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"
42 #include "ofp-print.h"
44 #include "ofproto-sflow.h"
46 #include "openflow/nicira-ext.h"
47 #include "openflow/openflow.h"
48 #include "openvswitch/datapath-protocol.h"
52 #include "poll-loop.h"
56 #include "stream-ssl.h"
64 VLOG_DEFINE_THIS_MODULE(ofproto);
66 COVERAGE_DEFINE(odp_overflow);
67 COVERAGE_DEFINE(ofproto_add_wc_flow);
68 COVERAGE_DEFINE(ofproto_agg_request);
69 COVERAGE_DEFINE(ofproto_costly_flags);
70 COVERAGE_DEFINE(ofproto_ctlr_action);
71 COVERAGE_DEFINE(ofproto_del_wc_flow);
72 COVERAGE_DEFINE(ofproto_dp_missed);
73 COVERAGE_DEFINE(ofproto_error);
74 COVERAGE_DEFINE(ofproto_expiration);
75 COVERAGE_DEFINE(ofproto_expired);
76 COVERAGE_DEFINE(ofproto_flows_req);
77 COVERAGE_DEFINE(ofproto_flush);
78 COVERAGE_DEFINE(ofproto_invalidated);
79 COVERAGE_DEFINE(ofproto_mod_wc_flow);
80 COVERAGE_DEFINE(ofproto_no_packet_in);
81 COVERAGE_DEFINE(ofproto_odp_unchanged);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_revalidate);
90 COVERAGE_DEFINE(ofproto_revalidate_moved);
91 COVERAGE_DEFINE(ofproto_revalidate_rule);
92 COVERAGE_DEFINE(ofproto_subrule_create);
93 COVERAGE_DEFINE(ofproto_unexpected_rule);
94 COVERAGE_DEFINE(ofproto_uninstallable);
95 COVERAGE_DEFINE(ofproto_update_port);
97 #include "sflow_api.h"
100 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
101 struct netdev *netdev;
102 struct ofp_phy_port opp; /* In host byte order. */
106 static void ofport_free(struct ofport *);
107 static void hton_ofp_phy_port(struct ofp_phy_port *);
109 static int xlate_actions(const union ofp_action *in, size_t n_in,
110 const struct flow *, struct ofproto *,
111 const struct ofpbuf *packet,
112 struct odp_actions *out, tag_type *tags,
113 bool *may_set_up_flow, uint16_t *nf_output_iface);
115 /* An OpenFlow flow. */
117 long long int used; /* Time last used; time created if not used. */
118 long long int created; /* Creation time. */
122 * - Do include packets and bytes from facets that have been deleted or
123 * whose own statistics have been folded into the rule.
125 * - Do include packets and bytes sent "by hand" that were accounted to
126 * the rule without any facet being involved (this is a rare corner
127 * case in rule_execute()).
129 * - Do not include packet or bytes that can be obtained from any facet's
130 * packet_count or byte_count member or that can be obtained from the
131 * datapath by, e.g., dpif_flow_get() for any facet.
133 uint64_t packet_count; /* Number of packets received. */
134 uint64_t byte_count; /* Number of bytes received. */
136 ovs_be64 flow_cookie; /* Controller-issued identifier. */
138 struct cls_rule cr; /* In owning ofproto's classifier. */
139 uint16_t idle_timeout; /* In seconds from time of last use. */
140 uint16_t hard_timeout; /* In seconds from time of creation. */
141 bool send_flow_removed; /* Send a flow removed message? */
142 int n_actions; /* Number of elements in actions[]. */
143 union ofp_action *actions; /* OpenFlow actions. */
144 struct list facets; /* List of "struct facet"s. */
147 static struct rule *rule_from_cls_rule(const struct cls_rule *);
148 static bool rule_is_hidden(const struct rule *);
150 static struct rule *rule_create(const struct cls_rule *,
151 const union ofp_action *, size_t n_actions,
152 uint16_t idle_timeout, uint16_t hard_timeout,
153 ovs_be64 flow_cookie, bool send_flow_removed);
154 static void rule_destroy(struct ofproto *, struct rule *);
155 static void rule_free(struct rule *);
157 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
158 static void rule_insert(struct ofproto *, struct rule *);
159 static void rule_remove(struct ofproto *, struct rule *);
161 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
163 /* An exact-match instantiation of an OpenFlow flow. */
165 long long int used; /* Time last used; time created if not used. */
169 * - Do include packets and bytes sent "by hand", e.g. with
172 * - Do include packets and bytes that were obtained from the datapath
173 * when a flow was deleted (e.g. dpif_flow_del()) or when its
174 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
176 * - Do not include any packets or bytes that can currently be obtained
177 * from the datapath by, e.g., dpif_flow_get().
179 uint64_t packet_count; /* Number of packets received. */
180 uint64_t byte_count; /* Number of bytes received. */
182 /* Number of bytes passed to account_cb. This may include bytes that can
183 * currently obtained from the datapath (thus, it can be greater than
185 uint64_t accounted_bytes;
187 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
188 struct list list_node; /* In owning rule's 'facets' list. */
189 struct rule *rule; /* Owning rule. */
190 struct flow flow; /* Exact-match flow. */
191 bool installed; /* Installed in datapath? */
192 bool may_install; /* True ordinarily; false if actions must
193 * be reassessed for every packet. */
194 int n_actions; /* Number of elements in actions[]. */
195 union odp_action *actions; /* Datapath actions. */
196 tag_type tags; /* Tags (set only by hooks). */
197 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
200 static struct facet *facet_create(struct ofproto *, struct rule *,
202 const struct ofpbuf *packet);
203 static void facet_remove(struct ofproto *, struct facet *);
204 static void facet_free(struct facet *);
206 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
207 static bool facet_revalidate(struct ofproto *, struct facet *);
209 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
210 static void facet_uninstall(struct ofproto *, struct facet *);
211 static void facet_flush_stats(struct ofproto *, struct facet *);
213 static void facet_make_actions(struct ofproto *, struct facet *,
214 const struct ofpbuf *packet);
215 static void facet_update_stats(struct ofproto *, struct facet *,
216 const struct odp_flow_stats *);
218 /* ofproto supports two kinds of OpenFlow connections:
220 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
221 * maintains persistent connections to these controllers and by default
222 * sends them asynchronous messages such as packet-ins.
224 * - "Service" connections, e.g. from ovs-ofctl. When these connections
225 * drop, it is the other side's responsibility to reconnect them if
226 * necessary. ofproto does not send them asynchronous messages by default.
228 * Currently, active (tcp, ssl, unix) connections are always "primary"
229 * connections and passive (ptcp, pssl, punix) connections are always "service"
230 * connections. There is no inherent reason for this, but it reflects the
234 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
235 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
238 /* A listener for incoming OpenFlow "service" connections. */
240 struct hmap_node node; /* In struct ofproto's "services" hmap. */
241 struct pvconn *pvconn; /* OpenFlow connection listener. */
243 /* These are not used by ofservice directly. They are settings for
244 * accepted "struct ofconn"s from the pvconn. */
245 int probe_interval; /* Max idle time before probing, in seconds. */
246 int rate_limit; /* Max packet-in rate in packets per second. */
247 int burst_limit; /* Limit on accumulating packet credits. */
250 static struct ofservice *ofservice_lookup(struct ofproto *,
252 static int ofservice_create(struct ofproto *,
253 const struct ofproto_controller *);
254 static void ofservice_reconfigure(struct ofservice *,
255 const struct ofproto_controller *);
256 static void ofservice_destroy(struct ofproto *, struct ofservice *);
258 /* An OpenFlow connection. */
260 struct ofproto *ofproto; /* The ofproto that owns this connection. */
261 struct list node; /* In struct ofproto's "all_conns" list. */
262 struct rconn *rconn; /* OpenFlow connection. */
263 enum ofconn_type type; /* Type. */
264 int flow_format; /* One of NXFF_*. */
266 /* OFPT_PACKET_IN related data. */
267 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
268 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
269 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
270 int miss_send_len; /* Bytes to send of buffered packets. */
272 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
273 * requests, and the maximum number before we stop reading OpenFlow
275 #define OFCONN_REPLY_MAX 100
276 struct rconn_packet_counter *reply_counter;
278 /* type == OFCONN_PRIMARY only. */
279 enum nx_role role; /* Role. */
280 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
281 struct discovery *discovery; /* Controller discovery object, if enabled. */
282 struct status_category *ss; /* Switch status category. */
283 enum ofproto_band band; /* In-band or out-of-band? */
286 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
287 * "schedulers" array. Their values are 0 and 1, and their meanings and values
288 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
289 * case anything ever changes, check their values here. */
290 #define N_SCHEDULERS 2
291 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
292 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
293 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
294 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
296 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
298 static void ofconn_destroy(struct ofconn *);
299 static void ofconn_run(struct ofconn *);
300 static void ofconn_wait(struct ofconn *);
301 static bool ofconn_receives_async_msgs(const struct ofconn *);
302 static char *ofconn_make_name(const struct ofproto *, const char *target);
303 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
305 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
306 struct rconn_packet_counter *counter);
308 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
309 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
313 uint64_t datapath_id; /* Datapath ID. */
314 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
315 char *mfr_desc; /* Manufacturer. */
316 char *hw_desc; /* Hardware. */
317 char *sw_desc; /* Software version. */
318 char *serial_desc; /* Serial number. */
319 char *dp_desc; /* Datapath description. */
323 struct netdev_monitor *netdev_monitor;
324 struct hmap ports; /* Contains "struct ofport"s. */
325 struct shash port_by_name;
329 struct switch_status *switch_status;
330 struct fail_open *fail_open;
331 struct netflow *netflow;
332 struct ofproto_sflow *sflow;
334 /* In-band control. */
335 struct in_band *in_band;
336 long long int next_in_band_update;
337 struct sockaddr_in *extra_in_band_remotes;
338 size_t n_extra_remotes;
342 struct classifier cls;
343 long long int next_expiration;
347 bool need_revalidate;
348 struct tag_set revalidate_set;
350 /* OpenFlow connections. */
351 struct hmap controllers; /* Controller "struct ofconn"s. */
352 struct list all_conns; /* Contains "struct ofconn"s. */
353 enum ofproto_fail_mode fail_mode;
355 /* OpenFlow listeners. */
356 struct hmap services; /* Contains "struct ofservice"s. */
357 struct pvconn **snoops;
360 /* Hooks for ovs-vswitchd. */
361 const struct ofhooks *ofhooks;
364 /* Used by default ofhooks. */
365 struct mac_learning *ml;
368 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
370 static const struct ofhooks default_ofhooks;
372 static uint64_t pick_datapath_id(const struct ofproto *);
373 static uint64_t pick_fallback_dpid(void);
375 static int ofproto_expire(struct ofproto *);
377 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
379 static void handle_openflow(struct ofconn *, struct ofpbuf *);
381 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
382 static void update_port(struct ofproto *, const char *devname);
383 static int init_ports(struct ofproto *);
384 static void reinit_ports(struct ofproto *);
387 ofproto_create(const char *datapath, const char *datapath_type,
388 const struct ofhooks *ofhooks, void *aux,
389 struct ofproto **ofprotop)
391 struct odp_stats stats;
398 /* Connect to datapath and start listening for messages. */
399 error = dpif_open(datapath, datapath_type, &dpif);
401 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
404 error = dpif_get_dp_stats(dpif, &stats);
406 VLOG_ERR("failed to obtain stats for datapath %s: %s",
407 datapath, strerror(error));
411 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
413 VLOG_ERR("failed to listen on datapath %s: %s",
414 datapath, strerror(error));
418 dpif_flow_flush(dpif);
419 dpif_recv_purge(dpif);
421 /* Initialize settings. */
422 p = xzalloc(sizeof *p);
423 p->fallback_dpid = pick_fallback_dpid();
424 p->datapath_id = p->fallback_dpid;
425 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
426 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
427 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
428 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
429 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
431 /* Initialize datapath. */
433 p->netdev_monitor = netdev_monitor_create();
434 hmap_init(&p->ports);
435 shash_init(&p->port_by_name);
436 p->max_ports = stats.max_ports;
438 /* Initialize submodules. */
439 p->switch_status = switch_status_create(p);
444 /* Initialize in-band control. */
446 p->in_band_queue = -1;
448 /* Initialize flow table. */
449 classifier_init(&p->cls);
450 p->next_expiration = time_msec() + 1000;
452 /* Initialize facet table. */
453 hmap_init(&p->facets);
454 p->need_revalidate = false;
455 tag_set_init(&p->revalidate_set);
457 /* Initialize OpenFlow connections. */
458 list_init(&p->all_conns);
459 hmap_init(&p->controllers);
460 hmap_init(&p->services);
464 /* Initialize hooks. */
466 p->ofhooks = ofhooks;
470 p->ofhooks = &default_ofhooks;
472 p->ml = mac_learning_create();
475 /* Pick final datapath ID. */
476 p->datapath_id = pick_datapath_id(p);
477 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
484 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
486 uint64_t old_dpid = p->datapath_id;
487 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
488 if (p->datapath_id != old_dpid) {
489 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
491 /* Force all active connections to reconnect, since there is no way to
492 * notify a controller that the datapath ID has changed. */
493 ofproto_reconnect_controllers(p);
498 is_discovery_controller(const struct ofproto_controller *c)
500 return !strcmp(c->target, "discover");
504 is_in_band_controller(const struct ofproto_controller *c)
506 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
509 /* Creates a new controller in 'ofproto'. Some of the settings are initially
510 * drawn from 'c', but update_controller() needs to be called later to finish
511 * the new ofconn's configuration. */
513 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
515 struct discovery *discovery;
516 struct ofconn *ofconn;
518 if (is_discovery_controller(c)) {
519 int error = discovery_create(c->accept_re, c->update_resolv_conf,
520 ofproto->dpif, ofproto->switch_status,
529 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
530 ofconn->pktbuf = pktbuf_create();
531 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
533 ofconn->discovery = discovery;
535 char *name = ofconn_make_name(ofproto, c->target);
536 rconn_connect(ofconn->rconn, c->target, name);
539 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
540 hash_string(c->target, 0));
543 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
544 * target or turn discovery on or off (these are done by creating new ofconns
545 * and deleting old ones), but it can update the rest of an ofconn's
548 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
552 ofconn->band = (is_in_band_controller(c)
553 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
555 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
557 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
558 rconn_set_probe_interval(ofconn->rconn, probe_interval);
560 if (ofconn->discovery) {
561 discovery_set_update_resolv_conf(ofconn->discovery,
562 c->update_resolv_conf);
563 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
566 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
570 ofconn_get_target(const struct ofconn *ofconn)
572 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
575 static struct ofconn *
576 find_controller_by_target(struct ofproto *ofproto, const char *target)
578 struct ofconn *ofconn;
580 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
581 hash_string(target, 0), &ofproto->controllers) {
582 if (!strcmp(ofconn_get_target(ofconn), target)) {
590 update_in_band_remotes(struct ofproto *ofproto)
592 const struct ofconn *ofconn;
593 struct sockaddr_in *addrs;
594 size_t max_addrs, n_addrs;
598 /* Allocate enough memory for as many remotes as we could possibly have. */
599 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
600 addrs = xmalloc(max_addrs * sizeof *addrs);
603 /* Add all the remotes. */
605 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
606 struct sockaddr_in *sin = &addrs[n_addrs];
608 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
612 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
613 if (sin->sin_addr.s_addr) {
614 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
617 if (ofconn->discovery) {
621 for (i = 0; i < ofproto->n_extra_remotes; i++) {
622 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
625 /* Create or update or destroy in-band.
627 * Ordinarily we only enable in-band if there's at least one remote
628 * address, but discovery needs the in-band rules for DHCP to be installed
629 * even before we know any remote addresses. */
630 if (n_addrs || discovery) {
631 if (!ofproto->in_band) {
632 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
635 if (ofproto->in_band) {
636 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
638 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
639 ofproto->next_in_band_update = time_msec() + 1000;
641 in_band_destroy(ofproto->in_band);
642 ofproto->in_band = NULL;
650 update_fail_open(struct ofproto *p)
652 struct ofconn *ofconn;
654 if (!hmap_is_empty(&p->controllers)
655 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
656 struct rconn **rconns;
660 p->fail_open = fail_open_create(p, p->switch_status);
664 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
665 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
666 rconns[n++] = ofconn->rconn;
669 fail_open_set_controllers(p->fail_open, rconns, n);
670 /* p->fail_open takes ownership of 'rconns'. */
672 fail_open_destroy(p->fail_open);
678 ofproto_set_controllers(struct ofproto *p,
679 const struct ofproto_controller *controllers,
680 size_t n_controllers)
682 struct shash new_controllers;
683 struct ofconn *ofconn, *next_ofconn;
684 struct ofservice *ofservice, *next_ofservice;
688 /* Create newly configured controllers and services.
689 * Create a name to ofproto_controller mapping in 'new_controllers'. */
690 shash_init(&new_controllers);
691 for (i = 0; i < n_controllers; i++) {
692 const struct ofproto_controller *c = &controllers[i];
694 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
695 if (!find_controller_by_target(p, c->target)) {
696 add_controller(p, c);
698 } else if (!pvconn_verify_name(c->target)) {
699 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
703 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
704 dpif_name(p->dpif), c->target);
708 shash_add_once(&new_controllers, c->target, &controllers[i]);
711 /* Delete controllers that are no longer configured.
712 * Update configuration of all now-existing controllers. */
714 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
715 struct ofproto_controller *c;
717 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
719 ofconn_destroy(ofconn);
721 update_controller(ofconn, c);
728 /* Delete services that are no longer configured.
729 * Update configuration of all now-existing services. */
730 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
731 struct ofproto_controller *c;
733 c = shash_find_data(&new_controllers,
734 pvconn_get_name(ofservice->pvconn));
736 ofservice_destroy(p, ofservice);
738 ofservice_reconfigure(ofservice, c);
742 shash_destroy(&new_controllers);
744 update_in_band_remotes(p);
747 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
748 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
749 struct ofconn, hmap_node);
750 ofconn->ss = switch_status_register(p->switch_status, "remote",
751 rconn_status_cb, ofconn->rconn);
756 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
758 p->fail_mode = fail_mode;
762 /* Drops the connections between 'ofproto' and all of its controllers, forcing
763 * them to reconnect. */
765 ofproto_reconnect_controllers(struct ofproto *ofproto)
767 struct ofconn *ofconn;
769 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
770 rconn_reconnect(ofconn->rconn);
775 any_extras_changed(const struct ofproto *ofproto,
776 const struct sockaddr_in *extras, size_t n)
780 if (n != ofproto->n_extra_remotes) {
784 for (i = 0; i < n; i++) {
785 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
786 const struct sockaddr_in *new = &extras[i];
788 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
789 old->sin_port != new->sin_port) {
797 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
798 * in-band control should guarantee access, in the same way that in-band
799 * control guarantees access to OpenFlow controllers. */
801 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
802 const struct sockaddr_in *extras, size_t n)
804 if (!any_extras_changed(ofproto, extras, n)) {
808 free(ofproto->extra_in_band_remotes);
809 ofproto->n_extra_remotes = n;
810 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
812 update_in_band_remotes(ofproto);
815 /* Sets the OpenFlow queue used by flows set up by in-band control on
816 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
817 * flows will use the default queue. */
819 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
821 if (queue_id != ofproto->in_band_queue) {
822 ofproto->in_band_queue = queue_id;
823 update_in_band_remotes(ofproto);
828 ofproto_set_desc(struct ofproto *p,
829 const char *mfr_desc, const char *hw_desc,
830 const char *sw_desc, const char *serial_desc,
833 struct ofp_desc_stats *ods;
836 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
837 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
838 sizeof ods->mfr_desc);
841 p->mfr_desc = xstrdup(mfr_desc);
844 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
845 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
846 sizeof ods->hw_desc);
849 p->hw_desc = xstrdup(hw_desc);
852 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
853 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
854 sizeof ods->sw_desc);
857 p->sw_desc = xstrdup(sw_desc);
860 if (strlen(serial_desc) >= sizeof ods->serial_num) {
861 VLOG_WARN("truncating serial_desc, must be less than %zu "
863 sizeof ods->serial_num);
865 free(p->serial_desc);
866 p->serial_desc = xstrdup(serial_desc);
869 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
870 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
871 sizeof ods->dp_desc);
874 p->dp_desc = xstrdup(dp_desc);
879 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
880 const struct svec *svec)
882 struct pvconn **pvconns = *pvconnsp;
883 size_t n_pvconns = *n_pvconnsp;
887 for (i = 0; i < n_pvconns; i++) {
888 pvconn_close(pvconns[i]);
892 pvconns = xmalloc(svec->n * sizeof *pvconns);
894 for (i = 0; i < svec->n; i++) {
895 const char *name = svec->names[i];
896 struct pvconn *pvconn;
899 error = pvconn_open(name, &pvconn);
901 pvconns[n_pvconns++] = pvconn;
903 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
911 *n_pvconnsp = n_pvconns;
917 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
919 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
923 ofproto_set_netflow(struct ofproto *ofproto,
924 const struct netflow_options *nf_options)
926 if (nf_options && nf_options->collectors.n) {
927 if (!ofproto->netflow) {
928 ofproto->netflow = netflow_create();
930 return netflow_set_options(ofproto->netflow, nf_options);
932 netflow_destroy(ofproto->netflow);
933 ofproto->netflow = NULL;
939 ofproto_set_sflow(struct ofproto *ofproto,
940 const struct ofproto_sflow_options *oso)
942 struct ofproto_sflow *os = ofproto->sflow;
945 struct ofport *ofport;
947 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
948 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
949 ofproto_sflow_add_port(os, ofport->odp_port,
950 netdev_get_name(ofport->netdev));
953 ofproto_sflow_set_options(os, oso);
955 ofproto_sflow_destroy(os);
956 ofproto->sflow = NULL;
961 ofproto_get_datapath_id(const struct ofproto *ofproto)
963 return ofproto->datapath_id;
967 ofproto_has_primary_controller(const struct ofproto *ofproto)
969 return !hmap_is_empty(&ofproto->controllers);
972 enum ofproto_fail_mode
973 ofproto_get_fail_mode(const struct ofproto *p)
979 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
983 for (i = 0; i < ofproto->n_snoops; i++) {
984 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
989 ofproto_destroy(struct ofproto *p)
991 struct ofservice *ofservice, *next_ofservice;
992 struct ofconn *ofconn, *next_ofconn;
993 struct ofport *ofport, *next_ofport;
1000 /* Destroy fail-open and in-band early, since they touch the classifier. */
1001 fail_open_destroy(p->fail_open);
1002 p->fail_open = NULL;
1004 in_band_destroy(p->in_band);
1006 free(p->extra_in_band_remotes);
1008 ofproto_flush_flows(p);
1009 classifier_destroy(&p->cls);
1010 hmap_destroy(&p->facets);
1012 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1013 ofconn_destroy(ofconn);
1015 hmap_destroy(&p->controllers);
1017 dpif_close(p->dpif);
1018 netdev_monitor_destroy(p->netdev_monitor);
1019 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1020 hmap_remove(&p->ports, &ofport->hmap_node);
1021 ofport_free(ofport);
1023 shash_destroy(&p->port_by_name);
1025 switch_status_destroy(p->switch_status);
1026 netflow_destroy(p->netflow);
1027 ofproto_sflow_destroy(p->sflow);
1029 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1030 ofservice_destroy(p, ofservice);
1032 hmap_destroy(&p->services);
1034 for (i = 0; i < p->n_snoops; i++) {
1035 pvconn_close(p->snoops[i]);
1039 mac_learning_destroy(p->ml);
1044 free(p->serial_desc);
1047 hmap_destroy(&p->ports);
1053 ofproto_run(struct ofproto *p)
1055 int error = ofproto_run1(p);
1057 error = ofproto_run2(p, false);
1063 process_port_change(struct ofproto *ofproto, int error, char *devname)
1065 if (error == ENOBUFS) {
1066 reinit_ports(ofproto);
1067 } else if (!error) {
1068 update_port(ofproto, devname);
1073 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1074 * means that 'ofconn' is more interesting for monitoring than a lower return
1077 snoop_preference(const struct ofconn *ofconn)
1079 switch (ofconn->role) {
1080 case NX_ROLE_MASTER:
1087 /* Shouldn't happen. */
1092 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1093 * Connects this vconn to a controller. */
1095 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1097 struct ofconn *ofconn, *best;
1099 /* Pick a controller for monitoring. */
1101 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1102 if (ofconn->type == OFCONN_PRIMARY
1103 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1109 rconn_add_monitor(best->rconn, vconn);
1111 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1117 ofproto_run1(struct ofproto *p)
1119 struct ofconn *ofconn, *next_ofconn;
1120 struct ofservice *ofservice;
1125 if (shash_is_empty(&p->port_by_name)) {
1129 for (i = 0; i < 50; i++) {
1132 error = dpif_recv(p->dpif, &buf);
1134 if (error == ENODEV) {
1135 /* Someone destroyed the datapath behind our back. The caller
1136 * better destroy us and give up, because we're just going to
1137 * spin from here on out. */
1138 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1139 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1140 dpif_name(p->dpif));
1146 handle_odp_msg(p, buf);
1149 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1150 process_port_change(p, error, devname);
1152 while ((error = netdev_monitor_poll(p->netdev_monitor,
1153 &devname)) != EAGAIN) {
1154 process_port_change(p, error, devname);
1158 if (time_msec() >= p->next_in_band_update) {
1159 update_in_band_remotes(p);
1161 in_band_run(p->in_band);
1164 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1168 /* Fail-open maintenance. Do this after processing the ofconns since
1169 * fail-open checks the status of the controller rconn. */
1171 fail_open_run(p->fail_open);
1174 HMAP_FOR_EACH (ofservice, node, &p->services) {
1175 struct vconn *vconn;
1178 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1180 struct rconn *rconn;
1183 rconn = rconn_create(ofservice->probe_interval, 0);
1184 name = ofconn_make_name(p, vconn_get_name(vconn));
1185 rconn_connect_unreliably(rconn, vconn, name);
1188 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1189 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1190 ofservice->burst_limit);
1191 } else if (retval != EAGAIN) {
1192 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1196 for (i = 0; i < p->n_snoops; i++) {
1197 struct vconn *vconn;
1200 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1202 add_snooper(p, vconn);
1203 } else if (retval != EAGAIN) {
1204 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1208 if (time_msec() >= p->next_expiration) {
1209 int delay = ofproto_expire(p);
1210 p->next_expiration = time_msec() + delay;
1211 COVERAGE_INC(ofproto_expiration);
1215 netflow_run(p->netflow);
1218 ofproto_sflow_run(p->sflow);
1225 ofproto_run2(struct ofproto *p, bool revalidate_all)
1227 /* Figure out what we need to revalidate now, if anything. */
1228 struct tag_set revalidate_set = p->revalidate_set;
1229 if (p->need_revalidate) {
1230 revalidate_all = true;
1233 /* Clear the revalidation flags. */
1234 tag_set_init(&p->revalidate_set);
1235 p->need_revalidate = false;
1237 /* Now revalidate if there's anything to do. */
1238 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1239 struct facet *facet, *next;
1241 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1243 || tag_set_intersects(&revalidate_set, facet->tags)) {
1244 facet_revalidate(p, facet);
1253 ofproto_wait(struct ofproto *p)
1255 struct ofservice *ofservice;
1256 struct ofconn *ofconn;
1259 dpif_recv_wait(p->dpif);
1260 dpif_port_poll_wait(p->dpif);
1261 netdev_monitor_poll_wait(p->netdev_monitor);
1262 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1263 ofconn_wait(ofconn);
1266 poll_timer_wait_until(p->next_in_band_update);
1267 in_band_wait(p->in_band);
1270 fail_open_wait(p->fail_open);
1273 ofproto_sflow_wait(p->sflow);
1275 if (!tag_set_is_empty(&p->revalidate_set)) {
1276 poll_immediate_wake();
1278 if (p->need_revalidate) {
1279 /* Shouldn't happen, but if it does just go around again. */
1280 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1281 poll_immediate_wake();
1282 } else if (p->next_expiration != LLONG_MAX) {
1283 poll_timer_wait_until(p->next_expiration);
1285 HMAP_FOR_EACH (ofservice, node, &p->services) {
1286 pvconn_wait(ofservice->pvconn);
1288 for (i = 0; i < p->n_snoops; i++) {
1289 pvconn_wait(p->snoops[i]);
1294 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1296 tag_set_add(&ofproto->revalidate_set, tag);
1300 ofproto_get_revalidate_set(struct ofproto *ofproto)
1302 return &ofproto->revalidate_set;
1306 ofproto_is_alive(const struct ofproto *p)
1308 return !hmap_is_empty(&p->controllers);
1311 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1313 * This is almost the same as calling dpif_port_del() directly on the
1314 * datapath, but it also makes 'ofproto' close its open netdev for the port
1315 * (if any). This makes it possible to create a new netdev of a different
1316 * type under the same name, which otherwise the netdev library would refuse
1317 * to do because of the conflict. (The netdev would eventually get closed on
1318 * the next trip through ofproto_run(), but this interface is more direct.)
1320 * Returns 0 if successful, otherwise a positive errno. */
1322 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1324 struct ofport *ofport = get_port(ofproto, odp_port);
1325 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1328 error = dpif_port_del(ofproto->dpif, odp_port);
1330 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1331 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1332 } else if (ofport) {
1333 /* 'name' is ofport->opp.name and update_port() is going to destroy
1334 * 'ofport'. Just in case update_port() refers to 'name' after it
1335 * destroys 'ofport', make a copy of it around the update_port()
1337 char *devname = xstrdup(name);
1338 update_port(ofproto, devname);
1344 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1345 * true if 'odp_port' exists and should be included, false otherwise. */
1347 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1349 struct ofport *ofport = get_port(ofproto, odp_port);
1350 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1354 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1355 const union ofp_action *actions, size_t n_actions,
1356 const struct ofpbuf *packet)
1358 struct odp_actions odp_actions;
1361 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1367 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1369 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
1373 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1374 * performs the 'n_actions' actions in 'actions'. The new flow will not
1377 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1378 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1379 * controllers; otherwise, it will be hidden.
1381 * The caller retains ownership of 'cls_rule' and 'actions'. */
1383 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1384 const union ofp_action *actions, size_t n_actions)
1387 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1388 rule_insert(p, rule);
1392 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1396 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1399 rule_remove(ofproto, rule);
1404 ofproto_flush_flows(struct ofproto *ofproto)
1406 struct facet *facet, *next_facet;
1407 struct rule *rule, *next_rule;
1408 struct cls_cursor cursor;
1410 COVERAGE_INC(ofproto_flush);
1412 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1413 /* Mark the facet as not installed so that facet_remove() doesn't
1414 * bother trying to uninstall it. There is no point in uninstalling it
1415 * individually since we are about to blow away all the facets with
1416 * dpif_flow_flush(). */
1417 facet->installed = false;
1418 facet_remove(ofproto, facet);
1421 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1422 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1423 rule_remove(ofproto, rule);
1426 dpif_flow_flush(ofproto->dpif);
1427 if (ofproto->in_band) {
1428 in_band_flushed(ofproto->in_band);
1430 if (ofproto->fail_open) {
1431 fail_open_flushed(ofproto->fail_open);
1436 reinit_ports(struct ofproto *p)
1438 struct svec devnames;
1439 struct ofport *ofport;
1440 struct odp_port *odp_ports;
1444 COVERAGE_INC(ofproto_reinit_ports);
1446 svec_init(&devnames);
1447 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1448 svec_add (&devnames, (char *) ofport->opp.name);
1450 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1451 for (i = 0; i < n_odp_ports; i++) {
1452 svec_add (&devnames, odp_ports[i].devname);
1456 svec_sort_unique(&devnames);
1457 for (i = 0; i < devnames.n; i++) {
1458 update_port(p, devnames.names[i]);
1460 svec_destroy(&devnames);
1463 static struct ofport *
1464 make_ofport(const struct odp_port *odp_port)
1466 struct netdev_options netdev_options;
1467 enum netdev_flags flags;
1468 struct ofport *ofport;
1469 struct netdev *netdev;
1472 memset(&netdev_options, 0, sizeof netdev_options);
1473 netdev_options.name = odp_port->devname;
1474 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1476 error = netdev_open(&netdev_options, &netdev);
1478 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1479 "cannot be opened (%s)",
1480 odp_port->devname, odp_port->port,
1481 odp_port->devname, strerror(error));
1485 ofport = xmalloc(sizeof *ofport);
1486 ofport->netdev = netdev;
1487 ofport->odp_port = odp_port->port;
1488 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1489 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1490 memcpy(ofport->opp.name, odp_port->devname,
1491 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1492 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1494 netdev_get_flags(netdev, &flags);
1495 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1497 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1499 netdev_get_features(netdev,
1500 &ofport->opp.curr, &ofport->opp.advertised,
1501 &ofport->opp.supported, &ofport->opp.peer);
1506 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1508 if (get_port(p, odp_port->port)) {
1509 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1512 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1513 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1522 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1524 const struct ofp_phy_port *a = &a_->opp;
1525 const struct ofp_phy_port *b = &b_->opp;
1527 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1528 return (a->port_no == b->port_no
1529 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1530 && !strcmp((char *) a->name, (char *) b->name)
1531 && a->state == b->state
1532 && a->config == b->config
1533 && a->curr == b->curr
1534 && a->advertised == b->advertised
1535 && a->supported == b->supported
1536 && a->peer == b->peer);
1540 send_port_status(struct ofproto *p, const struct ofport *ofport,
1543 /* XXX Should limit the number of queued port status change messages. */
1544 struct ofconn *ofconn;
1545 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1546 struct ofp_port_status *ops;
1549 /* Primary controllers, even slaves, should always get port status
1550 updates. Otherwise obey ofconn_receives_async_msgs(). */
1551 if (ofconn->type != OFCONN_PRIMARY
1552 && !ofconn_receives_async_msgs(ofconn)) {
1556 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1557 ops->reason = reason;
1558 ops->desc = ofport->opp;
1559 hton_ofp_phy_port(&ops->desc);
1560 queue_tx(b, ofconn, NULL);
1565 ofport_install(struct ofproto *p, struct ofport *ofport)
1567 const char *netdev_name = (const char *) ofport->opp.name;
1569 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1570 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1571 shash_add(&p->port_by_name, netdev_name, ofport);
1573 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1578 ofport_remove(struct ofproto *p, struct ofport *ofport)
1580 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1581 hmap_remove(&p->ports, &ofport->hmap_node);
1582 shash_delete(&p->port_by_name,
1583 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1585 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1590 ofport_free(struct ofport *ofport)
1593 netdev_close(ofport->netdev);
1598 static struct ofport *
1599 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1601 struct ofport *port;
1603 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1604 hash_int(odp_port, 0), &ofproto->ports) {
1605 if (port->odp_port == odp_port) {
1613 update_port(struct ofproto *p, const char *devname)
1615 struct odp_port odp_port;
1616 struct ofport *old_ofport;
1617 struct ofport *new_ofport;
1620 COVERAGE_INC(ofproto_update_port);
1622 /* Query the datapath for port information. */
1623 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1625 /* Find the old ofport. */
1626 old_ofport = shash_find_data(&p->port_by_name, devname);
1629 /* There's no port named 'devname' but there might be a port with
1630 * the same port number. This could happen if a port is deleted
1631 * and then a new one added in its place very quickly, or if a port
1632 * is renamed. In the former case we want to send an OFPPR_DELETE
1633 * and an OFPPR_ADD, and in the latter case we want to send a
1634 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1635 * the old port's ifindex against the new port, or perhaps less
1636 * reliably but more portably by comparing the old port's MAC
1637 * against the new port's MAC. However, this code isn't that smart
1638 * and always sends an OFPPR_MODIFY (XXX). */
1639 old_ofport = get_port(p, odp_port.port);
1641 } else if (error != ENOENT && error != ENODEV) {
1642 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1643 "%s", strerror(error));
1647 /* Create a new ofport. */
1648 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1650 /* Eliminate a few pathological cases. */
1651 if (!old_ofport && !new_ofport) {
1653 } else if (old_ofport && new_ofport) {
1654 /* Most of the 'config' bits are OpenFlow soft state, but
1655 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1656 * OpenFlow bits from old_ofport. (make_ofport() only sets
1657 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1658 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1660 if (ofport_equal(old_ofport, new_ofport)) {
1661 /* False alarm--no change. */
1662 ofport_free(new_ofport);
1667 /* Now deal with the normal cases. */
1669 ofport_remove(p, old_ofport);
1672 ofport_install(p, new_ofport);
1674 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1675 (!old_ofport ? OFPPR_ADD
1676 : !new_ofport ? OFPPR_DELETE
1678 ofport_free(old_ofport);
1682 init_ports(struct ofproto *p)
1684 struct odp_port *ports;
1689 error = dpif_port_list(p->dpif, &ports, &n_ports);
1694 for (i = 0; i < n_ports; i++) {
1695 const struct odp_port *odp_port = &ports[i];
1696 if (!ofport_conflicts(p, odp_port)) {
1697 struct ofport *ofport = make_ofport(odp_port);
1699 ofport_install(p, ofport);
1707 static struct ofconn *
1708 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1710 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1711 ofconn->ofproto = p;
1712 list_push_back(&p->all_conns, &ofconn->node);
1713 ofconn->rconn = rconn;
1714 ofconn->type = type;
1715 ofconn->flow_format = NXFF_OPENFLOW10;
1716 ofconn->role = NX_ROLE_OTHER;
1717 ofconn->packet_in_counter = rconn_packet_counter_create ();
1718 ofconn->pktbuf = NULL;
1719 ofconn->miss_send_len = 0;
1720 ofconn->reply_counter = rconn_packet_counter_create ();
1725 ofconn_destroy(struct ofconn *ofconn)
1727 if (ofconn->type == OFCONN_PRIMARY) {
1728 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1730 discovery_destroy(ofconn->discovery);
1732 list_remove(&ofconn->node);
1733 switch_status_unregister(ofconn->ss);
1734 rconn_destroy(ofconn->rconn);
1735 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1736 rconn_packet_counter_destroy(ofconn->reply_counter);
1737 pktbuf_destroy(ofconn->pktbuf);
1742 ofconn_run(struct ofconn *ofconn)
1744 struct ofproto *p = ofconn->ofproto;
1748 if (ofconn->discovery) {
1749 char *controller_name;
1750 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1751 discovery_question_connectivity(ofconn->discovery);
1753 if (discovery_run(ofconn->discovery, &controller_name)) {
1754 if (controller_name) {
1755 char *ofconn_name = ofconn_make_name(p, controller_name);
1756 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1759 rconn_disconnect(ofconn->rconn);
1764 for (i = 0; i < N_SCHEDULERS; i++) {
1765 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1768 rconn_run(ofconn->rconn);
1770 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1771 /* Limit the number of iterations to prevent other tasks from
1773 for (iteration = 0; iteration < 50; iteration++) {
1774 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1779 fail_open_maybe_recover(p->fail_open);
1781 handle_openflow(ofconn, of_msg);
1782 ofpbuf_delete(of_msg);
1786 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1787 ofconn_destroy(ofconn);
1792 ofconn_wait(struct ofconn *ofconn)
1796 if (ofconn->discovery) {
1797 discovery_wait(ofconn->discovery);
1799 for (i = 0; i < N_SCHEDULERS; i++) {
1800 pinsched_wait(ofconn->schedulers[i]);
1802 rconn_run_wait(ofconn->rconn);
1803 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1804 rconn_recv_wait(ofconn->rconn);
1806 COVERAGE_INC(ofproto_ofconn_stuck);
1810 /* Returns true if 'ofconn' should receive asynchronous messages. */
1812 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1814 if (ofconn->type == OFCONN_PRIMARY) {
1815 /* Primary controllers always get asynchronous messages unless they
1816 * have configured themselves as "slaves". */
1817 return ofconn->role != NX_ROLE_SLAVE;
1819 /* Service connections don't get asynchronous messages unless they have
1820 * explicitly asked for them by setting a nonzero miss send length. */
1821 return ofconn->miss_send_len > 0;
1825 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1826 * and 'target', suitable for use in log messages for identifying the
1829 * The name is dynamically allocated. The caller should free it (with free())
1830 * when it is no longer needed. */
1832 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1834 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1838 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1842 for (i = 0; i < N_SCHEDULERS; i++) {
1843 struct pinsched **s = &ofconn->schedulers[i];
1847 *s = pinsched_create(rate, burst,
1848 ofconn->ofproto->switch_status);
1850 pinsched_set_limits(*s, rate, burst);
1853 pinsched_destroy(*s);
1860 ofservice_reconfigure(struct ofservice *ofservice,
1861 const struct ofproto_controller *c)
1863 ofservice->probe_interval = c->probe_interval;
1864 ofservice->rate_limit = c->rate_limit;
1865 ofservice->burst_limit = c->burst_limit;
1868 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1869 * positive errno value. */
1871 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1873 struct ofservice *ofservice;
1874 struct pvconn *pvconn;
1877 error = pvconn_open(c->target, &pvconn);
1882 ofservice = xzalloc(sizeof *ofservice);
1883 hmap_insert(&ofproto->services, &ofservice->node,
1884 hash_string(c->target, 0));
1885 ofservice->pvconn = pvconn;
1887 ofservice_reconfigure(ofservice, c);
1893 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1895 hmap_remove(&ofproto->services, &ofservice->node);
1896 pvconn_close(ofservice->pvconn);
1900 /* Finds and returns the ofservice within 'ofproto' that has the given
1901 * 'target', or a null pointer if none exists. */
1902 static struct ofservice *
1903 ofservice_lookup(struct ofproto *ofproto, const char *target)
1905 struct ofservice *ofservice;
1907 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1908 &ofproto->services) {
1909 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1916 /* Returns true if 'rule' should be hidden from the controller.
1918 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1919 * (e.g. by in-band control) and are intentionally hidden from the
1922 rule_is_hidden(const struct rule *rule)
1924 return rule->cr.priority > UINT16_MAX;
1927 /* Creates and returns a new rule initialized as specified.
1929 * The caller is responsible for inserting the rule into the classifier (with
1930 * rule_insert()). */
1931 static struct rule *
1932 rule_create(const struct cls_rule *cls_rule,
1933 const union ofp_action *actions, size_t n_actions,
1934 uint16_t idle_timeout, uint16_t hard_timeout,
1935 ovs_be64 flow_cookie, bool send_flow_removed)
1937 struct rule *rule = xzalloc(sizeof *rule);
1938 rule->cr = *cls_rule;
1939 rule->idle_timeout = idle_timeout;
1940 rule->hard_timeout = hard_timeout;
1941 rule->flow_cookie = flow_cookie;
1942 rule->used = rule->created = time_msec();
1943 rule->send_flow_removed = send_flow_removed;
1944 list_init(&rule->facets);
1945 if (n_actions > 0) {
1946 rule->n_actions = n_actions;
1947 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1953 static struct rule *
1954 rule_from_cls_rule(const struct cls_rule *cls_rule)
1956 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1960 rule_free(struct rule *rule)
1962 free(rule->actions);
1966 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1967 * destroying any that no longer has a rule (which is probably all of them).
1969 * The caller must have already removed 'rule' from the classifier. */
1971 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1973 struct facet *facet, *next_facet;
1974 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1975 facet_revalidate(ofproto, facet);
1980 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1981 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1984 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1986 const union ofp_action *oa;
1987 struct actions_iterator i;
1989 if (out_port == htons(OFPP_NONE)) {
1992 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1993 oa = actions_next(&i)) {
1994 if (action_outputs_to_port(oa, out_port)) {
2001 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2002 * 'packet', which arrived on 'in_port'.
2004 * Takes ownership of 'packet'. */
2006 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
2007 const union odp_action *actions, size_t n_actions,
2008 struct ofpbuf *packet)
2010 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
2011 /* As an optimization, avoid a round-trip from userspace to kernel to
2012 * userspace. This also avoids possibly filling up kernel packet
2013 * buffers along the way. */
2014 struct odp_msg *msg;
2016 msg = ofpbuf_push_uninit(packet, sizeof *msg);
2017 msg->type = _ODPL_ACTION_NR;
2018 msg->length = sizeof(struct odp_msg) + packet->size;
2019 msg->port = in_port;
2021 msg->arg = actions[0].controller.arg;
2023 send_packet_in(ofproto, packet);
2029 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
2030 ofpbuf_delete(packet);
2035 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2036 * statistics appropriately. 'packet' must have at least sizeof(struct
2037 * ofp_packet_in) bytes of headroom.
2039 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2040 * applying flow_extract() to 'packet' would yield the same flow as
2043 * 'facet' must have accurately composed ODP actions; that is, it must not be
2044 * in need of revalidation.
2046 * Takes ownership of 'packet'. */
2048 facet_execute(struct ofproto *ofproto, struct facet *facet,
2049 struct ofpbuf *packet)
2051 struct odp_flow_stats stats;
2053 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2055 flow_extract_stats(&facet->flow, packet, &stats);
2056 if (execute_odp_actions(ofproto, facet->flow.in_port,
2057 facet->actions, facet->n_actions, packet)) {
2058 facet_update_stats(ofproto, facet, &stats);
2059 facet->used = time_msec();
2060 netflow_flow_update_time(ofproto->netflow,
2061 &facet->nf_flow, facet->used);
2065 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2066 * statistics (or the statistics for one of its facets) appropriately.
2067 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2069 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2070 * with statistics for 'packet' either way.
2072 * Takes ownership of 'packet'. */
2074 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2075 struct ofpbuf *packet)
2077 struct facet *facet;
2078 struct odp_actions a;
2082 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2084 flow_extract(packet, 0, in_port, &flow);
2086 /* First look for a related facet. If we find one, account it to that. */
2087 facet = facet_lookup_valid(ofproto, &flow);
2088 if (facet && facet->rule == rule) {
2089 facet_execute(ofproto, facet, packet);
2093 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2094 * create a new facet for it and use that. */
2095 if (rule_lookup(ofproto, &flow) == rule) {
2096 facet = facet_create(ofproto, rule, &flow, packet);
2097 facet_execute(ofproto, facet, packet);
2098 facet_install(ofproto, facet, true);
2102 /* We can't account anything to a facet. If we were to try, then that
2103 * facet would have a non-matching rule, busting our invariants. */
2104 if (xlate_actions(rule->actions, rule->n_actions, &flow, ofproto,
2105 packet, &a, NULL, 0, NULL)) {
2106 ofpbuf_delete(packet);
2109 size = packet->size;
2110 if (execute_odp_actions(ofproto, in_port,
2111 a.actions, a.n_actions, packet)) {
2112 rule->used = time_msec();
2113 rule->packet_count++;
2114 rule->byte_count += size;
2118 /* Inserts 'rule' into 'p''s flow table. */
2120 rule_insert(struct ofproto *p, struct rule *rule)
2122 struct rule *displaced_rule;
2124 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2125 if (displaced_rule) {
2126 rule_destroy(p, displaced_rule);
2128 p->need_revalidate = true;
2131 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2132 * 'flow' and an example 'packet' within that flow.
2134 * The caller must already have determined that no facet with an identical
2135 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2136 * 'ofproto''s classifier table. */
2137 static struct facet *
2138 facet_create(struct ofproto *ofproto, struct rule *rule,
2139 const struct flow *flow, const struct ofpbuf *packet)
2141 struct facet *facet;
2143 facet = xzalloc(sizeof *facet);
2144 facet->used = time_msec();
2145 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2146 list_push_back(&rule->facets, &facet->list_node);
2148 facet->flow = *flow;
2149 netflow_flow_init(&facet->nf_flow);
2150 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2152 facet_make_actions(ofproto, facet, packet);
2158 facet_free(struct facet *facet)
2160 free(facet->actions);
2164 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2166 * - Removes 'rule' from the classifier.
2168 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2169 * destroys them), via rule_destroy().
2172 rule_remove(struct ofproto *ofproto, struct rule *rule)
2174 COVERAGE_INC(ofproto_del_rule);
2175 ofproto->need_revalidate = true;
2176 classifier_remove(&ofproto->cls, &rule->cr);
2177 rule_destroy(ofproto, rule);
2180 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2182 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2183 * rule's statistics, via facet_uninstall().
2185 * - Removes 'facet' from its rule and from ofproto->facets.
2188 facet_remove(struct ofproto *ofproto, struct facet *facet)
2190 facet_uninstall(ofproto, facet);
2191 facet_flush_stats(ofproto, facet);
2192 hmap_remove(&ofproto->facets, &facet->hmap_node);
2193 list_remove(&facet->list_node);
2197 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2199 facet_make_actions(struct ofproto *p, struct facet *facet,
2200 const struct ofpbuf *packet)
2202 const struct rule *rule = facet->rule;
2203 struct odp_actions a;
2206 xlate_actions(rule->actions, rule->n_actions, &facet->flow, p,
2207 packet, &a, &facet->tags, &facet->may_install,
2208 &facet->nf_flow.output_iface);
2210 actions_len = a.n_actions * sizeof *a.actions;
2211 if (facet->n_actions != a.n_actions
2212 || memcmp(facet->actions, a.actions, actions_len)) {
2213 free(facet->actions);
2214 facet->n_actions = a.n_actions;
2215 facet->actions = xmemdup(a.actions, actions_len);
2220 facet_put__(struct ofproto *ofproto, struct facet *facet, int flags,
2221 struct odp_flow_put *put)
2223 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2224 odp_flow_key_from_flow(&put->flow.key, &facet->flow);
2225 put->flow.actions = facet->actions;
2226 put->flow.n_actions = facet->n_actions;
2227 put->flow.flags = 0;
2229 return dpif_flow_put(ofproto->dpif, put);
2232 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2233 * 'zero_stats' is true, clears any existing statistics from the datapath for
2236 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2238 if (facet->may_install) {
2239 struct odp_flow_put put;
2242 flags = ODPPF_CREATE | ODPPF_MODIFY;
2244 flags |= ODPPF_ZERO_STATS;
2246 if (!facet_put__(p, facet, flags, &put)) {
2247 facet->installed = true;
2252 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2253 * to the accounting hook function in the ofhooks structure. */
2255 facet_account(struct ofproto *ofproto,
2256 struct facet *facet, uint64_t extra_bytes)
2258 uint64_t total_bytes = facet->byte_count + extra_bytes;
2260 if (ofproto->ofhooks->account_flow_cb
2261 && total_bytes > facet->accounted_bytes)
2263 ofproto->ofhooks->account_flow_cb(
2264 &facet->flow, facet->tags, facet->actions, facet->n_actions,
2265 total_bytes - facet->accounted_bytes, ofproto->aux);
2266 facet->accounted_bytes = total_bytes;
2270 /* If 'rule' is installed in the datapath, uninstalls it. */
2272 facet_uninstall(struct ofproto *p, struct facet *facet)
2274 if (facet->installed) {
2275 struct odp_flow odp_flow;
2277 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
2278 odp_flow.actions = NULL;
2279 odp_flow.n_actions = 0;
2281 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2282 facet_update_stats(p, facet, &odp_flow.stats);
2284 facet->installed = false;
2288 /* Returns true if the only action for 'facet' is to send to the controller.
2289 * (We don't report NetFlow expiration messages for such facets because they
2290 * are just part of the control logic for the network, not real traffic). */
2292 facet_is_controller_flow(struct facet *facet)
2295 && facet->rule->n_actions == 1
2296 && action_outputs_to_port(&facet->rule->actions[0],
2297 htons(OFPP_CONTROLLER)));
2300 /* Folds all of 'facet''s statistics into its rule. Also updates the
2301 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2303 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2305 facet_account(ofproto, facet, 0);
2307 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2308 struct ofexpired expired;
2309 expired.flow = facet->flow;
2310 expired.packet_count = facet->packet_count;
2311 expired.byte_count = facet->byte_count;
2312 expired.used = facet->used;
2313 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2316 facet->rule->packet_count += facet->packet_count;
2317 facet->rule->byte_count += facet->byte_count;
2319 /* Reset counters to prevent double counting if 'facet' ever gets
2321 facet->packet_count = 0;
2322 facet->byte_count = 0;
2323 facet->accounted_bytes = 0;
2325 netflow_flow_clear(&facet->nf_flow);
2328 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2329 * Returns it if found, otherwise a null pointer.
2331 * The returned facet might need revalidation; use facet_lookup_valid()
2332 * instead if that is important. */
2333 static struct facet *
2334 facet_find(struct ofproto *ofproto, const struct flow *flow)
2336 struct facet *facet;
2338 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2340 if (flow_equal(flow, &facet->flow)) {
2348 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2349 * Returns it if found, otherwise a null pointer.
2351 * The returned facet is guaranteed to be valid. */
2352 static struct facet *
2353 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2355 struct facet *facet = facet_find(ofproto, flow);
2357 /* The facet we found might not be valid, since we could be in need of
2358 * revalidation. If it is not valid, don't return it. */
2360 && ofproto->need_revalidate
2361 && !facet_revalidate(ofproto, facet)) {
2362 COVERAGE_INC(ofproto_invalidated);
2369 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2371 * - If the rule found is different from 'facet''s current rule, moves
2372 * 'facet' to the new rule and recompiles its actions.
2374 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2375 * where it is and recompiles its actions anyway.
2377 * - If there is none, destroys 'facet'.
2379 * Returns true if 'facet' still exists, false if it has been destroyed. */
2381 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2383 struct rule *new_rule;
2384 struct odp_actions a;
2386 uint16_t new_nf_output_iface;
2387 bool actions_changed;
2389 COVERAGE_INC(facet_revalidate);
2391 /* Determine the new rule. */
2392 new_rule = rule_lookup(ofproto, &facet->flow);
2394 /* No new rule, so delete the facet. */
2395 facet_remove(ofproto, facet);
2399 /* Calculate new ODP actions.
2401 * We are very cautious about actually modifying 'facet' state at this
2402 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2403 * so, we need to have the old state around to properly compose it. */
2404 xlate_actions(new_rule->actions, new_rule->n_actions, &facet->flow,
2405 ofproto, NULL, &a, &facet->tags, &facet->may_install,
2406 &new_nf_output_iface);
2407 actions_len = a.n_actions * sizeof *a.actions;
2408 actions_changed = (facet->n_actions != a.n_actions
2409 || memcmp(facet->actions, a.actions, actions_len));
2411 /* If the ODP actions changed or the installability changed, then we need
2412 * to talk to the datapath. */
2413 if (actions_changed || facet->may_install != facet->installed) {
2414 if (facet->may_install) {
2415 struct odp_flow_put put;
2417 memset(&put.flow.stats, 0, sizeof put.flow.stats);
2418 odp_flow_key_from_flow(&put.flow.key, &facet->flow);
2419 put.flow.actions = a.actions;
2420 put.flow.n_actions = a.n_actions;
2422 put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS;
2423 dpif_flow_put(ofproto->dpif, &put);
2425 facet_update_stats(ofproto, facet, &put.flow.stats);
2427 facet_uninstall(ofproto, facet);
2430 /* The datapath flow is gone or has zeroed stats, so push stats out of
2431 * 'facet' into 'rule'. */
2432 facet_flush_stats(ofproto, facet);
2435 /* Update 'facet' now that we've taken care of all the old state. */
2436 facet->nf_flow.output_iface = new_nf_output_iface;
2437 if (actions_changed) {
2438 free(facet->actions);
2439 facet->n_actions = a.n_actions;
2440 facet->actions = xmemdup(a.actions, actions_len);
2442 if (facet->rule != new_rule) {
2443 COVERAGE_INC(facet_changed_rule);
2444 list_remove(&facet->list_node);
2445 list_push_back(&new_rule->facets, &facet->list_node);
2446 facet->rule = new_rule;
2447 facet->used = new_rule->created;
2454 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2455 struct rconn_packet_counter *counter)
2457 update_openflow_length(msg);
2458 if (rconn_send(ofconn->rconn, msg, counter)) {
2464 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2467 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2469 COVERAGE_INC(ofproto_error);
2470 queue_tx(buf, ofconn, ofconn->reply_counter);
2475 hton_ofp_phy_port(struct ofp_phy_port *opp)
2477 opp->port_no = htons(opp->port_no);
2478 opp->config = htonl(opp->config);
2479 opp->state = htonl(opp->state);
2480 opp->curr = htonl(opp->curr);
2481 opp->advertised = htonl(opp->advertised);
2482 opp->supported = htonl(opp->supported);
2483 opp->peer = htonl(opp->peer);
2487 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2489 struct ofp_header *rq = oh;
2490 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2495 handle_features_request(struct ofconn *ofconn, struct ofp_header *oh)
2497 struct ofp_switch_features *osf;
2499 struct ofport *port;
2501 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2502 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2503 osf->n_buffers = htonl(pktbuf_capacity());
2505 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2506 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2507 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2508 (1u << OFPAT_SET_VLAN_VID) |
2509 (1u << OFPAT_SET_VLAN_PCP) |
2510 (1u << OFPAT_STRIP_VLAN) |
2511 (1u << OFPAT_SET_DL_SRC) |
2512 (1u << OFPAT_SET_DL_DST) |
2513 (1u << OFPAT_SET_NW_SRC) |
2514 (1u << OFPAT_SET_NW_DST) |
2515 (1u << OFPAT_SET_NW_TOS) |
2516 (1u << OFPAT_SET_TP_SRC) |
2517 (1u << OFPAT_SET_TP_DST) |
2518 (1u << OFPAT_ENQUEUE));
2520 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2521 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2524 queue_tx(buf, ofconn, ofconn->reply_counter);
2529 handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh)
2532 struct ofp_switch_config *osc;
2536 /* Figure out flags. */
2537 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2538 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2541 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2542 osc->flags = htons(flags);
2543 osc->miss_send_len = htons(ofconn->miss_send_len);
2544 queue_tx(buf, ofconn, ofconn->reply_counter);
2550 handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc)
2555 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2559 flags = ntohs(osc->flags);
2561 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2562 switch (flags & OFPC_FRAG_MASK) {
2563 case OFPC_FRAG_NORMAL:
2564 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2566 case OFPC_FRAG_DROP:
2567 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2570 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2576 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2582 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2584 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2585 a->controller.arg = max_len;
2588 struct action_xlate_ctx {
2590 struct flow flow; /* Flow to which these actions correspond. */
2591 int recurse; /* Recursion level, via xlate_table_action. */
2592 struct ofproto *ofproto;
2593 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2594 * null pointer if we are revalidating
2595 * without a packet to refer to. */
2598 struct odp_actions *out; /* Datapath actions. */
2599 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
2600 bool may_set_up_flow; /* True ordinarily; false if the actions must
2601 * be reassessed for every packet. */
2602 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2605 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2606 * flow translation. */
2607 #define MAX_RESUBMIT_RECURSION 8
2609 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2610 struct action_xlate_ctx *ctx);
2613 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2615 const struct ofport *ofport = get_port(ctx->ofproto, port);
2618 if (ofport->opp.config & OFPPC_NO_FWD) {
2619 /* Forwarding disabled on port. */
2624 * We don't have an ofport record for this port, but it doesn't hurt to
2625 * allow forwarding to it anyhow. Maybe such a port will appear later
2626 * and we're pre-populating the flow table.
2630 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2631 ctx->nf_output_iface = port;
2634 static struct rule *
2635 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2637 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2641 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2643 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2644 uint16_t old_in_port;
2647 /* Look up a flow with 'in_port' as the input port. Then restore the
2648 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2649 * have surprising behavior). */
2650 old_in_port = ctx->flow.in_port;
2651 ctx->flow.in_port = in_port;
2652 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2653 ctx->flow.in_port = old_in_port;
2657 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2661 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2663 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2664 MAX_RESUBMIT_RECURSION);
2669 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2670 uint16_t *nf_output_iface, struct odp_actions *actions)
2672 struct ofport *ofport;
2674 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2675 uint16_t odp_port = ofport->odp_port;
2676 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2677 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2680 *nf_output_iface = NF_OUT_FLOOD;
2684 xlate_output_action__(struct action_xlate_ctx *ctx,
2685 uint16_t port, uint16_t max_len)
2688 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2690 ctx->nf_output_iface = NF_OUT_DROP;
2694 add_output_action(ctx, ctx->flow.in_port);
2697 xlate_table_action(ctx, ctx->flow.in_port);
2700 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2701 ctx->out, &ctx->tags,
2702 &ctx->nf_output_iface,
2703 ctx->ofproto->aux)) {
2704 COVERAGE_INC(ofproto_uninstallable);
2705 ctx->may_set_up_flow = false;
2709 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2710 &ctx->nf_output_iface, ctx->out);
2713 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2714 &ctx->nf_output_iface, ctx->out);
2716 case OFPP_CONTROLLER:
2717 add_controller_action(ctx->out, max_len);
2720 add_output_action(ctx, ODPP_LOCAL);
2723 odp_port = ofp_port_to_odp_port(port);
2724 if (odp_port != ctx->flow.in_port) {
2725 add_output_action(ctx, odp_port);
2730 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2731 ctx->nf_output_iface = NF_OUT_FLOOD;
2732 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2733 ctx->nf_output_iface = prev_nf_output_iface;
2734 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2735 ctx->nf_output_iface != NF_OUT_FLOOD) {
2736 ctx->nf_output_iface = NF_OUT_MULTI;
2741 xlate_output_action(struct action_xlate_ctx *ctx,
2742 const struct ofp_action_output *oao)
2744 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2747 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2748 * optimization, because we're going to add another action that sets the
2749 * priority immediately after, or because there are no actions following the
2752 remove_pop_action(struct action_xlate_ctx *ctx)
2754 size_t n = ctx->out->n_actions;
2755 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2756 ctx->out->n_actions--;
2761 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2762 const struct ofp_action_enqueue *oae)
2764 uint16_t ofp_port, odp_port;
2768 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2771 /* Fall back to ordinary output action. */
2772 xlate_output_action__(ctx, ntohs(oae->port), 0);
2776 /* Figure out ODP output port. */
2777 ofp_port = ntohs(oae->port);
2778 if (ofp_port != OFPP_IN_PORT) {
2779 odp_port = ofp_port_to_odp_port(ofp_port);
2781 odp_port = ctx->flow.in_port;
2784 /* Add ODP actions. */
2785 remove_pop_action(ctx);
2786 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2788 add_output_action(ctx, odp_port);
2789 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2791 /* Update NetFlow output port. */
2792 if (ctx->nf_output_iface == NF_OUT_DROP) {
2793 ctx->nf_output_iface = odp_port;
2794 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2795 ctx->nf_output_iface = NF_OUT_MULTI;
2800 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2801 const struct nx_action_set_queue *nasq)
2806 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2809 /* Couldn't translate queue to a priority, so ignore. A warning
2810 * has already been logged. */
2814 remove_pop_action(ctx);
2815 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2820 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2822 ovs_be16 tci = ctx->flow.vlan_tci;
2823 if (!(tci & htons(VLAN_CFI))) {
2824 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2826 union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2827 oa->dl_tci.tci = tci & ~htons(VLAN_CFI);
2832 xlate_reg_move_action(struct action_xlate_ctx *ctx,
2833 const struct nx_action_reg_move *narm)
2835 ovs_be16 old_tci = ctx->flow.vlan_tci;
2837 nxm_execute_reg_move(narm, &ctx->flow);
2839 if (ctx->flow.vlan_tci != old_tci) {
2840 xlate_set_dl_tci(ctx);
2845 xlate_nicira_action(struct action_xlate_ctx *ctx,
2846 const struct nx_action_header *nah)
2848 const struct nx_action_resubmit *nar;
2849 const struct nx_action_set_tunnel *nast;
2850 const struct nx_action_set_queue *nasq;
2851 union odp_action *oa;
2852 int subtype = ntohs(nah->subtype);
2854 assert(nah->vendor == htonl(NX_VENDOR_ID));
2856 case NXAST_RESUBMIT:
2857 nar = (const struct nx_action_resubmit *) nah;
2858 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2861 case NXAST_SET_TUNNEL:
2862 nast = (const struct nx_action_set_tunnel *) nah;
2863 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2864 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2867 case NXAST_DROP_SPOOFED_ARP:
2868 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2869 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2873 case NXAST_SET_QUEUE:
2874 nasq = (const struct nx_action_set_queue *) nah;
2875 xlate_set_queue_action(ctx, nasq);
2878 case NXAST_POP_QUEUE:
2879 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2882 case NXAST_REG_MOVE:
2883 xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah);
2886 case NXAST_REG_LOAD:
2887 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2891 /* Nothing to do. */
2894 /* If you add a new action here that modifies flow data, don't forget to
2895 * update the flow key in ctx->flow at the same time. */
2898 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2904 do_xlate_actions(const union ofp_action *in, size_t n_in,
2905 struct action_xlate_ctx *ctx)
2907 struct actions_iterator iter;
2908 const union ofp_action *ia;
2909 const struct ofport *port;
2911 port = get_port(ctx->ofproto, ctx->flow.in_port);
2912 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2913 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2914 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2915 /* Drop this flow. */
2919 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2920 uint16_t type = ntohs(ia->type);
2921 union odp_action *oa;
2925 xlate_output_action(ctx, &ia->output);
2928 case OFPAT_SET_VLAN_VID:
2929 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2930 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2931 xlate_set_dl_tci(ctx);
2934 case OFPAT_SET_VLAN_PCP:
2935 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2936 ctx->flow.vlan_tci |= htons(
2937 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2938 xlate_set_dl_tci(ctx);
2941 case OFPAT_STRIP_VLAN:
2942 ctx->flow.vlan_tci = htons(0);
2943 xlate_set_dl_tci(ctx);
2946 case OFPAT_SET_DL_SRC:
2947 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2948 memcpy(oa->dl_addr.dl_addr,
2949 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2950 memcpy(ctx->flow.dl_src,
2951 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2954 case OFPAT_SET_DL_DST:
2955 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2956 memcpy(oa->dl_addr.dl_addr,
2957 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2958 memcpy(ctx->flow.dl_dst,
2959 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2962 case OFPAT_SET_NW_SRC:
2963 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2964 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2967 case OFPAT_SET_NW_DST:
2968 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2969 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2972 case OFPAT_SET_NW_TOS:
2973 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2974 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2977 case OFPAT_SET_TP_SRC:
2978 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2979 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2982 case OFPAT_SET_TP_DST:
2983 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2984 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2988 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2992 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2996 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
3003 xlate_actions(const union ofp_action *in, size_t n_in,
3004 const struct flow *flow, struct ofproto *ofproto,
3005 const struct ofpbuf *packet,
3006 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
3007 uint16_t *nf_output_iface)
3009 struct action_xlate_ctx ctx;
3011 COVERAGE_INC(ofproto_ofp2odp);
3012 odp_actions_init(out);
3015 ctx.ofproto = ofproto;
3016 ctx.packet = packet;
3019 ctx.may_set_up_flow = true;
3020 ctx.nf_output_iface = NF_OUT_DROP;
3021 do_xlate_actions(in, n_in, &ctx);
3022 remove_pop_action(&ctx);
3024 /* Check with in-band control to see if we're allowed to set up this
3026 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
3027 ctx.may_set_up_flow = false;
3033 if (may_set_up_flow) {
3034 *may_set_up_flow = ctx.may_set_up_flow;
3036 if (nf_output_iface) {
3037 *nf_output_iface = ctx.nf_output_iface;
3039 if (odp_actions_overflow(out)) {
3040 COVERAGE_INC(odp_overflow);
3041 odp_actions_init(out);
3042 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
3047 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3048 * error message code (composed with ofp_mkerr()) for the caller to propagate
3049 * upward. Otherwise, returns 0.
3051 * The log message mentions 'msg_type'. */
3053 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3055 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3056 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3057 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3060 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3067 handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh)
3069 struct ofproto *p = ofconn->ofproto;
3070 struct ofp_packet_out *opo;
3071 struct ofpbuf payload, *buffer;
3072 union ofp_action *ofp_actions;
3073 struct odp_actions odp_actions;
3074 struct ofpbuf request;
3076 size_t n_ofp_actions;
3080 COVERAGE_INC(ofproto_packet_out);
3082 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3087 /* Get ofp_packet_out. */
3089 request.size = ntohs(oh->length);
3090 opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions));
3092 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3096 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3097 &ofp_actions, &n_ofp_actions);
3103 if (opo->buffer_id != htonl(UINT32_MAX)) {
3104 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3106 if (error || !buffer) {
3115 /* Extract flow, check actions. */
3116 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3118 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3124 error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload,
3125 &odp_actions, NULL, NULL, NULL);
3127 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions,
3132 ofpbuf_delete(buffer);
3137 update_port_config(struct ofproto *p, struct ofport *port,
3138 uint32_t config, uint32_t mask)
3140 mask &= config ^ port->opp.config;
3141 if (mask & OFPPC_PORT_DOWN) {
3142 if (config & OFPPC_PORT_DOWN) {
3143 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3145 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3148 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3149 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3150 if (mask & REVALIDATE_BITS) {
3151 COVERAGE_INC(ofproto_costly_flags);
3152 port->opp.config ^= mask & REVALIDATE_BITS;
3153 p->need_revalidate = true;
3155 #undef REVALIDATE_BITS
3156 if (mask & OFPPC_NO_PACKET_IN) {
3157 port->opp.config ^= OFPPC_NO_PACKET_IN;
3162 handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh)
3164 struct ofproto *p = ofconn->ofproto;
3165 const struct ofp_port_mod *opm;
3166 struct ofport *port;
3169 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3173 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
3177 opm = (struct ofp_port_mod *) oh;
3179 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3181 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3182 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3183 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3185 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3186 if (opm->advertise) {
3187 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3193 static struct ofpbuf *
3194 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3196 struct ofp_stats_reply *osr;
3199 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3200 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3202 osr->flags = htons(0);
3206 static struct ofpbuf *
3207 start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3209 return make_ofp_stats_reply(request->header.xid, request->type, body_len);
3213 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3214 struct ofpbuf **msgp)
3216 struct ofpbuf *msg = *msgp;
3217 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3218 if (nbytes + msg->size > UINT16_MAX) {
3219 struct ofp_stats_reply *reply = msg->data;
3220 reply->flags = htons(OFPSF_REPLY_MORE);
3221 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3222 queue_tx(msg, ofconn, ofconn->reply_counter);
3224 return ofpbuf_put_uninit(*msgp, nbytes);
3227 static struct ofpbuf *
3228 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3230 struct nicira_stats_msg *nsm;
3233 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3234 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3235 nsm->type = htons(OFPST_VENDOR);
3236 nsm->flags = htons(0);
3237 nsm->vendor = htonl(NX_VENDOR_ID);
3238 nsm->subtype = htonl(subtype);
3242 static struct ofpbuf *
3243 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3245 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3249 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3250 struct ofpbuf **msgp)
3252 struct ofpbuf *msg = *msgp;
3253 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3254 if (nbytes + msg->size > UINT16_MAX) {
3255 struct nicira_stats_msg *reply = msg->data;
3256 reply->flags = htons(OFPSF_REPLY_MORE);
3257 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3258 queue_tx(msg, ofconn, ofconn->reply_counter);
3260 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3264 handle_desc_stats_request(struct ofconn *ofconn,
3265 struct ofp_stats_request *request)
3267 struct ofproto *p = ofconn->ofproto;
3268 struct ofp_desc_stats *ods;
3271 msg = start_ofp_stats_reply(request, sizeof *ods);
3272 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3273 memset(ods, 0, sizeof *ods);
3274 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3275 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3276 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3277 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3278 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3279 queue_tx(msg, ofconn, ofconn->reply_counter);
3285 handle_table_stats_request(struct ofconn *ofconn,
3286 struct ofp_stats_request *request)
3288 struct ofproto *p = ofconn->ofproto;
3289 struct ofp_table_stats *ots;
3292 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3294 /* Classifier table. */
3295 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3296 memset(ots, 0, sizeof *ots);
3297 strcpy(ots->name, "classifier");
3298 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3299 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3300 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3301 ots->active_count = htonl(classifier_count(&p->cls));
3302 ots->lookup_count = htonll(0); /* XXX */
3303 ots->matched_count = htonll(0); /* XXX */
3305 queue_tx(msg, ofconn, ofconn->reply_counter);
3310 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3311 struct ofpbuf **msgp)
3313 struct netdev_stats stats;
3314 struct ofp_port_stats *ops;
3316 /* Intentionally ignore return value, since errors will set
3317 * 'stats' to all-1s, which is correct for OpenFlow, and
3318 * netdev_get_stats() will log errors. */
3319 netdev_get_stats(port->netdev, &stats);
3321 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3322 ops->port_no = htons(port->opp.port_no);
3323 memset(ops->pad, 0, sizeof ops->pad);
3324 ops->rx_packets = htonll(stats.rx_packets);
3325 ops->tx_packets = htonll(stats.tx_packets);
3326 ops->rx_bytes = htonll(stats.rx_bytes);
3327 ops->tx_bytes = htonll(stats.tx_bytes);
3328 ops->rx_dropped = htonll(stats.rx_dropped);
3329 ops->tx_dropped = htonll(stats.tx_dropped);
3330 ops->rx_errors = htonll(stats.rx_errors);
3331 ops->tx_errors = htonll(stats.tx_errors);
3332 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3333 ops->rx_over_err = htonll(stats.rx_over_errors);
3334 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3335 ops->collisions = htonll(stats.collisions);
3339 handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr,
3342 struct ofproto *p = ofconn->ofproto;
3343 struct ofp_port_stats_request *psr;
3344 struct ofp_port_stats *ops;
3346 struct ofport *port;
3348 if (arg_size != sizeof *psr) {
3349 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3351 psr = (struct ofp_port_stats_request *) osr->body;
3353 msg = start_ofp_stats_reply(osr, sizeof *ops * 16);
3354 if (psr->port_no != htons(OFPP_NONE)) {
3355 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3357 append_port_stat(port, ofconn, &msg);
3360 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3361 append_port_stat(port, ofconn, &msg);
3365 queue_tx(msg, ofconn, ofconn->reply_counter);
3369 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3370 * '*packet_countp' and '*byte_countp'. The returned statistics include
3371 * statistics for all of 'rule''s facets. */
3373 query_stats(struct ofproto *p, struct rule *rule,
3374 uint64_t *packet_countp, uint64_t *byte_countp)
3376 uint64_t packet_count, byte_count;
3377 struct facet *facet;
3378 struct odp_flow *odp_flows;
3381 /* Start from historical data for 'rule' itself that are no longer tracked
3382 * by the datapath. This counts, for example, facets that have expired. */
3383 packet_count = rule->packet_count;
3384 byte_count = rule->byte_count;
3386 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3388 * Also, add any statistics that are not tracked by the datapath for each
3389 * facet. This includes, for example, statistics for packets that were
3390 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3392 odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows);
3394 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3395 struct odp_flow *odp_flow = &odp_flows[n_odp_flows++];
3396 odp_flow_key_from_flow(&odp_flow->key, &facet->flow);
3397 packet_count += facet->packet_count;
3398 byte_count += facet->byte_count;
3401 /* Fetch up-to-date statistics from the datapath and add them in. */
3402 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3405 for (i = 0; i < n_odp_flows; i++) {
3406 struct odp_flow *odp_flow = &odp_flows[i];
3407 packet_count += odp_flow->stats.n_packets;
3408 byte_count += odp_flow->stats.n_bytes;
3413 /* Return the stats to the caller. */
3414 *packet_countp = packet_count;
3415 *byte_countp = byte_count;
3419 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3421 long long int msecs = time_msec() - start;
3422 *sec = htonl(msecs / 1000);
3423 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3427 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3428 ovs_be16 out_port, struct ofpbuf **replyp)
3430 struct ofp_flow_stats *ofs;
3431 uint64_t packet_count, byte_count;
3432 size_t act_len, len;
3434 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3438 act_len = sizeof *rule->actions * rule->n_actions;
3439 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3441 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3443 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3444 ofs->length = htons(len);
3447 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match);
3448 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3449 ofs->cookie = rule->flow_cookie;
3450 ofs->priority = htons(rule->cr.priority);
3451 ofs->idle_timeout = htons(rule->idle_timeout);
3452 ofs->hard_timeout = htons(rule->hard_timeout);
3453 memset(ofs->pad2, 0, sizeof ofs->pad2);
3454 ofs->packet_count = htonll(packet_count);
3455 ofs->byte_count = htonll(byte_count);
3456 if (rule->n_actions > 0) {
3457 memcpy(ofs->actions, rule->actions, act_len);
3462 is_valid_table(uint8_t table_id)
3464 return table_id == 0 || table_id == 0xff;
3468 handle_flow_stats_request(struct ofconn *ofconn,
3469 const struct ofp_stats_request *osr, size_t arg_size)
3471 struct ofp_flow_stats_request *fsr;
3472 struct ofpbuf *reply;
3474 if (arg_size != sizeof *fsr) {
3475 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3477 fsr = (struct ofp_flow_stats_request *) osr->body;
3479 COVERAGE_INC(ofproto_flows_req);
3480 reply = start_ofp_stats_reply(osr, 1024);
3481 if (is_valid_table(fsr->table_id)) {
3482 struct cls_cursor cursor;
3483 struct cls_rule target;
3486 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3488 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3489 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3490 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3493 queue_tx(reply, ofconn, ofconn->reply_counter);
3499 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3500 ovs_be16 out_port, struct ofpbuf **replyp)
3502 struct nx_flow_stats *nfs;
3503 uint64_t packet_count, byte_count;
3504 size_t act_len, start_len;
3505 struct ofpbuf *reply;
3507 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3511 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3513 act_len = sizeof *rule->actions * rule->n_actions;
3515 start_len = (*replyp)->size;
3516 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3519 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3522 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3523 nfs->cookie = rule->flow_cookie;
3524 nfs->priority = htons(rule->cr.priority);
3525 nfs->idle_timeout = htons(rule->idle_timeout);
3526 nfs->hard_timeout = htons(rule->hard_timeout);
3527 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3528 memset(nfs->pad2, 0, sizeof nfs->pad2);
3529 nfs->packet_count = htonll(packet_count);
3530 nfs->byte_count = htonll(byte_count);
3531 if (rule->n_actions > 0) {
3532 ofpbuf_put(reply, rule->actions, act_len);
3534 nfs->length = htons(reply->size - start_len);
3538 handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b)
3540 struct nx_flow_stats_request *nfsr;
3541 struct cls_rule target;
3542 struct ofpbuf *reply;
3545 /* Dissect the message. */
3546 nfsr = ofpbuf_try_pull(b, sizeof *nfsr);
3548 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3550 error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target);
3555 COVERAGE_INC(ofproto_flows_req);
3556 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3557 if (is_valid_table(nfsr->table_id)) {
3558 struct cls_cursor cursor;
3561 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3562 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3563 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3566 queue_tx(reply, ofconn, ofconn->reply_counter);
3572 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3574 struct ofp_match match;
3575 uint64_t packet_count, byte_count;
3576 size_t act_len = sizeof *rule->actions * rule->n_actions;
3578 query_stats(ofproto, rule, &packet_count, &byte_count);
3579 ofputil_cls_rule_to_match(&rule->cr, NXFF_OPENFLOW10, &match);
3581 ds_put_format(results, "duration=%llds, ",
3582 (time_msec() - rule->created) / 1000);
3583 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3584 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3585 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3586 ofp_print_match(results, &match, true);
3588 ofp_print_actions(results, &rule->actions->header, act_len);
3590 ds_put_cstr(results, "drop");
3592 ds_put_cstr(results, "\n");
3595 /* Adds a pretty-printed description of all flows to 'results', including
3596 * those marked hidden by secchan (e.g., by in-band control). */
3598 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3600 struct cls_cursor cursor;
3603 cls_cursor_init(&cursor, &p->cls, NULL);
3604 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3605 flow_stats_ds(p, rule, results);
3610 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3611 ovs_be16 out_port, uint8_t table_id,
3612 struct ofp_aggregate_stats_reply *oasr)
3614 uint64_t total_packets = 0;
3615 uint64_t total_bytes = 0;
3618 COVERAGE_INC(ofproto_agg_request);
3620 if (is_valid_table(table_id)) {
3621 struct cls_cursor cursor;
3624 cls_cursor_init(&cursor, &ofproto->cls, target);
3625 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3626 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3627 uint64_t packet_count;
3628 uint64_t byte_count;
3630 query_stats(ofproto, rule, &packet_count, &byte_count);
3632 total_packets += packet_count;
3633 total_bytes += byte_count;
3639 oasr->flow_count = htonl(n_flows);
3640 oasr->packet_count = htonll(total_packets);
3641 oasr->byte_count = htonll(total_bytes);
3642 memset(oasr->pad, 0, sizeof oasr->pad);
3646 handle_aggregate_stats_request(struct ofconn *ofconn,
3647 const struct ofp_stats_request *osr,
3650 struct ofp_aggregate_stats_request *request;
3651 struct ofp_aggregate_stats_reply *reply;
3652 struct cls_rule target;
3655 if (arg_size != sizeof *request) {
3656 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3658 request = (struct ofp_aggregate_stats_request *) osr->body;
3660 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3663 msg = start_ofp_stats_reply(osr, sizeof *reply);
3664 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3665 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3666 request->table_id, reply);
3667 queue_tx(msg, ofconn, ofconn->reply_counter);
3672 handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b)
3674 struct nx_aggregate_stats_request *request;
3675 struct ofp_aggregate_stats_reply *reply;
3676 struct cls_rule target;
3680 /* Dissect the message. */
3681 request = ofpbuf_try_pull(b, sizeof *request);
3683 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3685 error = nx_pull_match(b, ntohs(request->match_len), 0, &target);
3691 COVERAGE_INC(ofproto_flows_req);
3692 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3693 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3694 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3695 request->table_id, reply);
3696 queue_tx(buf, ofconn, ofconn->reply_counter);
3701 struct queue_stats_cbdata {
3702 struct ofconn *ofconn;
3703 struct ofport *ofport;
3708 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3709 const struct netdev_queue_stats *stats)
3711 struct ofp_queue_stats *reply;
3713 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3714 reply->port_no = htons(cbdata->ofport->opp.port_no);
3715 memset(reply->pad, 0, sizeof reply->pad);
3716 reply->queue_id = htonl(queue_id);
3717 reply->tx_bytes = htonll(stats->tx_bytes);
3718 reply->tx_packets = htonll(stats->tx_packets);
3719 reply->tx_errors = htonll(stats->tx_errors);
3723 handle_queue_stats_dump_cb(uint32_t queue_id,
3724 struct netdev_queue_stats *stats,
3727 struct queue_stats_cbdata *cbdata = cbdata_;
3729 put_queue_stats(cbdata, queue_id, stats);
3733 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3734 struct queue_stats_cbdata *cbdata)
3736 cbdata->ofport = port;
3737 if (queue_id == OFPQ_ALL) {
3738 netdev_dump_queue_stats(port->netdev,
3739 handle_queue_stats_dump_cb, cbdata);
3741 struct netdev_queue_stats stats;
3743 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3744 put_queue_stats(cbdata, queue_id, &stats);
3750 handle_queue_stats_request(struct ofconn *ofconn,
3751 const struct ofp_stats_request *osr,
3754 struct ofproto *ofproto = ofconn->ofproto;
3755 struct ofp_queue_stats_request *qsr;
3756 struct queue_stats_cbdata cbdata;
3757 struct ofport *port;
3758 unsigned int port_no;
3761 if (arg_size != sizeof *qsr) {
3762 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3764 qsr = (struct ofp_queue_stats_request *) osr->body;
3766 COVERAGE_INC(ofproto_queue_req);
3768 cbdata.ofconn = ofconn;
3769 cbdata.msg = start_ofp_stats_reply(osr, 128);
3771 port_no = ntohs(qsr->port_no);
3772 queue_id = ntohl(qsr->queue_id);
3773 if (port_no == OFPP_ALL) {
3774 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3775 handle_queue_stats_for_port(port, queue_id, &cbdata);
3777 } else if (port_no < ofproto->max_ports) {
3778 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3780 handle_queue_stats_for_port(port, queue_id, &cbdata);
3783 ofpbuf_delete(cbdata.msg);
3784 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3786 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3792 handle_vendor_stats_request(struct ofconn *ofconn,
3793 struct ofp_stats_request *osr, size_t arg_size)
3795 struct nicira_stats_msg *nsm;
3800 VLOG_WARN_RL(&rl, "truncated vendor stats request body");
3801 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3804 memcpy(&vendor, osr->body, sizeof vendor);
3805 if (vendor != htonl(NX_VENDOR_ID)) {
3806 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3809 if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) {
3810 VLOG_WARN_RL(&rl, "truncated Nicira stats request");
3811 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3814 nsm = (struct nicira_stats_msg *) osr;
3816 b.size = ntohs(nsm->header.length);
3817 switch (ntohl(nsm->subtype)) {
3819 return handle_nxst_flow(ofconn, &b);
3821 case NXST_AGGREGATE:
3822 return handle_nxst_aggregate(ofconn, &b);
3825 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3830 handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh)
3832 struct ofp_stats_request *osr;
3836 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3841 osr = (struct ofp_stats_request *) oh;
3843 switch (ntohs(osr->type)) {
3845 return handle_desc_stats_request(ofconn, osr);
3848 return handle_flow_stats_request(ofconn, osr, arg_size);
3850 case OFPST_AGGREGATE:
3851 return handle_aggregate_stats_request(ofconn, osr, arg_size);
3854 return handle_table_stats_request(ofconn, osr);
3857 return handle_port_stats_request(ofconn, osr, arg_size);
3860 return handle_queue_stats_request(ofconn, osr, arg_size);
3863 return handle_vendor_stats_request(ofconn, osr, arg_size);
3866 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3870 static long long int
3871 msec_from_nsec(uint64_t sec, uint32_t nsec)
3873 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3877 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3878 const struct odp_flow_stats *stats)
3880 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3881 if (used > facet->used) {
3883 if (used > facet->rule->used) {
3884 facet->rule->used = used;
3886 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3890 /* Folds the statistics from 'stats' into the counters in 'facet'.
3892 * Because of the meaning of a facet's counters, it only makes sense to do this
3893 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3894 * packet that was sent by hand or if it represents statistics that have been
3895 * cleared out of the datapath. */
3897 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3898 const struct odp_flow_stats *stats)
3900 if (stats->n_packets) {
3901 facet_update_time(ofproto, facet, stats);
3902 facet->packet_count += stats->n_packets;
3903 facet->byte_count += stats->n_bytes;
3904 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3912 uint16_t idle_timeout;
3913 uint16_t hard_timeout;
3917 union ofp_action *actions;
3921 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3922 * in which no matching flow already exists in the flow table.
3924 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3925 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3926 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3928 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3931 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3933 struct ofproto *p = ofconn->ofproto;
3934 struct ofpbuf *packet;
3939 if (fm->flags & OFPFF_CHECK_OVERLAP
3940 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3941 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3945 if (fm->buffer_id != UINT32_MAX) {
3946 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3950 in_port = UINT16_MAX;
3953 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3954 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3955 fm->flags & OFPFF_SEND_FLOW_REM);
3956 rule_insert(p, rule);
3958 rule_execute(p, rule, in_port, packet);
3963 static struct rule *
3964 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3966 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3970 send_buffered_packet(struct ofconn *ofconn,
3971 struct rule *rule, uint32_t buffer_id)
3973 struct ofpbuf *packet;
3977 if (buffer_id == UINT32_MAX) {
3981 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3986 rule_execute(ofconn->ofproto, rule, in_port, packet);
3991 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3993 struct modify_flows_cbdata {
3994 struct ofproto *ofproto;
3995 const struct flow_mod *fm;
3999 static int modify_flow(struct ofproto *, const struct flow_mod *,
4002 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4003 * encoded by ofp_mkerr() on failure.
4005 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4008 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
4010 struct ofproto *p = ofconn->ofproto;
4011 struct rule *match = NULL;
4012 struct cls_cursor cursor;
4015 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4016 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4017 if (!rule_is_hidden(rule)) {
4019 modify_flow(p, fm, rule);
4024 /* This credits the packet to whichever flow happened to match last.
4025 * That's weird. Maybe we should do a lookup for the flow that
4026 * actually matches the packet? Who knows. */
4027 send_buffered_packet(ofconn, match, fm->buffer_id);
4030 return add_flow(ofconn, fm);
4034 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4035 * code as encoded by ofp_mkerr() on failure.
4037 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4040 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4042 struct ofproto *p = ofconn->ofproto;
4043 struct rule *rule = find_flow_strict(p, fm);
4044 if (rule && !rule_is_hidden(rule)) {
4045 modify_flow(p, fm, rule);
4046 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4048 return add_flow(ofconn, fm);
4052 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4053 * been identified as a flow in 'p''s flow table to be modified, by changing
4054 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4055 * ofp_action[] structures). */
4057 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4059 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4061 rule->flow_cookie = fm->cookie;
4063 /* If the actions are the same, do nothing. */
4064 if (fm->n_actions == rule->n_actions
4066 || !memcmp(fm->actions, rule->actions, actions_len))) {
4070 /* Replace actions. */
4071 free(rule->actions);
4072 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4073 rule->n_actions = fm->n_actions;
4075 p->need_revalidate = true;
4080 /* OFPFC_DELETE implementation. */
4082 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4084 /* Implements OFPFC_DELETE. */
4086 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4088 struct rule *rule, *next_rule;
4089 struct cls_cursor cursor;
4091 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4092 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4093 delete_flow(p, rule, htons(fm->out_port));
4097 /* Implements OFPFC_DELETE_STRICT. */
4099 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4101 struct rule *rule = find_flow_strict(p, fm);
4103 delete_flow(p, rule, htons(fm->out_port));
4107 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4108 * been identified as a flow to delete from 'p''s flow table, by deleting the
4109 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4112 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4113 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4114 * specified 'out_port'. */
4116 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4118 if (rule_is_hidden(rule)) {
4122 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4126 rule_send_removed(p, rule, OFPRR_DELETE);
4127 rule_remove(p, rule);
4131 flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm)
4133 struct ofproto *p = ofconn->ofproto;
4136 error = reject_slave_controller(ofconn, "flow_mod");
4141 error = validate_actions(fm->actions, fm->n_actions,
4142 &fm->cr.flow, p->max_ports);
4147 /* We do not support the emergency flow cache. It will hopefully
4148 * get dropped from OpenFlow in the near future. */
4149 if (fm->flags & OFPFF_EMERG) {
4150 /* There isn't a good fit for an error code, so just state that the
4151 * flow table is full. */
4152 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4155 switch (fm->command) {
4157 return add_flow(ofconn, fm);
4160 return modify_flows_loose(ofconn, fm);
4162 case OFPFC_MODIFY_STRICT:
4163 return modify_flow_strict(ofconn, fm);
4166 delete_flows_loose(p, fm);
4169 case OFPFC_DELETE_STRICT:
4170 delete_flow_strict(p, fm);
4174 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4179 handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4181 struct ofp_match orig_match;
4182 struct ofp_flow_mod *ofm;
4188 b.size = ntohs(oh->length);
4190 /* Dissect the message. */
4191 ofm = ofpbuf_try_pull(&b, sizeof *ofm);
4193 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4195 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4200 /* Normalize ofm->match. If normalization actually changes anything, then
4201 * log the differences. */
4202 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
4203 orig_match = ofm->match;
4204 normalize_match(&ofm->match);
4205 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
4206 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4207 if (!VLOG_DROP_INFO(&normal_rl)) {
4208 char *old = ofp_match_to_literal_string(&orig_match);
4209 char *new = ofp_match_to_literal_string(&ofm->match);
4210 VLOG_INFO("%s: normalization changed ofp_match, details:",
4211 rconn_get_name(ofconn->rconn));
4212 VLOG_INFO(" pre: %s", old);
4213 VLOG_INFO("post: %s", new);
4219 /* Translate the message. */
4220 ofputil_cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
4221 ofconn->flow_format, ofm->cookie, &fm.cr);
4222 fm.cookie = ofm->cookie;
4223 fm.command = ntohs(ofm->command);
4224 fm.idle_timeout = ntohs(ofm->idle_timeout);
4225 fm.hard_timeout = ntohs(ofm->hard_timeout);
4226 fm.buffer_id = ntohl(ofm->buffer_id);
4227 fm.out_port = ntohs(ofm->out_port);
4228 fm.flags = ntohs(ofm->flags);
4230 /* Execute the command. */
4231 return flow_mod_core(ofconn, &fm);
4235 handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4237 struct nx_flow_mod *nfm;
4243 b.size = ntohs(oh->length);
4245 /* Dissect the message. */
4246 nfm = ofpbuf_try_pull(&b, sizeof *nfm);
4248 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4250 error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority),
4255 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4260 /* Translate the message. */
4261 fm.cookie = nfm->cookie;
4262 fm.command = ntohs(nfm->command);
4263 fm.idle_timeout = ntohs(nfm->idle_timeout);
4264 fm.hard_timeout = ntohs(nfm->hard_timeout);
4265 fm.buffer_id = ntohl(nfm->buffer_id);
4266 fm.out_port = ntohs(nfm->out_port);
4267 fm.flags = ntohs(nfm->flags);
4269 /* Execute the command. */
4270 return flow_mod_core(ofconn, &fm);
4274 handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg)
4278 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4283 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4288 handle_role_request(struct ofconn *ofconn, struct nicira_header *msg)
4290 struct nx_role_request *nrr;
4291 struct nx_role_request *reply;
4295 if (ntohs(msg->header.length) != sizeof *nrr) {
4296 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
4297 ntohs(msg->header.length), sizeof *nrr);
4298 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4300 nrr = (struct nx_role_request *) msg;
4302 if (ofconn->type != OFCONN_PRIMARY) {
4303 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4305 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4308 role = ntohl(nrr->role);
4309 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4310 && role != NX_ROLE_SLAVE) {
4311 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4313 /* There's no good error code for this. */
4314 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4317 if (role == NX_ROLE_MASTER) {
4318 struct ofconn *other;
4320 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4321 if (other->role == NX_ROLE_MASTER) {
4322 other->role = NX_ROLE_SLAVE;
4326 ofconn->role = role;
4328 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid,
4330 reply->role = htonl(role);
4331 queue_tx(buf, ofconn, ofconn->reply_counter);
4337 handle_nxt_set_flow_format(struct ofconn *ofconn,
4338 struct nxt_set_flow_format *msg)
4343 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4348 format = ntohl(msg->format);
4349 if (format == NXFF_OPENFLOW10
4350 || format == NXFF_TUN_ID_FROM_COOKIE
4351 || format == NXFF_NXM) {
4352 ofconn->flow_format = format;
4355 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4360 handle_vendor(struct ofconn *ofconn, void *msg)
4362 struct ofproto *p = ofconn->ofproto;
4363 struct ofp_vendor_header *ovh = msg;
4364 struct nicira_header *nh;
4366 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4367 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4368 "(expected at least %zu)",
4369 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4370 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4372 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4373 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4375 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4376 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4377 "(expected at least %zu)",
4378 ntohs(ovh->header.length), sizeof(struct nicira_header));
4379 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4383 switch (ntohl(nh->subtype)) {
4384 case NXT_STATUS_REQUEST:
4385 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4388 case NXT_TUN_ID_FROM_COOKIE:
4389 return handle_tun_id_from_cookie(ofconn, msg);
4391 case NXT_ROLE_REQUEST:
4392 return handle_role_request(ofconn, msg);
4394 case NXT_SET_FLOW_FORMAT:
4395 return handle_nxt_set_flow_format(ofconn, msg);
4398 return handle_nxt_flow_mod(ofconn, &ovh->header);
4401 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4405 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4407 struct ofp_header *ob;
4410 /* Currently, everything executes synchronously, so we can just
4411 * immediately send the barrier reply. */
4412 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4413 queue_tx(buf, ofconn, ofconn->reply_counter);
4418 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4420 struct ofp_header *oh = ofp_msg->data;
4423 COVERAGE_INC(ofproto_recv_openflow);
4425 case OFPT_ECHO_REQUEST:
4426 error = handle_echo_request(ofconn, oh);
4429 case OFPT_ECHO_REPLY:
4433 case OFPT_FEATURES_REQUEST:
4434 error = handle_features_request(ofconn, oh);
4437 case OFPT_GET_CONFIG_REQUEST:
4438 error = handle_get_config_request(ofconn, oh);
4441 case OFPT_SET_CONFIG:
4442 error = handle_set_config(ofconn, ofp_msg->data);
4445 case OFPT_PACKET_OUT:
4446 error = handle_packet_out(ofconn, ofp_msg->data);
4450 error = handle_port_mod(ofconn, oh);
4454 error = handle_ofpt_flow_mod(ofconn, ofp_msg->data);
4457 case OFPT_STATS_REQUEST:
4458 error = handle_stats_request(ofconn, oh);
4462 error = handle_vendor(ofconn, ofp_msg->data);
4465 case OFPT_BARRIER_REQUEST:
4466 error = handle_barrier_request(ofconn, oh);
4470 if (VLOG_IS_WARN_ENABLED()) {
4471 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4472 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4475 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4480 send_error_oh(ofconn, ofp_msg->data, error);
4485 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4487 struct odp_msg *msg = packet->data;
4488 struct ofpbuf payload;
4489 struct facet *facet;
4492 payload.data = msg + 1;
4493 payload.size = msg->length - sizeof *msg;
4494 flow_extract(&payload, msg->arg, msg->port, &flow);
4496 packet->l2 = payload.l2;
4497 packet->l3 = payload.l3;
4498 packet->l4 = payload.l4;
4499 packet->l7 = payload.l7;
4501 /* Check with in-band control to see if this packet should be sent
4502 * to the local port regardless of the flow table. */
4503 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4504 union odp_action action;
4506 memset(&action, 0, sizeof(action));
4507 action.output.type = ODPAT_OUTPUT;
4508 action.output.port = ODPP_LOCAL;
4509 dpif_execute(p->dpif, &action, 1, &payload);
4512 facet = facet_lookup_valid(p, &flow);
4514 struct rule *rule = rule_lookup(p, &flow);
4516 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4517 struct ofport *port = get_port(p, msg->port);
4519 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4520 COVERAGE_INC(ofproto_no_packet_in);
4521 /* XXX install 'drop' flow entry */
4522 ofpbuf_delete(packet);
4526 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4530 COVERAGE_INC(ofproto_packet_in);
4531 send_packet_in(p, packet);
4535 facet = facet_create(p, rule, &flow, packet);
4536 } else if (!facet->may_install) {
4537 /* The facet is not installable, that is, we need to process every
4538 * packet, so process the current packet's actions into 'facet'. */
4539 facet_make_actions(p, facet, packet);
4542 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4544 * Extra-special case for fail-open mode.
4546 * We are in fail-open mode and the packet matched the fail-open rule,
4547 * but we are connected to a controller too. We should send the packet
4548 * up to the controller in the hope that it will try to set up a flow
4549 * and thereby allow us to exit fail-open.
4551 * See the top-level comment in fail-open.c for more information.
4553 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4554 DPIF_RECV_MSG_PADDING));
4557 ofpbuf_pull(packet, sizeof *msg);
4558 facet_execute(p, facet, packet);
4559 facet_install(p, facet, false);
4563 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4565 struct odp_msg *msg = packet->data;
4567 switch (msg->type) {
4568 case _ODPL_ACTION_NR:
4569 COVERAGE_INC(ofproto_ctlr_action);
4570 send_packet_in(p, packet);
4573 case _ODPL_SFLOW_NR:
4575 ofproto_sflow_received(p->sflow, msg);
4577 ofpbuf_delete(packet);
4581 handle_odp_miss_msg(p, packet);
4585 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4591 /* Flow expiration. */
4593 static int ofproto_dp_max_idle(const struct ofproto *);
4594 static void ofproto_update_used(struct ofproto *);
4595 static void rule_expire(struct ofproto *, struct rule *);
4596 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4598 /* This function is called periodically by ofproto_run(). Its job is to
4599 * collect updates for the flows that have been installed into the datapath,
4600 * most importantly when they last were used, and then use that information to
4601 * expire flows that have not been used recently.
4603 * Returns the number of milliseconds after which it should be called again. */
4605 ofproto_expire(struct ofproto *ofproto)
4607 struct rule *rule, *next_rule;
4608 struct cls_cursor cursor;
4611 /* Update 'used' for each flow in the datapath. */
4612 ofproto_update_used(ofproto);
4614 /* Expire facets that have been idle too long. */
4615 dp_max_idle = ofproto_dp_max_idle(ofproto);
4616 ofproto_expire_facets(ofproto, dp_max_idle);
4618 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4619 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4620 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4621 rule_expire(ofproto, rule);
4624 /* Let the hook know that we're at a stable point: all outstanding data
4625 * in existing flows has been accounted to the account_cb. Thus, the
4626 * hook can now reasonably do operations that depend on having accurate
4627 * flow volume accounting (currently, that's just bond rebalancing). */
4628 if (ofproto->ofhooks->account_checkpoint_cb) {
4629 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4632 return MIN(dp_max_idle, 1000);
4635 /* Update 'used' member of installed facets. */
4637 ofproto_update_used(struct ofproto *p)
4639 struct odp_flow *flows;
4644 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4649 for (i = 0; i < n_flows; i++) {
4650 struct odp_flow *f = &flows[i];
4651 struct facet *facet;
4654 odp_flow_key_to_flow(&f->key, &flow);
4655 facet = facet_find(p, &flow);
4657 if (facet && facet->installed) {
4658 facet_update_time(p, facet, &f->stats);
4659 facet_account(p, facet, f->stats.n_bytes);
4661 /* There's a flow in the datapath that we know nothing about.
4663 COVERAGE_INC(ofproto_unexpected_rule);
4664 dpif_flow_del(p->dpif, f);
4671 /* Calculates and returns the number of milliseconds of idle time after which
4672 * facets should expire from the datapath and we should fold their statistics
4673 * into their parent rules in userspace. */
4675 ofproto_dp_max_idle(const struct ofproto *ofproto)
4678 * Idle time histogram.
4680 * Most of the time a switch has a relatively small number of facets. When
4681 * this is the case we might as well keep statistics for all of them in
4682 * userspace and to cache them in the kernel datapath for performance as
4685 * As the number of facets increases, the memory required to maintain
4686 * statistics about them in userspace and in the kernel becomes
4687 * significant. However, with a large number of facets it is likely that
4688 * only a few of them are "heavy hitters" that consume a large amount of
4689 * bandwidth. At this point, only heavy hitters are worth caching in the
4690 * kernel and maintaining in userspaces; other facets we can discard.
4692 * The technique used to compute the idle time is to build a histogram with
4693 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4694 * that is installed in the kernel gets dropped in the appropriate bucket.
4695 * After the histogram has been built, we compute the cutoff so that only
4696 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4697 * cached. At least the most-recently-used bucket of facets is kept, so
4698 * actually an arbitrary number of facets can be kept in any given
4699 * expiration run (though the next run will delete most of those unless
4700 * they receive additional data).
4702 * This requires a second pass through the facets, in addition to the pass
4703 * made by ofproto_update_used(), because the former function never looks
4704 * at uninstallable facets.
4706 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4707 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4708 int buckets[N_BUCKETS] = { 0 };
4709 struct facet *facet;
4714 total = hmap_count(&ofproto->facets);
4715 if (total <= 1000) {
4716 return N_BUCKETS * BUCKET_WIDTH;
4719 /* Build histogram. */
4721 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4722 long long int idle = now - facet->used;
4723 int bucket = (idle <= 0 ? 0
4724 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4725 : (unsigned int) idle / BUCKET_WIDTH);
4729 /* Find the first bucket whose flows should be expired. */
4730 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4731 if (buckets[bucket]) {
4734 subtotal += buckets[bucket++];
4735 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4740 if (VLOG_IS_DBG_ENABLED()) {
4744 ds_put_cstr(&s, "keep");
4745 for (i = 0; i < N_BUCKETS; i++) {
4747 ds_put_cstr(&s, ", drop");
4750 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4753 VLOG_INFO("%s: %s (msec:count)",
4754 dpif_name(ofproto->dpif), ds_cstr(&s));
4758 return bucket * BUCKET_WIDTH;
4762 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4764 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4765 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4766 struct ofexpired expired;
4767 struct odp_flow odp_flow;
4769 /* Get updated flow stats.
4771 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4772 * updated TCP flags and (2) the dpif_flow_list_all() in
4773 * ofproto_update_used() zeroed TCP flags. */
4774 memset(&odp_flow, 0, sizeof odp_flow);
4775 if (facet->installed) {
4776 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
4777 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4778 dpif_flow_get(ofproto->dpif, &odp_flow);
4780 if (odp_flow.stats.n_packets) {
4781 facet_update_time(ofproto, facet, &odp_flow.stats);
4782 netflow_flow_update_flags(&facet->nf_flow,
4783 odp_flow.stats.tcp_flags);
4787 expired.flow = facet->flow;
4788 expired.packet_count = facet->packet_count +
4789 odp_flow.stats.n_packets;
4790 expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes;
4791 expired.used = facet->used;
4793 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4798 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4800 long long int cutoff = time_msec() - dp_max_idle;
4801 struct facet *facet, *next_facet;
4803 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4804 facet_active_timeout(ofproto, facet);
4805 if (facet->used < cutoff) {
4806 facet_remove(ofproto, facet);
4811 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4812 * then delete it entirely. */
4814 rule_expire(struct ofproto *ofproto, struct rule *rule)
4816 struct facet *facet, *next_facet;
4820 /* Has 'rule' expired? */
4822 if (rule->hard_timeout
4823 && now > rule->created + rule->hard_timeout * 1000) {
4824 reason = OFPRR_HARD_TIMEOUT;
4825 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4826 && now >rule->used + rule->idle_timeout * 1000) {
4827 reason = OFPRR_IDLE_TIMEOUT;
4832 COVERAGE_INC(ofproto_expired);
4834 /* Update stats. (This is a no-op if the rule expired due to an idle
4835 * timeout, because that only happens when the rule has no facets left.) */
4836 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4837 facet_remove(ofproto, facet);
4840 /* Get rid of the rule. */
4841 if (!rule_is_hidden(rule)) {
4842 rule_send_removed(ofproto, rule, reason);
4844 rule_remove(ofproto, rule);
4847 static struct ofpbuf *
4848 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4851 struct ofp_flow_removed *ofr;
4854 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4855 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match);
4856 ofr->cookie = rule->flow_cookie;
4857 ofr->priority = htons(rule->cr.priority);
4858 ofr->reason = reason;
4859 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4860 ofr->idle_timeout = htons(rule->idle_timeout);
4861 ofr->packet_count = htonll(rule->packet_count);
4862 ofr->byte_count = htonll(rule->byte_count);
4867 static struct ofpbuf *
4868 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4870 struct nx_flow_removed *nfr;
4874 nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf);
4876 match_len = nx_put_match(buf, &rule->cr);
4878 nfr->cookie = rule->flow_cookie;
4879 nfr->priority = htons(rule->cr.priority);
4880 nfr->reason = reason;
4881 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4882 nfr->idle_timeout = htons(rule->idle_timeout);
4883 nfr->match_len = htons(match_len);
4884 nfr->packet_count = htonll(rule->packet_count);
4885 nfr->byte_count = htonll(rule->byte_count);
4891 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4893 struct ofconn *ofconn;
4895 if (!rule->send_flow_removed) {
4899 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4902 if (!rconn_is_connected(ofconn->rconn)
4903 || !ofconn_receives_async_msgs(ofconn)) {
4907 msg = (ofconn->flow_format == NXFF_NXM
4908 ? compose_nx_flow_removed(rule, reason)
4909 : compose_ofp_flow_removed(ofconn, rule, reason));
4911 /* Account flow expirations under ofconn->reply_counter, the counter
4912 * for replies to OpenFlow requests. That works because preventing
4913 * OpenFlow requests from being processed also prevents new flows from
4914 * being added (and expiring). (It also prevents processing OpenFlow
4915 * requests that would not add new flows, so it is imperfect.) */
4916 queue_tx(msg, ofconn, ofconn->reply_counter);
4920 /* pinsched callback for sending 'packet' on 'ofconn'. */
4922 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4924 struct ofconn *ofconn = ofconn_;
4926 rconn_send_with_limit(ofconn->rconn, packet,
4927 ofconn->packet_in_counter, 100);
4930 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4931 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4932 * packet scheduler for sending.
4934 * 'max_len' specifies the maximum number of bytes of the packet to send on
4935 * 'ofconn' (INT_MAX specifies no limit).
4937 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4938 * ownership is transferred to this function. */
4940 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4943 struct ofproto *ofproto = ofconn->ofproto;
4944 struct ofp_packet_in *opi = packet->data;
4945 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4946 int send_len, trim_size;
4950 if (opi->reason == OFPR_ACTION) {
4951 buffer_id = UINT32_MAX;
4952 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4953 buffer_id = pktbuf_get_null();
4954 } else if (!ofconn->pktbuf) {
4955 buffer_id = UINT32_MAX;
4957 struct ofpbuf payload;
4958 payload.data = opi->data;
4959 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4960 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4963 /* Figure out how much of the packet to send. */
4964 send_len = ntohs(opi->total_len);
4965 if (buffer_id != UINT32_MAX) {
4966 send_len = MIN(send_len, ofconn->miss_send_len);
4968 send_len = MIN(send_len, max_len);
4970 /* Adjust packet length and clone if necessary. */
4971 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4973 packet = ofpbuf_clone_data(packet->data, trim_size);
4976 packet->size = trim_size;
4979 /* Update packet headers. */
4980 opi->buffer_id = htonl(buffer_id);
4981 update_openflow_length(packet);
4983 /* Hand over to packet scheduler. It might immediately call into
4984 * do_send_packet_in() or it might buffer it for a while (until a later
4985 * call to pinsched_run()). */
4986 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4987 packet, do_send_packet_in, ofconn);
4990 /* Replace struct odp_msg header in 'packet' by equivalent struct
4991 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4992 * returned by dpif_recv()).
4994 * The conversion is not complete: the caller still needs to trim any unneeded
4995 * payload off the end of the buffer, set the length in the OpenFlow header,
4996 * and set buffer_id. Those require us to know the controller settings and so
4997 * must be done on a per-controller basis.
4999 * Returns the maximum number of bytes of the packet that should be sent to
5000 * the controller (INT_MAX if no limit). */
5002 do_convert_to_packet_in(struct ofpbuf *packet)
5004 struct odp_msg *msg = packet->data;
5005 struct ofp_packet_in *opi;
5011 /* Extract relevant header fields */
5012 if (msg->type == _ODPL_ACTION_NR) {
5013 reason = OFPR_ACTION;
5016 reason = OFPR_NO_MATCH;
5019 total_len = msg->length - sizeof *msg;
5020 in_port = odp_port_to_ofp_port(msg->port);
5022 /* Repurpose packet buffer by overwriting header. */
5023 ofpbuf_pull(packet, sizeof(struct odp_msg));
5024 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
5025 opi->header.version = OFP_VERSION;
5026 opi->header.type = OFPT_PACKET_IN;
5027 opi->total_len = htons(total_len);
5028 opi->in_port = htons(in_port);
5029 opi->reason = reason;
5034 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
5035 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
5036 * as necessary according to their individual configurations.
5038 * 'packet' must have sufficient headroom to convert it into a struct
5039 * ofp_packet_in (e.g. as returned by dpif_recv()).
5041 * Takes ownership of 'packet'. */
5043 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
5045 struct ofconn *ofconn, *prev;
5048 max_len = do_convert_to_packet_in(packet);
5051 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5052 if (ofconn_receives_async_msgs(ofconn)) {
5054 schedule_packet_in(prev, packet, max_len, true);
5060 schedule_packet_in(prev, packet, max_len, false);
5062 ofpbuf_delete(packet);
5067 pick_datapath_id(const struct ofproto *ofproto)
5069 const struct ofport *port;
5071 port = get_port(ofproto, ODPP_LOCAL);
5073 uint8_t ea[ETH_ADDR_LEN];
5076 error = netdev_get_etheraddr(port->netdev, ea);
5078 return eth_addr_to_uint64(ea);
5080 VLOG_WARN("could not get MAC address for %s (%s)",
5081 netdev_get_name(port->netdev), strerror(error));
5083 return ofproto->fallback_dpid;
5087 pick_fallback_dpid(void)
5089 uint8_t ea[ETH_ADDR_LEN];
5090 eth_addr_nicira_random(ea);
5091 return eth_addr_to_uint64(ea);
5095 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5096 struct odp_actions *actions, tag_type *tags,
5097 uint16_t *nf_output_iface, void *ofproto_)
5099 struct ofproto *ofproto = ofproto_;
5102 /* Drop frames for reserved multicast addresses. */
5103 if (eth_addr_is_reserved(flow->dl_dst)) {
5107 /* Learn source MAC (but don't try to learn from revalidation). */
5108 if (packet != NULL) {
5109 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5111 GRAT_ARP_LOCK_NONE);
5113 /* The log messages here could actually be useful in debugging,
5114 * so keep the rate limit relatively high. */
5115 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5116 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5117 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5118 ofproto_revalidate(ofproto, rev_tag);
5122 /* Determine output port. */
5123 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5126 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5127 nf_output_iface, actions);
5128 } else if (out_port != flow->in_port) {
5129 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
5130 *nf_output_iface = out_port;
5138 static const struct ofhooks default_ofhooks = {
5139 default_normal_ofhook_cb,