {
3334 struct ofproto *ofproto;
3336 uint64_t packet_count;
3337 uint64_t byte_count;
3342 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3344 struct rule *rule = rule_from_cls_rule(rule_);
3345 struct aggregate_stats_cbdata *cbdata = cbdata_;
3346 uint64_t packet_count, byte_count;
3348 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3352 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3354 cbdata->packet_count += packet_count;
3355 cbdata->byte_count += byte_count;
3360 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3361 const struct ofp_stats_request *osr,
3364 struct ofp_aggregate_stats_request *asr;
3365 struct ofp_aggregate_stats_reply *reply;
3366 struct aggregate_stats_cbdata cbdata;
3367 struct cls_rule target;
3370 if (arg_size != sizeof *asr) {
3371 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3373 asr = (struct ofp_aggregate_stats_request *) osr->body;
3375 COVERAGE_INC(ofproto_agg_request);
3377 cbdata.out_port = asr->out_port;
3378 cbdata.packet_count = 0;
3379 cbdata.byte_count = 0;
3381 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3382 classifier_for_each_match(&p->cls, &target,
3383 table_id_to_include(asr->table_id),
3384 aggregate_stats_cb, &cbdata);
3386 msg = start_stats_reply(osr, sizeof *reply);
3387 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3388 reply->flow_count = htonl(cbdata.n_flows);
3389 reply->packet_count = htonll(cbdata.packet_count);
3390 reply->byte_count = htonll(cbdata.byte_count);
3391 queue_tx(msg, ofconn, ofconn->reply_counter);
3395 struct queue_stats_cbdata {
3396 struct ofconn *ofconn;
3397 struct ofport *ofport;
3402 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3403 const struct netdev_queue_stats *stats)
3405 struct ofp_queue_stats *reply;
3407 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3408 reply->port_no = htons(cbdata->ofport->opp.port_no);
3409 memset(reply->pad, 0, sizeof reply->pad);
3410 reply->queue_id = htonl(queue_id);
3411 reply->tx_bytes = htonll(stats->tx_bytes);
3412 reply->tx_packets = htonll(stats->tx_packets);
3413 reply->tx_errors = htonll(stats->tx_errors);
3417 handle_queue_stats_dump_cb(uint32_t queue_id,
3418 struct netdev_queue_stats *stats,
3421 struct queue_stats_cbdata *cbdata = cbdata_;
3423 put_queue_stats(cbdata, queue_id, stats);
3427 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3428 struct queue_stats_cbdata *cbdata)
3430 cbdata->ofport = port;
3431 if (queue_id == OFPQ_ALL) {
3432 netdev_dump_queue_stats(port->netdev,
3433 handle_queue_stats_dump_cb, cbdata);
3435 struct netdev_queue_stats stats;
3437 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3438 put_queue_stats(cbdata, queue_id, &stats);
3444 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3445 const struct ofp_stats_request *osr,
3448 struct ofp_queue_stats_request *qsr;
3449 struct queue_stats_cbdata cbdata;
3450 struct ofport *port;
3451 unsigned int port_no;
3454 if (arg_size != sizeof *qsr) {
3455 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3457 qsr = (struct ofp_queue_stats_request *) osr->body;
3459 COVERAGE_INC(ofproto_queue_req);
3461 cbdata.ofconn = ofconn;
3462 cbdata.msg = start_stats_reply(osr, 128);
3464 port_no = ntohs(qsr->port_no);
3465 queue_id = ntohl(qsr->queue_id);
3466 if (port_no == OFPP_ALL) {
3467 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3468 handle_queue_stats_for_port(port, queue_id, &cbdata);
3470 } else if (port_no < ofproto->max_ports) {
3471 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3473 handle_queue_stats_for_port(port, queue_id, &cbdata);
3476 ofpbuf_delete(cbdata.msg);
3477 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3479 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3485 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3486 struct ofp_header *oh)
3488 struct ofp_stats_request *osr;
3492 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3497 osr = (struct ofp_stats_request *) oh;
3499 switch (ntohs(osr->type)) {
3501 return handle_desc_stats_request(p, ofconn, osr);
3504 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3506 case OFPST_AGGREGATE:
3507 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3510 return handle_table_stats_request(p, ofconn, osr);
3513 return handle_port_stats_request(p, ofconn, osr, arg_size);
3516 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3519 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3522 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3526 static long long int
3527 msec_from_nsec(uint64_t sec, uint32_t nsec)
3529 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3533 update_time(struct ofproto *ofproto, struct rule *rule,
3534 const struct odp_flow_stats *stats)
3536 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3537 if (used > rule->used) {
3539 if (rule->super && used > rule->super->used) {
3540 rule->super->used = used;
3542 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3547 update_stats(struct ofproto *ofproto, struct rule *rule,
3548 const struct odp_flow_stats *stats)
3550 if (stats->n_packets) {
3551 update_time(ofproto, rule, stats);
3552 rule->packet_count += stats->n_packets;
3553 rule->byte_count += stats->n_bytes;
3554 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3558 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3559 * in which no matching flow already exists in the flow table.
3561 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3562 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3563 * code as encoded by ofp_mkerr() on failure.
3565 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3568 add_flow(struct ofproto *p, struct ofconn *ofconn,
3569 const struct ofp_flow_mod *ofm, size_t n_actions)
3571 struct ofpbuf *packet;
3576 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3579 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3580 p->tun_id_from_cookie, ofm->cookie, &cr);
3581 if (classifier_rule_overlaps(&p->cls, &cr)) {
3582 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3586 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3587 n_actions, ntohs(ofm->idle_timeout),
3588 ntohs(ofm->hard_timeout), ofm->cookie,
3589 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3590 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3591 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3594 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3595 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3599 in_port = UINT16_MAX;
3602 rule_insert(p, rule, packet, in_port);
3606 static struct rule *
3607 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3609 struct cls_rule target;
3611 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3612 p->tun_id_from_cookie, ofm->cookie, &target);
3613 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &target));
3617 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3618 struct rule *rule, const struct ofp_flow_mod *ofm)
3620 struct ofpbuf *packet;
3625 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3629 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3635 flow_extract(packet, 0, in_port, &flow);
3636 rule_execute(ofproto, rule, packet, &flow);
3641 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3643 struct modify_flows_cbdata {
3644 struct ofproto *ofproto;
3645 const struct ofp_flow_mod *ofm;
3650 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3651 size_t n_actions, struct rule *);
3652 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3654 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3655 * encoded by ofp_mkerr() on failure.
3657 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3660 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3661 const struct ofp_flow_mod *ofm, size_t n_actions)
3663 struct modify_flows_cbdata cbdata;
3664 struct cls_rule target;
3668 cbdata.n_actions = n_actions;
3669 cbdata.match = NULL;
3671 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3674 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3675 modify_flows_cb, &cbdata);
3677 /* This credits the packet to whichever flow happened to happened to
3678 * match last. That's weird. Maybe we should do a lookup for the
3679 * flow that actually matches the packet? Who knows. */
3680 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3683 return add_flow(p, ofconn, ofm, n_actions);
3687 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3688 * code as encoded by ofp_mkerr() on failure.
3690 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3693 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3694 struct ofp_flow_mod *ofm, size_t n_actions)
3696 struct rule *rule = find_flow_strict(p, ofm);
3697 if (rule && !rule_is_hidden(rule)) {
3698 modify_flow(p, ofm, n_actions, rule);
3699 return send_buffered_packet(p, ofconn, rule, ofm);
3701 return add_flow(p, ofconn, ofm, n_actions);
3705 /* Callback for modify_flows_loose(). */
3707 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3709 struct rule *rule = rule_from_cls_rule(rule_);
3710 struct modify_flows_cbdata *cbdata = cbdata_;
3712 if (!rule_is_hidden(rule)) {
3713 cbdata->match = rule;
3714 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3718 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3719 * been identified as a flow in 'p''s flow table to be modified, by changing
3720 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3721 * ofp_action[] structures). */
3723 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3724 size_t n_actions, struct rule *rule)
3726 size_t actions_len = n_actions * sizeof *rule->actions;
3728 rule->flow_cookie = ofm->cookie;
3730 /* If the actions are the same, do nothing. */
3731 if (n_actions == rule->n_actions
3732 && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len)))
3737 /* Replace actions. */
3738 free(rule->actions);
3739 rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL;
3740 rule->n_actions = n_actions;
3742 /* Make sure that the datapath gets updated properly. */
3743 if (rule->cr.wc.wildcards) {
3744 COVERAGE_INC(ofproto_mod_wc_flow);
3745 p->need_revalidate = true;
3747 rule_update_actions(p, rule);
3753 /* OFPFC_DELETE implementation. */
3755 struct delete_flows_cbdata {
3756 struct ofproto *ofproto;
3760 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3761 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3763 /* Implements OFPFC_DELETE. */
3765 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3767 struct delete_flows_cbdata cbdata;
3768 struct cls_rule target;
3771 cbdata.out_port = ofm->out_port;
3773 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3776 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3777 delete_flows_cb, &cbdata);
3780 /* Implements OFPFC_DELETE_STRICT. */
3782 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3784 struct rule *rule = find_flow_strict(p, ofm);
3786 delete_flow(p, rule, ofm->out_port);
3790 /* Callback for delete_flows_loose(). */
3792 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3794 struct rule *rule = rule_from_cls_rule(rule_);
3795 struct delete_flows_cbdata *cbdata = cbdata_;
3797 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3800 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3801 * been identified as a flow to delete from 'p''s flow table, by deleting the
3802 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3805 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3806 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3807 * specified 'out_port'. */
3809 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3811 if (rule_is_hidden(rule)) {
3815 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3819 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3820 rule_remove(p, rule);
3824 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3825 struct ofp_flow_mod *ofm)
3827 struct ofp_match orig_match;
3831 error = reject_slave_controller(ofconn, &ofm->header);
3835 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3836 sizeof *ofm->actions, &n_actions);
3841 /* We do not support the emergency flow cache. It will hopefully
3842 * get dropped from OpenFlow in the near future. */
3843 if (ofm->flags & htons(OFPFF_EMERG)) {
3844 /* There isn't a good fit for an error code, so just state that the
3845 * flow table is full. */
3846 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3849 /* Normalize ofp->match. If normalization actually changes anything, then
3850 * log the differences. */
3851 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3852 orig_match = ofm->match;
3853 normalize_match(&ofm->match);
3854 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3855 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3856 if (!VLOG_DROP_INFO(&normal_rl)) {
3857 char *old = ofp_match_to_literal_string(&orig_match);
3858 char *new = ofp_match_to_literal_string(&ofm->match);
3859 VLOG_INFO("%s: normalization changed ofp_match, details:",
3860 rconn_get_name(ofconn->rconn));
3861 VLOG_INFO(" pre: %s", old);
3862 VLOG_INFO("post: %s", new);
3868 if (!ofm->match.wildcards) {
3869 ofm->priority = htons(UINT16_MAX);
3872 error = validate_actions((const union ofp_action *) ofm->actions,
3873 n_actions, p->max_ports);
3878 switch (ntohs(ofm->command)) {
3880 return add_flow(p, ofconn, ofm, n_actions);
3883 return modify_flows_loose(p, ofconn, ofm, n_actions);
3885 case OFPFC_MODIFY_STRICT:
3886 return modify_flow_strict(p, ofconn, ofm, n_actions);
3889 delete_flows_loose(p, ofm);
3892 case OFPFC_DELETE_STRICT:
3893 delete_flow_strict(p, ofm);
3897 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3902 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3906 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3911 p->tun_id_from_cookie = !!msg->set;
3916 handle_role_request(struct ofproto *ofproto,
3917 struct ofconn *ofconn, struct nicira_header *msg)
3919 struct nx_role_request *nrr;
3920 struct nx_role_request *reply;
3924 if (ntohs(msg->header.length) != sizeof *nrr) {
3925 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3926 ntohs(msg->header.length), sizeof *nrr);
3927 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3929 nrr = (struct nx_role_request *) msg;
3931 if (ofconn->type != OFCONN_PRIMARY) {
3932 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3934 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3937 role = ntohl(nrr->role);
3938 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3939 && role != NX_ROLE_SLAVE) {
3940 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3942 /* There's no good error code for this. */
3943 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3946 if (role == NX_ROLE_MASTER) {
3947 struct ofconn *other;
3949 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
3950 if (other->role == NX_ROLE_MASTER) {
3951 other->role = NX_ROLE_SLAVE;
3955 ofconn->role = role;
3957 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3959 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3960 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3961 reply->role = htonl(role);
3962 queue_tx(buf, ofconn, ofconn->reply_counter);
3968 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3970 struct ofp_vendor_header *ovh = msg;
3971 struct nicira_header *nh;
3973 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3974 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3975 "(expected at least %zu)",
3976 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3977 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3979 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3980 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3982 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3983 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3984 "(expected at least %zu)",
3985 ntohs(ovh->header.length), sizeof(struct nicira_header));
3986 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3990 switch (ntohl(nh->subtype)) {
3991 case NXT_STATUS_REQUEST:
3992 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3995 case NXT_TUN_ID_FROM_COOKIE:
3996 return handle_tun_id_from_cookie(p, msg);
3998 case NXT_ROLE_REQUEST:
3999 return handle_role_request(p, ofconn, msg);
4002 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4006 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4008 struct ofp_header *ob;
4011 /* Currently, everything executes synchronously, so we can just
4012 * immediately send the barrier reply. */
4013 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4014 queue_tx(buf, ofconn, ofconn->reply_counter);
4019 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4020 struct ofpbuf *ofp_msg)
4022 struct ofp_header *oh = ofp_msg->data;
4025 COVERAGE_INC(ofproto_recv_openflow);
4027 case OFPT_ECHO_REQUEST:
4028 error = handle_echo_request(ofconn, oh);
4031 case OFPT_ECHO_REPLY:
4035 case OFPT_FEATURES_REQUEST:
4036 error = handle_features_request(p, ofconn, oh);
4039 case OFPT_GET_CONFIG_REQUEST:
4040 error = handle_get_config_request(p, ofconn, oh);
4043 case OFPT_SET_CONFIG:
4044 error = handle_set_config(p, ofconn, ofp_msg->data);
4047 case OFPT_PACKET_OUT:
4048 error = handle_packet_out(p, ofconn, ofp_msg->data);
4052 error = handle_port_mod(p, ofconn, oh);
4056 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4059 case OFPT_STATS_REQUEST:
4060 error = handle_stats_request(p, ofconn, oh);
4064 error = handle_vendor(p, ofconn, ofp_msg->data);
4067 case OFPT_BARRIER_REQUEST:
4068 error = handle_barrier_request(ofconn, oh);
4072 if (VLOG_IS_WARN_ENABLED()) {
4073 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4074 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4077 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4082 send_error_oh(ofconn, ofp_msg->data, error);
4087 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4089 struct odp_msg *msg = packet->data;
4091 struct ofpbuf payload;
4094 payload.data = msg + 1;
4095 payload.size = msg->length - sizeof *msg;
4096 flow_extract(&payload, msg->arg, msg->port, &flow);
4098 /* Check with in-band control to see if this packet should be sent
4099 * to the local port regardless of the flow table. */
4100 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4101 union odp_action action;
4103 memset(&action, 0, sizeof(action));
4104 action.output.type = ODPAT_OUTPUT;
4105 action.output.port = ODPP_LOCAL;
4106 dpif_execute(p->dpif, &action, 1, &payload);
4109 rule = lookup_valid_rule(p, &flow);
4111 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4112 struct ofport *port = get_port(p, msg->port);
4114 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4115 COVERAGE_INC(ofproto_no_packet_in);
4116 /* XXX install 'drop' flow entry */
4117 ofpbuf_delete(packet);
4121 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4124 COVERAGE_INC(ofproto_packet_in);
4125 send_packet_in(p, packet);
4129 if (rule->cr.wc.wildcards) {
4130 rule = rule_create_subrule(p, rule, &flow);
4131 rule_make_actions(p, rule, packet);
4133 if (!rule->may_install) {
4134 /* The rule is not installable, that is, we need to process every
4135 * packet, so process the current packet and set its actions into
4137 rule_make_actions(p, rule, packet);
4139 /* XXX revalidate rule if it needs it */
4143 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4145 * Extra-special case for fail-open mode.
4147 * We are in fail-open mode and the packet matched the fail-open rule,
4148 * but we are connected to a controller too. We should send the packet
4149 * up to the controller in the hope that it will try to set up a flow
4150 * and thereby allow us to exit fail-open.
4152 * See the top-level comment in fail-open.c for more information.
4154 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4155 DPIF_RECV_MSG_PADDING));
4158 ofpbuf_pull(packet, sizeof *msg);
4159 rule_execute(p, rule, packet, &flow);
4160 rule_reinstall(p, rule);
4164 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4166 struct odp_msg *msg = packet->data;
4168 switch (msg->type) {
4169 case _ODPL_ACTION_NR:
4170 COVERAGE_INC(ofproto_ctlr_action);
4171 send_packet_in(p, packet);
4174 case _ODPL_SFLOW_NR:
4176 ofproto_sflow_received(p->sflow, msg);
4178 ofpbuf_delete(packet);
4182 handle_odp_miss_msg(p, packet);
4186 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4192 /* Flow expiration. */
4194 struct expire_cbdata {
4195 struct ofproto *ofproto;
4199 static int ofproto_dp_max_idle(const struct ofproto *);
4200 static void ofproto_update_used(struct ofproto *);
4201 static void rule_expire(struct cls_rule *, void *cbdata);
4203 /* This function is called periodically by ofproto_run(). Its job is to
4204 * collect updates for the flows that have been installed into the datapath,
4205 * most importantly when they last were used, and then use that information to
4206 * expire flows that have not been used recently.
4208 * Returns the number of milliseconds after which it should be called again. */
4210 ofproto_expire(struct ofproto *ofproto)
4212 struct expire_cbdata cbdata;
4214 /* Update 'used' for each flow in the datapath. */
4215 ofproto_update_used(ofproto);
4217 /* Expire idle flows.
4219 * A wildcarded flow is idle only when all of its subrules have expired due
4220 * to becoming idle, so iterate through the exact-match flows first. */
4221 cbdata.ofproto = ofproto;
4222 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4223 classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
4224 classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
4226 /* Let the hook know that we're at a stable point: all outstanding data
4227 * in existing flows has been accounted to the account_cb. Thus, the
4228 * hook can now reasonably do operations that depend on having accurate
4229 * flow volume accounting (currently, that's just bond rebalancing). */
4230 if (ofproto->ofhooks->account_checkpoint_cb) {
4231 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4234 return MIN(cbdata.dp_max_idle, 1000);
4237 /* Update 'used' member of each flow currently installed into the datapath. */
4239 ofproto_update_used(struct ofproto *p)
4241 struct odp_flow *flows;
4246 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4251 for (i = 0; i < n_flows; i++) {
4252 struct odp_flow *f = &flows[i];
4253 struct cls_rule target;
4257 odp_flow_key_to_flow(&f->key, &flow);
4258 cls_rule_from_flow(&flow, 0, UINT16_MAX, &target);
4260 rule = rule_from_cls_rule(classifier_find_rule_exactly(&p->cls,
4263 if (rule && rule->installed) {
4264 update_time(p, rule, &f->stats);
4265 rule_account(p, rule, f->stats.n_bytes);
4267 /* There's a flow in the datapath that we know nothing about.
4269 COVERAGE_INC(ofproto_unexpected_rule);
4270 dpif_flow_del(p->dpif, f);
4277 /* Calculates and returns the number of milliseconds of idle time after which
4278 * flows should expire from the datapath and we should fold their statistics
4279 * into their parent rules in userspace. */
4281 ofproto_dp_max_idle(const struct ofproto *ofproto)
4284 * Idle time histogram.
4286 * Most of the time a switch has a relatively small number of flows. When
4287 * this is the case we might as well keep statistics for all of them in
4288 * userspace and to cache them in the kernel datapath for performance as
4291 * As the number of flows increases, the memory required to maintain
4292 * statistics about them in userspace and in the kernel becomes
4293 * significant. However, with a large number of flows it is likely that
4294 * only a few of them are "heavy hitters" that consume a large amount of
4295 * bandwidth. At this point, only heavy hitters are worth caching in the
4296 * kernel and maintaining in userspaces; other flows we can discard.
4298 * The technique used to compute the idle time is to build a histogram with
4299 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4300 * is installed in the kernel gets dropped in the appropriate bucket.
4301 * After the histogram has been built, we compute the cutoff so that only
4302 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4303 * cached. At least the most-recently-used bucket of flows is kept, so
4304 * actually an arbitrary number of flows can be kept in any given
4305 * expiration run (though the next run will delete most of those unless
4306 * they receive additional data).
4308 * This requires a second pass through the exact-match flows, in addition
4309 * to the pass made by ofproto_update_used(), because the former function
4310 * never looks at uninstallable flows.
4312 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4313 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4314 int buckets[N_BUCKETS] = { 0 };
4320 total = classifier_count_exact(&ofproto->cls);
4321 if (total <= 1000) {
4322 return N_BUCKETS * BUCKET_WIDTH;
4325 /* Build histogram. */
4327 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) {
4328 long long int idle = now - rule->used;
4329 int bucket = (idle <= 0 ? 0
4330 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4331 : (unsigned int) idle / BUCKET_WIDTH);
4335 /* Find the first bucket whose flows should be expired. */
4336 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4337 if (buckets[bucket]) {
4340 subtotal += buckets[bucket++];
4341 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4346 if (VLOG_IS_DBG_ENABLED()) {
4350 ds_put_cstr(&s, "keep");
4351 for (i = 0; i < N_BUCKETS; i++) {
4353 ds_put_cstr(&s, ", drop");
4356 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4359 VLOG_INFO("%s: %s (msec:count)",
4360 dpif_name(ofproto->dpif), ds_cstr(&s));
4364 return bucket * BUCKET_WIDTH;
4368 rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
4370 if (ofproto->netflow && !is_controller_rule(rule) &&
4371 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4372 struct ofexpired expired;
4373 struct odp_flow odp_flow;
4375 /* Get updated flow stats.
4377 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4378 * updated TCP flags and (2) the dpif_flow_list_all() in
4379 * ofproto_update_used() zeroed TCP flags. */
4380 memset(&odp_flow, 0, sizeof odp_flow);
4381 if (rule->installed) {
4382 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
4383 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4384 dpif_flow_get(ofproto->dpif, &odp_flow);
4386 if (odp_flow.stats.n_packets) {
4387 update_time(ofproto, rule, &odp_flow.stats);
4388 netflow_flow_update_flags(&rule->nf_flow,
4389 odp_flow.stats.tcp_flags);
4393 expired.flow = rule->cr.flow;
4394 expired.packet_count = rule->packet_count +
4395 odp_flow.stats.n_packets;
4396 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4397 expired.used = rule->used;
4399 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4403 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4404 * rules, then delete it entirely.
4406 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4407 * the datapath and fold its statistics back into its super-rule.
4409 * (This is a callback function for classifier_for_each().) */
4411 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4413 struct expire_cbdata *cbdata = cbdata_;
4414 struct ofproto *ofproto = cbdata->ofproto;
4415 struct rule *rule = rule_from_cls_rule(cls_rule);
4416 long long int hard_expire, idle_expire, expire, now;
4418 /* Calculate OpenFlow expiration times for 'rule'. */
4419 hard_expire = (rule->hard_timeout
4420 ? rule->created + rule->hard_timeout * 1000
4422 idle_expire = (rule->idle_timeout
4423 && (rule->super || list_is_empty(&rule->list))
4424 ? rule->used + rule->idle_timeout * 1000
4426 expire = MIN(hard_expire, idle_expire);
4430 /* 'rule' has not expired according to OpenFlow rules. */
4431 if (!rule->cr.wc.wildcards) {
4432 if (now >= rule->used + cbdata->dp_max_idle) {
4433 /* This rule is idle, so drop it to free up resources. */
4435 /* It's not part of the OpenFlow flow table, so we can
4436 * delete it entirely and fold its statistics into its
4438 rule_remove(ofproto, rule);
4440 /* It is part of the OpenFlow flow table, so we have to
4441 * keep the rule but we can at least uninstall it from the
4443 rule_uninstall(ofproto, rule);
4446 /* Send NetFlow active timeout if appropriate. */
4447 rule_active_timeout(cbdata->ofproto, rule);
4451 /* 'rule' has expired according to OpenFlow rules. */
4452 COVERAGE_INC(ofproto_expired);
4454 /* Update stats. (This is a no-op if the rule expired due to an idle
4455 * timeout, because that only happens when the rule has no subrules
4457 if (rule->cr.wc.wildcards) {
4458 struct rule *subrule, *next;
4459 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
4460 rule_remove(cbdata->ofproto, subrule);
4463 rule_uninstall(cbdata->ofproto, rule);
4466 /* Get rid of the rule. */
4467 if (!rule_is_hidden(rule)) {
4468 send_flow_removed(cbdata->ofproto, rule, now,
4470 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4472 rule_remove(cbdata->ofproto, rule);
4477 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4479 struct rule *sub = rule_from_cls_rule(sub_);
4480 struct revalidate_cbdata *cbdata = cbdata_;
4482 if (cbdata->revalidate_all
4483 || (cbdata->revalidate_subrules && sub->super)
4484 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4485 revalidate_rule(cbdata->ofproto, sub);
4490 revalidate_rule(struct ofproto *p, struct rule *rule)
4492 const struct flow *flow = &rule->cr.flow;
4494 COVERAGE_INC(ofproto_revalidate_rule);
4497 super = rule_from_cls_rule(classifier_lookup(&p->cls, flow,
4500 rule_remove(p, rule);
4502 } else if (super != rule->super) {
4503 COVERAGE_INC(ofproto_revalidate_moved);
4504 list_remove(&rule->list);
4505 list_push_back(&super->list, &rule->list);
4506 rule->super = super;
4507 rule->hard_timeout = super->hard_timeout;
4508 rule->idle_timeout = super->idle_timeout;
4509 rule->created = super->created;
4514 rule_update_actions(p, rule);
4518 static struct ofpbuf *
4519 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4520 long long int now, uint8_t reason)
4522 struct ofp_flow_removed *ofr;
4524 long long int tdiff = now - rule->created;
4525 uint32_t sec = tdiff / 1000;
4526 uint32_t msec = tdiff - (sec * 1000);
4528 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4529 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4531 ofr->cookie = rule->flow_cookie;
4532 ofr->priority = htons(rule->cr.priority);
4533 ofr->reason = reason;
4534 ofr->duration_sec = htonl(sec);
4535 ofr->duration_nsec = htonl(msec * 1000000);
4536 ofr->idle_timeout = htons(rule->idle_timeout);
4537 ofr->packet_count = htonll(rule->packet_count);
4538 ofr->byte_count = htonll(rule->byte_count);
4544 send_flow_removed(struct ofproto *p, struct rule *rule,
4545 long long int now, uint8_t reason)
4547 struct ofconn *ofconn;
4548 struct ofconn *prev;
4549 struct ofpbuf *buf = NULL;
4551 if (!rule->send_flow_removed) {
4555 /* We limit the maximum number of queued flow expirations it by accounting
4556 * them under the counter for replies. That works because preventing
4557 * OpenFlow requests from being processed also prevents new flows from
4558 * being added (and expiring). (It also prevents processing OpenFlow
4559 * requests that would not add new flows, so it is imperfect.) */
4562 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4563 if (rconn_is_connected(ofconn->rconn)
4564 && ofconn_receives_async_msgs(ofconn)) {
4566 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4568 buf = compose_flow_removed(p, rule, now, reason);
4574 queue_tx(buf, prev, prev->reply_counter);
4578 /* pinsched callback for sending 'packet' on 'ofconn'. */
4580 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4582 struct ofconn *ofconn = ofconn_;
4584 rconn_send_with_limit(ofconn->rconn, packet,
4585 ofconn->packet_in_counter, 100);
4588 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4589 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4590 * packet scheduler for sending.
4592 * 'max_len' specifies the maximum number of bytes of the packet to send on
4593 * 'ofconn' (INT_MAX specifies no limit).
4595 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4596 * ownership is transferred to this function. */
4598 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4601 struct ofproto *ofproto = ofconn->ofproto;
4602 struct ofp_packet_in *opi = packet->data;
4603 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4604 int send_len, trim_size;
4608 if (opi->reason == OFPR_ACTION) {
4609 buffer_id = UINT32_MAX;
4610 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4611 buffer_id = pktbuf_get_null();
4612 } else if (!ofconn->pktbuf) {
4613 buffer_id = UINT32_MAX;
4615 struct ofpbuf payload;
4616 payload.data = opi->data;
4617 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4618 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4621 /* Figure out how much of the packet to send. */
4622 send_len = ntohs(opi->total_len);
4623 if (buffer_id != UINT32_MAX) {
4624 send_len = MIN(send_len, ofconn->miss_send_len);
4626 send_len = MIN(send_len, max_len);
4628 /* Adjust packet length and clone if necessary. */
4629 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4631 packet = ofpbuf_clone_data(packet->data, trim_size);
4634 packet->size = trim_size;
4637 /* Update packet headers. */
4638 opi->buffer_id = htonl(buffer_id);
4639 update_openflow_length(packet);
4641 /* Hand over to packet scheduler. It might immediately call into
4642 * do_send_packet_in() or it might buffer it for a while (until a later
4643 * call to pinsched_run()). */
4644 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4645 packet, do_send_packet_in, ofconn);
4648 /* Replace struct odp_msg header in 'packet' by equivalent struct
4649 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4650 * returned by dpif_recv()).
4652 * The conversion is not complete: the caller still needs to trim any unneeded
4653 * payload off the end of the buffer, set the length in the OpenFlow header,
4654 * and set buffer_id. Those require us to know the controller settings and so
4655 * must be done on a per-controller basis.
4657 * Returns the maximum number of bytes of the packet that should be sent to
4658 * the controller (INT_MAX if no limit). */
4660 do_convert_to_packet_in(struct ofpbuf *packet)
4662 struct odp_msg *msg = packet->data;
4663 struct ofp_packet_in *opi;
4669 /* Extract relevant header fields */
4670 if (msg->type == _ODPL_ACTION_NR) {
4671 reason = OFPR_ACTION;
4674 reason = OFPR_NO_MATCH;
4677 total_len = msg->length - sizeof *msg;
4678 in_port = odp_port_to_ofp_port(msg->port);
4680 /* Repurpose packet buffer by overwriting header. */
4681 ofpbuf_pull(packet, sizeof(struct odp_msg));
4682 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4683 opi->header.version = OFP_VERSION;
4684 opi->header.type = OFPT_PACKET_IN;
4685 opi->total_len = htons(total_len);
4686 opi->in_port = htons(in_port);
4687 opi->reason = reason;
4692 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4693 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4694 * as necessary according to their individual configurations.
4696 * 'packet' must have sufficient headroom to convert it into a struct
4697 * ofp_packet_in (e.g. as returned by dpif_recv()).
4699 * Takes ownership of 'packet'. */
4701 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4703 struct ofconn *ofconn, *prev;
4706 max_len = do_convert_to_packet_in(packet);
4709 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4710 if (ofconn_receives_async_msgs(ofconn)) {
4712 schedule_packet_in(prev, packet, max_len, true);
4718 schedule_packet_in(prev, packet, max_len, false);
4720 ofpbuf_delete(packet);
4725 pick_datapath_id(const struct ofproto *ofproto)
4727 const struct ofport *port;
4729 port = get_port(ofproto, ODPP_LOCAL);
4731 uint8_t ea[ETH_ADDR_LEN];
4734 error = netdev_get_etheraddr(port->netdev, ea);
4736 return eth_addr_to_uint64(ea);
4738 VLOG_WARN("could not get MAC address for %s (%s)",
4739 netdev_get_name(port->netdev), strerror(error));
4741 return ofproto->fallback_dpid;
4745 pick_fallback_dpid(void)
4747 uint8_t ea[ETH_ADDR_LEN];
4748 eth_addr_nicira_random(ea);
4749 return eth_addr_to_uint64(ea);
4753 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4754 struct odp_actions *actions, tag_type *tags,
4755 uint16_t *nf_output_iface, void *ofproto_)
4757 struct ofproto *ofproto = ofproto_;
4760 /* Drop frames for reserved multicast addresses. */
4761 if (eth_addr_is_reserved(flow->dl_dst)) {
4765 /* Learn source MAC (but don't try to learn from revalidation). */
4766 if (packet != NULL) {
4767 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4769 GRAT_ARP_LOCK_NONE);
4771 /* The log messages here could actually be useful in debugging,
4772 * so keep the rate limit relatively high. */
4773 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4774 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4775 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4776 ofproto_revalidate(ofproto, rev_tag);
4780 /* Determine output port. */
4781 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4784 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4785 nf_output_iface, actions);
4786 } else if (out_port != flow->in_port) {
4787 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4788 *nf_output_iface = out_port;
4796 static const struct ofhooks default_ofhooks = {
4797 default_normal_ofhook_cb,