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(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(odp_overflow);
69 COVERAGE_DEFINE(ofproto_agg_request);
70 COVERAGE_DEFINE(ofproto_costly_flags);
71 COVERAGE_DEFINE(ofproto_ctlr_action);
72 COVERAGE_DEFINE(ofproto_del_rule);
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_no_packet_in);
80 COVERAGE_DEFINE(ofproto_ofconn_stuck);
81 COVERAGE_DEFINE(ofproto_ofp2odp);
82 COVERAGE_DEFINE(ofproto_packet_in);
83 COVERAGE_DEFINE(ofproto_packet_out);
84 COVERAGE_DEFINE(ofproto_queue_req);
85 COVERAGE_DEFINE(ofproto_recv_openflow);
86 COVERAGE_DEFINE(ofproto_reinit_ports);
87 COVERAGE_DEFINE(ofproto_unexpected_rule);
88 COVERAGE_DEFINE(ofproto_uninstallable);
89 COVERAGE_DEFINE(ofproto_update_port);
91 #include "sflow_api.h"
96 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
97 struct netdev *netdev;
98 struct ofp_phy_port opp; /* In host byte order. */
102 static void ofport_free(struct ofport *);
103 static void hton_ofp_phy_port(struct ofp_phy_port *);
105 struct action_xlate_ctx {
106 /* action_xlate_ctx_init() initializes these members. */
109 struct ofproto *ofproto;
111 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
112 * this flow when actions change header fields. */
115 /* The packet corresponding to 'flow', or a null pointer if we are
116 * revalidating without a packet to refer to. */
117 const struct ofpbuf *packet;
119 /* xlate_actions() initializes and uses these members. The client might want
120 * to look at them after it returns. */
122 /* Datapath action set. This is xlate_actions()'s primary output. */
123 struct odp_actions out;
125 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
126 bool may_set_up_flow; /* True ordinarily; false if the actions must
127 * be reassessed for every packet. */
128 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
130 /* xlate_actions() initializes and uses these members, but the client has no
131 * reason to look at them. */
133 int recurse; /* Recursion level, via xlate_table_action. */
136 static void action_xlate_ctx_init(struct action_xlate_ctx *,
137 struct ofproto *, const struct flow *,
138 const struct ofpbuf *);
139 static int xlate_actions(struct action_xlate_ctx *ctx,
140 const union ofp_action *in, size_t n_in);
142 /* An OpenFlow flow. */
144 long long int used; /* Time last used; time created if not used. */
145 long long int created; /* Creation time. */
149 * - Do include packets and bytes from facets that have been deleted or
150 * whose own statistics have been folded into the rule.
152 * - Do include packets and bytes sent "by hand" that were accounted to
153 * the rule without any facet being involved (this is a rare corner
154 * case in rule_execute()).
156 * - Do not include packet or bytes that can be obtained from any facet's
157 * packet_count or byte_count member or that can be obtained from the
158 * datapath by, e.g., dpif_flow_get() for any facet.
160 uint64_t packet_count; /* Number of packets received. */
161 uint64_t byte_count; /* Number of bytes received. */
163 ovs_be64 flow_cookie; /* Controller-issued identifier. */
165 struct cls_rule cr; /* In owning ofproto's classifier. */
166 uint16_t idle_timeout; /* In seconds from time of last use. */
167 uint16_t hard_timeout; /* In seconds from time of creation. */
168 bool send_flow_removed; /* Send a flow removed message? */
169 int n_actions; /* Number of elements in actions[]. */
170 union ofp_action *actions; /* OpenFlow actions. */
171 struct list facets; /* List of "struct facet"s. */
174 static struct rule *rule_from_cls_rule(const struct cls_rule *);
175 static bool rule_is_hidden(const struct rule *);
177 static struct rule *rule_create(const struct cls_rule *,
178 const union ofp_action *, size_t n_actions,
179 uint16_t idle_timeout, uint16_t hard_timeout,
180 ovs_be64 flow_cookie, bool send_flow_removed);
181 static void rule_destroy(struct ofproto *, struct rule *);
182 static void rule_free(struct rule *);
184 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
185 static void rule_insert(struct ofproto *, struct rule *);
186 static void rule_remove(struct ofproto *, struct rule *);
188 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
190 /* An exact-match instantiation of an OpenFlow flow. */
192 long long int used; /* Time last used; time created if not used. */
196 * - Do include packets and bytes sent "by hand", e.g. with
199 * - Do include packets and bytes that were obtained from the datapath
200 * when a flow was deleted (e.g. dpif_flow_del()) or when its
201 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
203 * - Do not include any packets or bytes that can currently be obtained
204 * from the datapath by, e.g., dpif_flow_get().
206 uint64_t packet_count; /* Number of packets received. */
207 uint64_t byte_count; /* Number of bytes received. */
209 /* Number of bytes passed to account_cb. This may include bytes that can
210 * currently obtained from the datapath (thus, it can be greater than
212 uint64_t accounted_bytes;
214 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
215 struct list list_node; /* In owning rule's 'facets' list. */
216 struct rule *rule; /* Owning rule. */
217 struct flow flow; /* Exact-match flow. */
218 bool installed; /* Installed in datapath? */
219 bool may_install; /* True ordinarily; false if actions must
220 * be reassessed for every packet. */
221 int n_actions; /* Number of elements in actions[]. */
222 union odp_action *actions; /* Datapath actions. */
223 tag_type tags; /* Tags (set only by hooks). */
224 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
227 static struct facet *facet_create(struct ofproto *, struct rule *,
229 const struct ofpbuf *packet);
230 static void facet_remove(struct ofproto *, struct facet *);
231 static void facet_free(struct facet *);
233 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
234 static bool facet_revalidate(struct ofproto *, struct facet *);
236 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
237 static void facet_uninstall(struct ofproto *, struct facet *);
238 static void facet_flush_stats(struct ofproto *, struct facet *);
240 static void facet_make_actions(struct ofproto *, struct facet *,
241 const struct ofpbuf *packet);
242 static void facet_update_stats(struct ofproto *, struct facet *,
243 const struct odp_flow_stats *);
245 /* ofproto supports two kinds of OpenFlow connections:
247 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
248 * maintains persistent connections to these controllers and by default
249 * sends them asynchronous messages such as packet-ins.
251 * - "Service" connections, e.g. from ovs-ofctl. When these connections
252 * drop, it is the other side's responsibility to reconnect them if
253 * necessary. ofproto does not send them asynchronous messages by default.
255 * Currently, active (tcp, ssl, unix) connections are always "primary"
256 * connections and passive (ptcp, pssl, punix) connections are always "service"
257 * connections. There is no inherent reason for this, but it reflects the
261 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
262 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
265 /* A listener for incoming OpenFlow "service" connections. */
267 struct hmap_node node; /* In struct ofproto's "services" hmap. */
268 struct pvconn *pvconn; /* OpenFlow connection listener. */
270 /* These are not used by ofservice directly. They are settings for
271 * accepted "struct ofconn"s from the pvconn. */
272 int probe_interval; /* Max idle time before probing, in seconds. */
273 int rate_limit; /* Max packet-in rate in packets per second. */
274 int burst_limit; /* Limit on accumulating packet credits. */
277 static struct ofservice *ofservice_lookup(struct ofproto *,
279 static int ofservice_create(struct ofproto *,
280 const struct ofproto_controller *);
281 static void ofservice_reconfigure(struct ofservice *,
282 const struct ofproto_controller *);
283 static void ofservice_destroy(struct ofproto *, struct ofservice *);
285 /* An OpenFlow connection. */
287 struct ofproto *ofproto; /* The ofproto that owns this connection. */
288 struct list node; /* In struct ofproto's "all_conns" list. */
289 struct rconn *rconn; /* OpenFlow connection. */
290 enum ofconn_type type; /* Type. */
291 enum nx_flow_format flow_format; /* Currently selected flow format. */
293 /* OFPT_PACKET_IN related data. */
294 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
295 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
296 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
297 int miss_send_len; /* Bytes to send of buffered packets. */
299 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
300 * requests, and the maximum number before we stop reading OpenFlow
302 #define OFCONN_REPLY_MAX 100
303 struct rconn_packet_counter *reply_counter;
305 /* type == OFCONN_PRIMARY only. */
306 enum nx_role role; /* Role. */
307 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
308 struct discovery *discovery; /* Controller discovery object, if enabled. */
309 struct status_category *ss; /* Switch status category. */
310 enum ofproto_band band; /* In-band or out-of-band? */
313 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
314 * "schedulers" array. Their values are 0 and 1, and their meanings and values
315 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
316 * case anything ever changes, check their values here. */
317 #define N_SCHEDULERS 2
318 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
319 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
320 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
321 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
323 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
325 static void ofconn_destroy(struct ofconn *);
326 static void ofconn_run(struct ofconn *);
327 static void ofconn_wait(struct ofconn *);
328 static bool ofconn_receives_async_msgs(const struct ofconn *);
329 static char *ofconn_make_name(const struct ofproto *, const char *target);
330 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
332 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
333 struct rconn_packet_counter *counter);
335 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
336 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
340 uint64_t datapath_id; /* Datapath ID. */
341 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
342 char *mfr_desc; /* Manufacturer. */
343 char *hw_desc; /* Hardware. */
344 char *sw_desc; /* Software version. */
345 char *serial_desc; /* Serial number. */
346 char *dp_desc; /* Datapath description. */
350 struct netdev_monitor *netdev_monitor;
351 struct hmap ports; /* Contains "struct ofport"s. */
352 struct shash port_by_name;
356 struct switch_status *switch_status;
357 struct fail_open *fail_open;
358 struct netflow *netflow;
359 struct ofproto_sflow *sflow;
361 /* In-band control. */
362 struct in_band *in_band;
363 long long int next_in_band_update;
364 struct sockaddr_in *extra_in_band_remotes;
365 size_t n_extra_remotes;
369 struct classifier cls;
370 long long int next_expiration;
374 bool need_revalidate;
375 struct tag_set revalidate_set;
377 /* OpenFlow connections. */
378 struct hmap controllers; /* Controller "struct ofconn"s. */
379 struct list all_conns; /* Contains "struct ofconn"s. */
380 enum ofproto_fail_mode fail_mode;
382 /* OpenFlow listeners. */
383 struct hmap services; /* Contains "struct ofservice"s. */
384 struct pvconn **snoops;
387 /* Hooks for ovs-vswitchd. */
388 const struct ofhooks *ofhooks;
391 /* Used by default ofhooks. */
392 struct mac_learning *ml;
395 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
397 static const struct ofhooks default_ofhooks;
399 static uint64_t pick_datapath_id(const struct ofproto *);
400 static uint64_t pick_fallback_dpid(void);
402 static int ofproto_expire(struct ofproto *);
404 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
406 static void handle_openflow(struct ofconn *, struct ofpbuf *);
408 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
409 static void update_port(struct ofproto *, const char *devname);
410 static int init_ports(struct ofproto *);
411 static void reinit_ports(struct ofproto *);
414 ofproto_create(const char *datapath, const char *datapath_type,
415 const struct ofhooks *ofhooks, void *aux,
416 struct ofproto **ofprotop)
418 struct odp_stats stats;
425 /* Connect to datapath and start listening for messages. */
426 error = dpif_open(datapath, datapath_type, &dpif);
428 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
431 error = dpif_get_dp_stats(dpif, &stats);
433 VLOG_ERR("failed to obtain stats for datapath %s: %s",
434 datapath, strerror(error));
438 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
440 VLOG_ERR("failed to listen on datapath %s: %s",
441 datapath, strerror(error));
445 dpif_flow_flush(dpif);
446 dpif_recv_purge(dpif);
448 /* Initialize settings. */
449 p = xzalloc(sizeof *p);
450 p->fallback_dpid = pick_fallback_dpid();
451 p->datapath_id = p->fallback_dpid;
452 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
453 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
454 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
455 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
456 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
458 /* Initialize datapath. */
460 p->netdev_monitor = netdev_monitor_create();
461 hmap_init(&p->ports);
462 shash_init(&p->port_by_name);
463 p->max_ports = stats.max_ports;
465 /* Initialize submodules. */
466 p->switch_status = switch_status_create(p);
471 /* Initialize in-band control. */
473 p->in_band_queue = -1;
475 /* Initialize flow table. */
476 classifier_init(&p->cls);
477 p->next_expiration = time_msec() + 1000;
479 /* Initialize facet table. */
480 hmap_init(&p->facets);
481 p->need_revalidate = false;
482 tag_set_init(&p->revalidate_set);
484 /* Initialize OpenFlow connections. */
485 list_init(&p->all_conns);
486 hmap_init(&p->controllers);
487 hmap_init(&p->services);
491 /* Initialize hooks. */
493 p->ofhooks = ofhooks;
497 p->ofhooks = &default_ofhooks;
499 p->ml = mac_learning_create();
502 /* Pick final datapath ID. */
503 p->datapath_id = pick_datapath_id(p);
504 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
511 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
513 uint64_t old_dpid = p->datapath_id;
514 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
515 if (p->datapath_id != old_dpid) {
516 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
518 /* Force all active connections to reconnect, since there is no way to
519 * notify a controller that the datapath ID has changed. */
520 ofproto_reconnect_controllers(p);
525 is_discovery_controller(const struct ofproto_controller *c)
527 return !strcmp(c->target, "discover");
531 is_in_band_controller(const struct ofproto_controller *c)
533 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
536 /* Creates a new controller in 'ofproto'. Some of the settings are initially
537 * drawn from 'c', but update_controller() needs to be called later to finish
538 * the new ofconn's configuration. */
540 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
542 struct discovery *discovery;
543 struct ofconn *ofconn;
545 if (is_discovery_controller(c)) {
546 int error = discovery_create(c->accept_re, c->update_resolv_conf,
547 ofproto->dpif, ofproto->switch_status,
556 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
557 ofconn->pktbuf = pktbuf_create();
558 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
560 ofconn->discovery = discovery;
562 char *name = ofconn_make_name(ofproto, c->target);
563 rconn_connect(ofconn->rconn, c->target, name);
566 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
567 hash_string(c->target, 0));
570 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
571 * target or turn discovery on or off (these are done by creating new ofconns
572 * and deleting old ones), but it can update the rest of an ofconn's
575 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
579 ofconn->band = (is_in_band_controller(c)
580 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
582 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
584 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
585 rconn_set_probe_interval(ofconn->rconn, probe_interval);
587 if (ofconn->discovery) {
588 discovery_set_update_resolv_conf(ofconn->discovery,
589 c->update_resolv_conf);
590 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
593 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
597 ofconn_get_target(const struct ofconn *ofconn)
599 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
602 static struct ofconn *
603 find_controller_by_target(struct ofproto *ofproto, const char *target)
605 struct ofconn *ofconn;
607 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
608 hash_string(target, 0), &ofproto->controllers) {
609 if (!strcmp(ofconn_get_target(ofconn), target)) {
617 update_in_band_remotes(struct ofproto *ofproto)
619 const struct ofconn *ofconn;
620 struct sockaddr_in *addrs;
621 size_t max_addrs, n_addrs;
625 /* Allocate enough memory for as many remotes as we could possibly have. */
626 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
627 addrs = xmalloc(max_addrs * sizeof *addrs);
630 /* Add all the remotes. */
632 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
633 struct sockaddr_in *sin = &addrs[n_addrs];
635 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
639 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
640 if (sin->sin_addr.s_addr) {
641 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
644 if (ofconn->discovery) {
648 for (i = 0; i < ofproto->n_extra_remotes; i++) {
649 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
652 /* Create or update or destroy in-band.
654 * Ordinarily we only enable in-band if there's at least one remote
655 * address, but discovery needs the in-band rules for DHCP to be installed
656 * even before we know any remote addresses. */
657 if (n_addrs || discovery) {
658 if (!ofproto->in_band) {
659 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
662 if (ofproto->in_band) {
663 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
665 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
666 ofproto->next_in_band_update = time_msec() + 1000;
668 in_band_destroy(ofproto->in_band);
669 ofproto->in_band = NULL;
677 update_fail_open(struct ofproto *p)
679 struct ofconn *ofconn;
681 if (!hmap_is_empty(&p->controllers)
682 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
683 struct rconn **rconns;
687 p->fail_open = fail_open_create(p, p->switch_status);
691 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
692 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
693 rconns[n++] = ofconn->rconn;
696 fail_open_set_controllers(p->fail_open, rconns, n);
697 /* p->fail_open takes ownership of 'rconns'. */
699 fail_open_destroy(p->fail_open);
705 ofproto_set_controllers(struct ofproto *p,
706 const struct ofproto_controller *controllers,
707 size_t n_controllers)
709 struct shash new_controllers;
710 struct ofconn *ofconn, *next_ofconn;
711 struct ofservice *ofservice, *next_ofservice;
715 /* Create newly configured controllers and services.
716 * Create a name to ofproto_controller mapping in 'new_controllers'. */
717 shash_init(&new_controllers);
718 for (i = 0; i < n_controllers; i++) {
719 const struct ofproto_controller *c = &controllers[i];
721 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
722 if (!find_controller_by_target(p, c->target)) {
723 add_controller(p, c);
725 } else if (!pvconn_verify_name(c->target)) {
726 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
730 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
731 dpif_name(p->dpif), c->target);
735 shash_add_once(&new_controllers, c->target, &controllers[i]);
738 /* Delete controllers that are no longer configured.
739 * Update configuration of all now-existing controllers. */
741 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
742 struct ofproto_controller *c;
744 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
746 ofconn_destroy(ofconn);
748 update_controller(ofconn, c);
755 /* Delete services that are no longer configured.
756 * Update configuration of all now-existing services. */
757 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
758 struct ofproto_controller *c;
760 c = shash_find_data(&new_controllers,
761 pvconn_get_name(ofservice->pvconn));
763 ofservice_destroy(p, ofservice);
765 ofservice_reconfigure(ofservice, c);
769 shash_destroy(&new_controllers);
771 update_in_band_remotes(p);
774 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
775 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
776 struct ofconn, hmap_node);
777 ofconn->ss = switch_status_register(p->switch_status, "remote",
778 rconn_status_cb, ofconn->rconn);
783 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
785 p->fail_mode = fail_mode;
789 /* Drops the connections between 'ofproto' and all of its controllers, forcing
790 * them to reconnect. */
792 ofproto_reconnect_controllers(struct ofproto *ofproto)
794 struct ofconn *ofconn;
796 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
797 rconn_reconnect(ofconn->rconn);
802 any_extras_changed(const struct ofproto *ofproto,
803 const struct sockaddr_in *extras, size_t n)
807 if (n != ofproto->n_extra_remotes) {
811 for (i = 0; i < n; i++) {
812 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
813 const struct sockaddr_in *new = &extras[i];
815 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
816 old->sin_port != new->sin_port) {
824 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
825 * in-band control should guarantee access, in the same way that in-band
826 * control guarantees access to OpenFlow controllers. */
828 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
829 const struct sockaddr_in *extras, size_t n)
831 if (!any_extras_changed(ofproto, extras, n)) {
835 free(ofproto->extra_in_band_remotes);
836 ofproto->n_extra_remotes = n;
837 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
839 update_in_band_remotes(ofproto);
842 /* Sets the OpenFlow queue used by flows set up by in-band control on
843 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
844 * flows will use the default queue. */
846 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
848 if (queue_id != ofproto->in_band_queue) {
849 ofproto->in_band_queue = queue_id;
850 update_in_band_remotes(ofproto);
855 ofproto_set_desc(struct ofproto *p,
856 const char *mfr_desc, const char *hw_desc,
857 const char *sw_desc, const char *serial_desc,
860 struct ofp_desc_stats *ods;
863 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
864 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
865 sizeof ods->mfr_desc);
868 p->mfr_desc = xstrdup(mfr_desc);
871 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
872 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
873 sizeof ods->hw_desc);
876 p->hw_desc = xstrdup(hw_desc);
879 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
880 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
881 sizeof ods->sw_desc);
884 p->sw_desc = xstrdup(sw_desc);
887 if (strlen(serial_desc) >= sizeof ods->serial_num) {
888 VLOG_WARN("truncating serial_desc, must be less than %zu "
890 sizeof ods->serial_num);
892 free(p->serial_desc);
893 p->serial_desc = xstrdup(serial_desc);
896 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
897 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
898 sizeof ods->dp_desc);
901 p->dp_desc = xstrdup(dp_desc);
906 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
907 const struct svec *svec)
909 struct pvconn **pvconns = *pvconnsp;
910 size_t n_pvconns = *n_pvconnsp;
914 for (i = 0; i < n_pvconns; i++) {
915 pvconn_close(pvconns[i]);
919 pvconns = xmalloc(svec->n * sizeof *pvconns);
921 for (i = 0; i < svec->n; i++) {
922 const char *name = svec->names[i];
923 struct pvconn *pvconn;
926 error = pvconn_open(name, &pvconn);
928 pvconns[n_pvconns++] = pvconn;
930 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
938 *n_pvconnsp = n_pvconns;
944 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
946 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
950 ofproto_set_netflow(struct ofproto *ofproto,
951 const struct netflow_options *nf_options)
953 if (nf_options && nf_options->collectors.n) {
954 if (!ofproto->netflow) {
955 ofproto->netflow = netflow_create();
957 return netflow_set_options(ofproto->netflow, nf_options);
959 netflow_destroy(ofproto->netflow);
960 ofproto->netflow = NULL;
966 ofproto_set_sflow(struct ofproto *ofproto,
967 const struct ofproto_sflow_options *oso)
969 struct ofproto_sflow *os = ofproto->sflow;
972 struct ofport *ofport;
974 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
975 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
976 ofproto_sflow_add_port(os, ofport->odp_port,
977 netdev_get_name(ofport->netdev));
980 ofproto_sflow_set_options(os, oso);
982 ofproto_sflow_destroy(os);
983 ofproto->sflow = NULL;
988 ofproto_get_datapath_id(const struct ofproto *ofproto)
990 return ofproto->datapath_id;
994 ofproto_has_primary_controller(const struct ofproto *ofproto)
996 return !hmap_is_empty(&ofproto->controllers);
999 enum ofproto_fail_mode
1000 ofproto_get_fail_mode(const struct ofproto *p)
1002 return p->fail_mode;
1006 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1010 for (i = 0; i < ofproto->n_snoops; i++) {
1011 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1016 ofproto_destroy(struct ofproto *p)
1018 struct ofservice *ofservice, *next_ofservice;
1019 struct ofconn *ofconn, *next_ofconn;
1020 struct ofport *ofport, *next_ofport;
1027 /* Destroy fail-open and in-band early, since they touch the classifier. */
1028 fail_open_destroy(p->fail_open);
1029 p->fail_open = NULL;
1031 in_band_destroy(p->in_band);
1033 free(p->extra_in_band_remotes);
1035 ofproto_flush_flows(p);
1036 classifier_destroy(&p->cls);
1037 hmap_destroy(&p->facets);
1039 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1040 ofconn_destroy(ofconn);
1042 hmap_destroy(&p->controllers);
1044 dpif_close(p->dpif);
1045 netdev_monitor_destroy(p->netdev_monitor);
1046 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1047 hmap_remove(&p->ports, &ofport->hmap_node);
1048 ofport_free(ofport);
1050 shash_destroy(&p->port_by_name);
1052 switch_status_destroy(p->switch_status);
1053 netflow_destroy(p->netflow);
1054 ofproto_sflow_destroy(p->sflow);
1056 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1057 ofservice_destroy(p, ofservice);
1059 hmap_destroy(&p->services);
1061 for (i = 0; i < p->n_snoops; i++) {
1062 pvconn_close(p->snoops[i]);
1066 mac_learning_destroy(p->ml);
1071 free(p->serial_desc);
1074 hmap_destroy(&p->ports);
1080 ofproto_run(struct ofproto *p)
1082 int error = ofproto_run1(p);
1084 error = ofproto_run2(p, false);
1090 process_port_change(struct ofproto *ofproto, int error, char *devname)
1092 if (error == ENOBUFS) {
1093 reinit_ports(ofproto);
1094 } else if (!error) {
1095 update_port(ofproto, devname);
1100 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1101 * means that 'ofconn' is more interesting for monitoring than a lower return
1104 snoop_preference(const struct ofconn *ofconn)
1106 switch (ofconn->role) {
1107 case NX_ROLE_MASTER:
1114 /* Shouldn't happen. */
1119 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1120 * Connects this vconn to a controller. */
1122 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1124 struct ofconn *ofconn, *best;
1126 /* Pick a controller for monitoring. */
1128 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1129 if (ofconn->type == OFCONN_PRIMARY
1130 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1136 rconn_add_monitor(best->rconn, vconn);
1138 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1144 ofproto_run1(struct ofproto *p)
1146 struct ofconn *ofconn, *next_ofconn;
1147 struct ofservice *ofservice;
1152 if (shash_is_empty(&p->port_by_name)) {
1156 for (i = 0; i < 50; i++) {
1159 error = dpif_recv(p->dpif, &buf);
1161 if (error == ENODEV) {
1162 /* Someone destroyed the datapath behind our back. The caller
1163 * better destroy us and give up, because we're just going to
1164 * spin from here on out. */
1165 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1166 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1167 dpif_name(p->dpif));
1173 handle_odp_msg(p, buf);
1176 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1177 process_port_change(p, error, devname);
1179 while ((error = netdev_monitor_poll(p->netdev_monitor,
1180 &devname)) != EAGAIN) {
1181 process_port_change(p, error, devname);
1185 if (time_msec() >= p->next_in_band_update) {
1186 update_in_band_remotes(p);
1188 in_band_run(p->in_band);
1191 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1195 /* Fail-open maintenance. Do this after processing the ofconns since
1196 * fail-open checks the status of the controller rconn. */
1198 fail_open_run(p->fail_open);
1201 HMAP_FOR_EACH (ofservice, node, &p->services) {
1202 struct vconn *vconn;
1205 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1207 struct rconn *rconn;
1210 rconn = rconn_create(ofservice->probe_interval, 0);
1211 name = ofconn_make_name(p, vconn_get_name(vconn));
1212 rconn_connect_unreliably(rconn, vconn, name);
1215 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1216 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1217 ofservice->burst_limit);
1218 } else if (retval != EAGAIN) {
1219 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1223 for (i = 0; i < p->n_snoops; i++) {
1224 struct vconn *vconn;
1227 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1229 add_snooper(p, vconn);
1230 } else if (retval != EAGAIN) {
1231 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1235 if (time_msec() >= p->next_expiration) {
1236 int delay = ofproto_expire(p);
1237 p->next_expiration = time_msec() + delay;
1238 COVERAGE_INC(ofproto_expiration);
1242 netflow_run(p->netflow);
1245 ofproto_sflow_run(p->sflow);
1252 ofproto_run2(struct ofproto *p, bool revalidate_all)
1254 /* Figure out what we need to revalidate now, if anything. */
1255 struct tag_set revalidate_set = p->revalidate_set;
1256 if (p->need_revalidate) {
1257 revalidate_all = true;
1260 /* Clear the revalidation flags. */
1261 tag_set_init(&p->revalidate_set);
1262 p->need_revalidate = false;
1264 /* Now revalidate if there's anything to do. */
1265 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1266 struct facet *facet, *next;
1268 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1270 || tag_set_intersects(&revalidate_set, facet->tags)) {
1271 facet_revalidate(p, facet);
1280 ofproto_wait(struct ofproto *p)
1282 struct ofservice *ofservice;
1283 struct ofconn *ofconn;
1286 dpif_recv_wait(p->dpif);
1287 dpif_port_poll_wait(p->dpif);
1288 netdev_monitor_poll_wait(p->netdev_monitor);
1289 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1290 ofconn_wait(ofconn);
1293 poll_timer_wait_until(p->next_in_band_update);
1294 in_band_wait(p->in_band);
1297 fail_open_wait(p->fail_open);
1300 ofproto_sflow_wait(p->sflow);
1302 if (!tag_set_is_empty(&p->revalidate_set)) {
1303 poll_immediate_wake();
1305 if (p->need_revalidate) {
1306 /* Shouldn't happen, but if it does just go around again. */
1307 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1308 poll_immediate_wake();
1309 } else if (p->next_expiration != LLONG_MAX) {
1310 poll_timer_wait_until(p->next_expiration);
1312 HMAP_FOR_EACH (ofservice, node, &p->services) {
1313 pvconn_wait(ofservice->pvconn);
1315 for (i = 0; i < p->n_snoops; i++) {
1316 pvconn_wait(p->snoops[i]);
1321 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1323 tag_set_add(&ofproto->revalidate_set, tag);
1327 ofproto_get_revalidate_set(struct ofproto *ofproto)
1329 return &ofproto->revalidate_set;
1333 ofproto_is_alive(const struct ofproto *p)
1335 return !hmap_is_empty(&p->controllers);
1338 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1340 * This is almost the same as calling dpif_port_del() directly on the
1341 * datapath, but it also makes 'ofproto' close its open netdev for the port
1342 * (if any). This makes it possible to create a new netdev of a different
1343 * type under the same name, which otherwise the netdev library would refuse
1344 * to do because of the conflict. (The netdev would eventually get closed on
1345 * the next trip through ofproto_run(), but this interface is more direct.)
1347 * Returns 0 if successful, otherwise a positive errno. */
1349 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1351 struct ofport *ofport = get_port(ofproto, odp_port);
1352 const char *name = ofport ? ofport->opp.name : "<unknown>";
1355 error = dpif_port_del(ofproto->dpif, odp_port);
1357 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1358 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1359 } else if (ofport) {
1360 /* 'name' is ofport->opp.name and update_port() is going to destroy
1361 * 'ofport'. Just in case update_port() refers to 'name' after it
1362 * destroys 'ofport', make a copy of it around the update_port()
1364 char *devname = xstrdup(name);
1365 update_port(ofproto, devname);
1371 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1372 * true if 'odp_port' exists and should be included, false otherwise. */
1374 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1376 struct ofport *ofport = get_port(ofproto, odp_port);
1377 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1381 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1382 const union ofp_action *actions, size_t n_actions,
1383 const struct ofpbuf *packet)
1385 struct action_xlate_ctx ctx;
1388 action_xlate_ctx_init(&ctx, p, flow, packet);
1389 error = xlate_actions(&ctx, actions, n_actions);
1394 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1396 dpif_execute(p->dpif, ctx.out.actions, ctx.out.n_actions, packet);
1400 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1401 * performs the 'n_actions' actions in 'actions'. The new flow will not
1404 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1405 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1406 * controllers; otherwise, it will be hidden.
1408 * The caller retains ownership of 'cls_rule' and 'actions'. */
1410 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1411 const union ofp_action *actions, size_t n_actions)
1414 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1415 rule_insert(p, rule);
1419 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1423 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1426 rule_remove(ofproto, rule);
1431 ofproto_flush_flows(struct ofproto *ofproto)
1433 struct facet *facet, *next_facet;
1434 struct rule *rule, *next_rule;
1435 struct cls_cursor cursor;
1437 COVERAGE_INC(ofproto_flush);
1439 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1440 /* Mark the facet as not installed so that facet_remove() doesn't
1441 * bother trying to uninstall it. There is no point in uninstalling it
1442 * individually since we are about to blow away all the facets with
1443 * dpif_flow_flush(). */
1444 facet->installed = false;
1445 facet_remove(ofproto, facet);
1448 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1449 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1450 rule_remove(ofproto, rule);
1453 dpif_flow_flush(ofproto->dpif);
1454 if (ofproto->in_band) {
1455 in_band_flushed(ofproto->in_band);
1457 if (ofproto->fail_open) {
1458 fail_open_flushed(ofproto->fail_open);
1463 reinit_ports(struct ofproto *p)
1465 struct svec devnames;
1466 struct ofport *ofport;
1467 struct odp_port *odp_ports;
1471 COVERAGE_INC(ofproto_reinit_ports);
1473 svec_init(&devnames);
1474 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1475 svec_add (&devnames, ofport->opp.name);
1477 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1478 for (i = 0; i < n_odp_ports; i++) {
1479 svec_add (&devnames, odp_ports[i].devname);
1483 svec_sort_unique(&devnames);
1484 for (i = 0; i < devnames.n; i++) {
1485 update_port(p, devnames.names[i]);
1487 svec_destroy(&devnames);
1490 static struct ofport *
1491 make_ofport(const struct odp_port *odp_port)
1493 struct netdev_options netdev_options;
1494 enum netdev_flags flags;
1495 struct ofport *ofport;
1496 struct netdev *netdev;
1499 memset(&netdev_options, 0, sizeof netdev_options);
1500 netdev_options.name = odp_port->devname;
1501 netdev_options.type = odp_port->type;
1502 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1504 error = netdev_open(&netdev_options, &netdev);
1506 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1507 "cannot be opened (%s)",
1508 odp_port->devname, odp_port->port,
1509 odp_port->devname, strerror(error));
1513 ofport = xmalloc(sizeof *ofport);
1514 ofport->netdev = netdev;
1515 ofport->odp_port = odp_port->port;
1516 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1517 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1518 memcpy(ofport->opp.name, odp_port->devname,
1519 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1520 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1522 netdev_get_flags(netdev, &flags);
1523 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1525 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1527 netdev_get_features(netdev,
1528 &ofport->opp.curr, &ofport->opp.advertised,
1529 &ofport->opp.supported, &ofport->opp.peer);
1534 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1536 if (get_port(p, odp_port->port)) {
1537 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1540 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1541 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1550 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1552 const struct ofp_phy_port *a = &a_->opp;
1553 const struct ofp_phy_port *b = &b_->opp;
1555 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1556 return (a->port_no == b->port_no
1557 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1558 && !strcmp(a->name, b->name)
1559 && a->state == b->state
1560 && a->config == b->config
1561 && a->curr == b->curr
1562 && a->advertised == b->advertised
1563 && a->supported == b->supported
1564 && a->peer == b->peer);
1568 send_port_status(struct ofproto *p, const struct ofport *ofport,
1571 /* XXX Should limit the number of queued port status change messages. */
1572 struct ofconn *ofconn;
1573 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1574 struct ofp_port_status *ops;
1577 /* Primary controllers, even slaves, should always get port status
1578 updates. Otherwise obey ofconn_receives_async_msgs(). */
1579 if (ofconn->type != OFCONN_PRIMARY
1580 && !ofconn_receives_async_msgs(ofconn)) {
1584 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1585 ops->reason = reason;
1586 ops->desc = ofport->opp;
1587 hton_ofp_phy_port(&ops->desc);
1588 queue_tx(b, ofconn, NULL);
1593 ofport_install(struct ofproto *p, struct ofport *ofport)
1595 const char *netdev_name = ofport->opp.name;
1597 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1598 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1599 shash_add(&p->port_by_name, netdev_name, ofport);
1601 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1606 ofport_remove(struct ofproto *p, struct ofport *ofport)
1608 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1609 hmap_remove(&p->ports, &ofport->hmap_node);
1610 shash_delete(&p->port_by_name,
1611 shash_find(&p->port_by_name, ofport->opp.name));
1613 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1618 ofport_free(struct ofport *ofport)
1621 netdev_close(ofport->netdev);
1626 static struct ofport *
1627 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1629 struct ofport *port;
1631 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1632 hash_int(odp_port, 0), &ofproto->ports) {
1633 if (port->odp_port == odp_port) {
1641 update_port(struct ofproto *p, const char *devname)
1643 struct odp_port odp_port;
1644 struct ofport *old_ofport;
1645 struct ofport *new_ofport;
1648 COVERAGE_INC(ofproto_update_port);
1650 /* Query the datapath for port information. */
1651 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1653 /* Find the old ofport. */
1654 old_ofport = shash_find_data(&p->port_by_name, devname);
1657 /* There's no port named 'devname' but there might be a port with
1658 * the same port number. This could happen if a port is deleted
1659 * and then a new one added in its place very quickly, or if a port
1660 * is renamed. In the former case we want to send an OFPPR_DELETE
1661 * and an OFPPR_ADD, and in the latter case we want to send a
1662 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1663 * the old port's ifindex against the new port, or perhaps less
1664 * reliably but more portably by comparing the old port's MAC
1665 * against the new port's MAC. However, this code isn't that smart
1666 * and always sends an OFPPR_MODIFY (XXX). */
1667 old_ofport = get_port(p, odp_port.port);
1669 } else if (error != ENOENT && error != ENODEV) {
1670 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1671 "%s", strerror(error));
1675 /* Create a new ofport. */
1676 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1678 /* Eliminate a few pathological cases. */
1679 if (!old_ofport && !new_ofport) {
1681 } else if (old_ofport && new_ofport) {
1682 /* Most of the 'config' bits are OpenFlow soft state, but
1683 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1684 * OpenFlow bits from old_ofport. (make_ofport() only sets
1685 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1686 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1688 if (ofport_equal(old_ofport, new_ofport)) {
1689 /* False alarm--no change. */
1690 ofport_free(new_ofport);
1695 /* Now deal with the normal cases. */
1697 ofport_remove(p, old_ofport);
1700 ofport_install(p, new_ofport);
1702 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1703 (!old_ofport ? OFPPR_ADD
1704 : !new_ofport ? OFPPR_DELETE
1706 ofport_free(old_ofport);
1710 init_ports(struct ofproto *p)
1712 struct odp_port *ports;
1717 error = dpif_port_list(p->dpif, &ports, &n_ports);
1722 for (i = 0; i < n_ports; i++) {
1723 const struct odp_port *odp_port = &ports[i];
1724 if (!ofport_conflicts(p, odp_port)) {
1725 struct ofport *ofport = make_ofport(odp_port);
1727 ofport_install(p, ofport);
1735 static struct ofconn *
1736 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1738 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1739 ofconn->ofproto = p;
1740 list_push_back(&p->all_conns, &ofconn->node);
1741 ofconn->rconn = rconn;
1742 ofconn->type = type;
1743 ofconn->flow_format = NXFF_OPENFLOW10;
1744 ofconn->role = NX_ROLE_OTHER;
1745 ofconn->packet_in_counter = rconn_packet_counter_create ();
1746 ofconn->pktbuf = NULL;
1747 ofconn->miss_send_len = 0;
1748 ofconn->reply_counter = rconn_packet_counter_create ();
1753 ofconn_destroy(struct ofconn *ofconn)
1755 if (ofconn->type == OFCONN_PRIMARY) {
1756 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1758 discovery_destroy(ofconn->discovery);
1760 list_remove(&ofconn->node);
1761 switch_status_unregister(ofconn->ss);
1762 rconn_destroy(ofconn->rconn);
1763 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1764 rconn_packet_counter_destroy(ofconn->reply_counter);
1765 pktbuf_destroy(ofconn->pktbuf);
1770 ofconn_run(struct ofconn *ofconn)
1772 struct ofproto *p = ofconn->ofproto;
1776 if (ofconn->discovery) {
1777 char *controller_name;
1778 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1779 discovery_question_connectivity(ofconn->discovery);
1781 if (discovery_run(ofconn->discovery, &controller_name)) {
1782 if (controller_name) {
1783 char *ofconn_name = ofconn_make_name(p, controller_name);
1784 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1787 rconn_disconnect(ofconn->rconn);
1792 for (i = 0; i < N_SCHEDULERS; i++) {
1793 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1796 rconn_run(ofconn->rconn);
1798 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1799 /* Limit the number of iterations to prevent other tasks from
1801 for (iteration = 0; iteration < 50; iteration++) {
1802 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1807 fail_open_maybe_recover(p->fail_open);
1809 handle_openflow(ofconn, of_msg);
1810 ofpbuf_delete(of_msg);
1814 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1815 ofconn_destroy(ofconn);
1820 ofconn_wait(struct ofconn *ofconn)
1824 if (ofconn->discovery) {
1825 discovery_wait(ofconn->discovery);
1827 for (i = 0; i < N_SCHEDULERS; i++) {
1828 pinsched_wait(ofconn->schedulers[i]);
1830 rconn_run_wait(ofconn->rconn);
1831 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1832 rconn_recv_wait(ofconn->rconn);
1834 COVERAGE_INC(ofproto_ofconn_stuck);
1838 /* Returns true if 'ofconn' should receive asynchronous messages. */
1840 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1842 if (ofconn->type == OFCONN_PRIMARY) {
1843 /* Primary controllers always get asynchronous messages unless they
1844 * have configured themselves as "slaves". */
1845 return ofconn->role != NX_ROLE_SLAVE;
1847 /* Service connections don't get asynchronous messages unless they have
1848 * explicitly asked for them by setting a nonzero miss send length. */
1849 return ofconn->miss_send_len > 0;
1853 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1854 * and 'target', suitable for use in log messages for identifying the
1857 * The name is dynamically allocated. The caller should free it (with free())
1858 * when it is no longer needed. */
1860 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1862 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1866 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1870 for (i = 0; i < N_SCHEDULERS; i++) {
1871 struct pinsched **s = &ofconn->schedulers[i];
1875 *s = pinsched_create(rate, burst,
1876 ofconn->ofproto->switch_status);
1878 pinsched_set_limits(*s, rate, burst);
1881 pinsched_destroy(*s);
1888 ofservice_reconfigure(struct ofservice *ofservice,
1889 const struct ofproto_controller *c)
1891 ofservice->probe_interval = c->probe_interval;
1892 ofservice->rate_limit = c->rate_limit;
1893 ofservice->burst_limit = c->burst_limit;
1896 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1897 * positive errno value. */
1899 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1901 struct ofservice *ofservice;
1902 struct pvconn *pvconn;
1905 error = pvconn_open(c->target, &pvconn);
1910 ofservice = xzalloc(sizeof *ofservice);
1911 hmap_insert(&ofproto->services, &ofservice->node,
1912 hash_string(c->target, 0));
1913 ofservice->pvconn = pvconn;
1915 ofservice_reconfigure(ofservice, c);
1921 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1923 hmap_remove(&ofproto->services, &ofservice->node);
1924 pvconn_close(ofservice->pvconn);
1928 /* Finds and returns the ofservice within 'ofproto' that has the given
1929 * 'target', or a null pointer if none exists. */
1930 static struct ofservice *
1931 ofservice_lookup(struct ofproto *ofproto, const char *target)
1933 struct ofservice *ofservice;
1935 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1936 &ofproto->services) {
1937 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1944 /* Returns true if 'rule' should be hidden from the controller.
1946 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1947 * (e.g. by in-band control) and are intentionally hidden from the
1950 rule_is_hidden(const struct rule *rule)
1952 return rule->cr.priority > UINT16_MAX;
1955 /* Creates and returns a new rule initialized as specified.
1957 * The caller is responsible for inserting the rule into the classifier (with
1958 * rule_insert()). */
1959 static struct rule *
1960 rule_create(const struct cls_rule *cls_rule,
1961 const union ofp_action *actions, size_t n_actions,
1962 uint16_t idle_timeout, uint16_t hard_timeout,
1963 ovs_be64 flow_cookie, bool send_flow_removed)
1965 struct rule *rule = xzalloc(sizeof *rule);
1966 rule->cr = *cls_rule;
1967 rule->idle_timeout = idle_timeout;
1968 rule->hard_timeout = hard_timeout;
1969 rule->flow_cookie = flow_cookie;
1970 rule->used = rule->created = time_msec();
1971 rule->send_flow_removed = send_flow_removed;
1972 list_init(&rule->facets);
1973 if (n_actions > 0) {
1974 rule->n_actions = n_actions;
1975 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1981 static struct rule *
1982 rule_from_cls_rule(const struct cls_rule *cls_rule)
1984 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1988 rule_free(struct rule *rule)
1990 free(rule->actions);
1994 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1995 * destroying any that no longer has a rule (which is probably all of them).
1997 * The caller must have already removed 'rule' from the classifier. */
1999 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2001 struct facet *facet, *next_facet;
2002 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2003 facet_revalidate(ofproto, facet);
2008 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2009 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2012 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2014 const union ofp_action *oa;
2015 struct actions_iterator i;
2017 if (out_port == htons(OFPP_NONE)) {
2020 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2021 oa = actions_next(&i)) {
2022 if (action_outputs_to_port(oa, out_port)) {
2029 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2030 * 'packet', which arrived on 'in_port'.
2032 * Takes ownership of 'packet'. */
2034 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
2035 const union odp_action *actions, size_t n_actions,
2036 struct ofpbuf *packet)
2038 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
2039 /* As an optimization, avoid a round-trip from userspace to kernel to
2040 * userspace. This also avoids possibly filling up kernel packet
2041 * buffers along the way. */
2042 struct odp_msg *msg;
2044 msg = ofpbuf_push_uninit(packet, sizeof *msg);
2045 msg->type = _ODPL_ACTION_NR;
2046 msg->length = sizeof(struct odp_msg) + packet->size;
2047 msg->port = in_port;
2049 msg->arg = actions[0].controller.arg;
2051 send_packet_in(ofproto, packet);
2057 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
2058 ofpbuf_delete(packet);
2063 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2064 * statistics appropriately. 'packet' must have at least sizeof(struct
2065 * ofp_packet_in) bytes of headroom.
2067 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2068 * applying flow_extract() to 'packet' would yield the same flow as
2071 * 'facet' must have accurately composed ODP actions; that is, it must not be
2072 * in need of revalidation.
2074 * Takes ownership of 'packet'. */
2076 facet_execute(struct ofproto *ofproto, struct facet *facet,
2077 struct ofpbuf *packet)
2079 struct odp_flow_stats stats;
2081 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2083 flow_extract_stats(&facet->flow, packet, &stats);
2084 if (execute_odp_actions(ofproto, facet->flow.in_port,
2085 facet->actions, facet->n_actions, packet)) {
2086 facet_update_stats(ofproto, facet, &stats);
2087 facet->used = time_msec();
2088 netflow_flow_update_time(ofproto->netflow,
2089 &facet->nf_flow, facet->used);
2093 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2094 * statistics (or the statistics for one of its facets) appropriately.
2095 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2097 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2098 * with statistics for 'packet' either way.
2100 * Takes ownership of 'packet'. */
2102 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2103 struct ofpbuf *packet)
2105 struct action_xlate_ctx ctx;
2106 struct facet *facet;
2110 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2112 flow_extract(packet, 0, in_port, &flow);
2114 /* First look for a related facet. If we find one, account it to that. */
2115 facet = facet_lookup_valid(ofproto, &flow);
2116 if (facet && facet->rule == rule) {
2117 facet_execute(ofproto, facet, packet);
2121 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2122 * create a new facet for it and use that. */
2123 if (rule_lookup(ofproto, &flow) == rule) {
2124 facet = facet_create(ofproto, rule, &flow, packet);
2125 facet_execute(ofproto, facet, packet);
2126 facet_install(ofproto, facet, true);
2130 /* We can't account anything to a facet. If we were to try, then that
2131 * facet would have a non-matching rule, busting our invariants. */
2132 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2133 if (xlate_actions(&ctx, rule->actions, rule->n_actions)) {
2134 ofpbuf_delete(packet);
2138 size = packet->size;
2139 if (execute_odp_actions(ofproto, in_port,
2140 ctx.out.actions, ctx.out.n_actions, packet)) {
2141 rule->used = time_msec();
2142 rule->packet_count++;
2143 rule->byte_count += size;
2147 /* Inserts 'rule' into 'p''s flow table. */
2149 rule_insert(struct ofproto *p, struct rule *rule)
2151 struct rule *displaced_rule;
2153 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2154 if (displaced_rule) {
2155 rule_destroy(p, displaced_rule);
2157 p->need_revalidate = true;
2160 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2161 * 'flow' and an example 'packet' within that flow.
2163 * The caller must already have determined that no facet with an identical
2164 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2165 * 'ofproto''s classifier table. */
2166 static struct facet *
2167 facet_create(struct ofproto *ofproto, struct rule *rule,
2168 const struct flow *flow, const struct ofpbuf *packet)
2170 struct facet *facet;
2172 facet = xzalloc(sizeof *facet);
2173 facet->used = time_msec();
2174 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2175 list_push_back(&rule->facets, &facet->list_node);
2177 facet->flow = *flow;
2178 netflow_flow_init(&facet->nf_flow);
2179 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2181 facet_make_actions(ofproto, facet, packet);
2187 facet_free(struct facet *facet)
2189 free(facet->actions);
2193 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2195 * - Removes 'rule' from the classifier.
2197 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2198 * destroys them), via rule_destroy().
2201 rule_remove(struct ofproto *ofproto, struct rule *rule)
2203 COVERAGE_INC(ofproto_del_rule);
2204 ofproto->need_revalidate = true;
2205 classifier_remove(&ofproto->cls, &rule->cr);
2206 rule_destroy(ofproto, rule);
2209 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2211 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2212 * rule's statistics, via facet_uninstall().
2214 * - Removes 'facet' from its rule and from ofproto->facets.
2217 facet_remove(struct ofproto *ofproto, struct facet *facet)
2219 facet_uninstall(ofproto, facet);
2220 facet_flush_stats(ofproto, facet);
2221 hmap_remove(&ofproto->facets, &facet->hmap_node);
2222 list_remove(&facet->list_node);
2226 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2228 facet_make_actions(struct ofproto *p, struct facet *facet,
2229 const struct ofpbuf *packet)
2231 const struct rule *rule = facet->rule;
2232 struct action_xlate_ctx ctx;
2235 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2236 xlate_actions(&ctx, rule->actions, rule->n_actions);
2237 facet->tags = ctx.tags;
2238 facet->may_install = ctx.may_set_up_flow;
2239 facet->nf_flow.output_iface = ctx.nf_output_iface;
2241 actions_len = ctx.out.n_actions * sizeof *ctx.out.actions;
2242 if (facet->n_actions != ctx.out.n_actions
2243 || memcmp(facet->actions, ctx.out.actions, actions_len)) {
2244 free(facet->actions);
2245 facet->n_actions = ctx.out.n_actions;
2246 facet->actions = xmemdup(ctx.out.actions, actions_len);
2251 facet_put__(struct ofproto *ofproto, struct facet *facet, int flags,
2252 struct odp_flow_put *put)
2254 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2255 odp_flow_key_from_flow(&put->flow.key, &facet->flow);
2256 put->flow.actions = facet->actions;
2257 put->flow.n_actions = facet->n_actions;
2258 put->flow.flags = 0;
2260 return dpif_flow_put(ofproto->dpif, put);
2263 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2264 * 'zero_stats' is true, clears any existing statistics from the datapath for
2267 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2269 if (facet->may_install) {
2270 struct odp_flow_put put;
2273 flags = ODPPF_CREATE | ODPPF_MODIFY;
2275 flags |= ODPPF_ZERO_STATS;
2277 if (!facet_put__(p, facet, flags, &put)) {
2278 facet->installed = true;
2283 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2284 * to the accounting hook function in the ofhooks structure. */
2286 facet_account(struct ofproto *ofproto,
2287 struct facet *facet, uint64_t extra_bytes)
2289 uint64_t total_bytes = facet->byte_count + extra_bytes;
2291 if (ofproto->ofhooks->account_flow_cb
2292 && total_bytes > facet->accounted_bytes)
2294 ofproto->ofhooks->account_flow_cb(
2295 &facet->flow, facet->tags, facet->actions, facet->n_actions,
2296 total_bytes - facet->accounted_bytes, ofproto->aux);
2297 facet->accounted_bytes = total_bytes;
2301 /* If 'rule' is installed in the datapath, uninstalls it. */
2303 facet_uninstall(struct ofproto *p, struct facet *facet)
2305 if (facet->installed) {
2306 struct odp_flow odp_flow;
2308 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
2309 odp_flow.actions = NULL;
2310 odp_flow.n_actions = 0;
2312 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2313 facet_update_stats(p, facet, &odp_flow.stats);
2315 facet->installed = false;
2319 /* Returns true if the only action for 'facet' is to send to the controller.
2320 * (We don't report NetFlow expiration messages for such facets because they
2321 * are just part of the control logic for the network, not real traffic). */
2323 facet_is_controller_flow(struct facet *facet)
2326 && facet->rule->n_actions == 1
2327 && action_outputs_to_port(&facet->rule->actions[0],
2328 htons(OFPP_CONTROLLER)));
2331 /* Folds all of 'facet''s statistics into its rule. Also updates the
2332 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2334 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2336 facet_account(ofproto, facet, 0);
2338 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2339 struct ofexpired expired;
2340 expired.flow = facet->flow;
2341 expired.packet_count = facet->packet_count;
2342 expired.byte_count = facet->byte_count;
2343 expired.used = facet->used;
2344 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2347 facet->rule->packet_count += facet->packet_count;
2348 facet->rule->byte_count += facet->byte_count;
2350 /* Reset counters to prevent double counting if 'facet' ever gets
2352 facet->packet_count = 0;
2353 facet->byte_count = 0;
2354 facet->accounted_bytes = 0;
2356 netflow_flow_clear(&facet->nf_flow);
2359 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2360 * Returns it if found, otherwise a null pointer.
2362 * The returned facet might need revalidation; use facet_lookup_valid()
2363 * instead if that is important. */
2364 static struct facet *
2365 facet_find(struct ofproto *ofproto, const struct flow *flow)
2367 struct facet *facet;
2369 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2371 if (flow_equal(flow, &facet->flow)) {
2379 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2380 * Returns it if found, otherwise a null pointer.
2382 * The returned facet is guaranteed to be valid. */
2383 static struct facet *
2384 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2386 struct facet *facet = facet_find(ofproto, flow);
2388 /* The facet we found might not be valid, since we could be in need of
2389 * revalidation. If it is not valid, don't return it. */
2391 && ofproto->need_revalidate
2392 && !facet_revalidate(ofproto, facet)) {
2393 COVERAGE_INC(ofproto_invalidated);
2400 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2402 * - If the rule found is different from 'facet''s current rule, moves
2403 * 'facet' to the new rule and recompiles its actions.
2405 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2406 * where it is and recompiles its actions anyway.
2408 * - If there is none, destroys 'facet'.
2410 * Returns true if 'facet' still exists, false if it has been destroyed. */
2412 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2414 struct action_xlate_ctx ctx;
2415 struct rule *new_rule;
2417 bool actions_changed;
2419 COVERAGE_INC(facet_revalidate);
2421 /* Determine the new rule. */
2422 new_rule = rule_lookup(ofproto, &facet->flow);
2424 /* No new rule, so delete the facet. */
2425 facet_remove(ofproto, facet);
2429 /* Calculate new ODP actions.
2431 * We are very cautious about actually modifying 'facet' state at this
2432 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2433 * so, we need to have the old state around to properly compose it. */
2434 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2435 xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2436 actions_len = ctx.out.n_actions * sizeof *ctx.out.actions;
2437 actions_changed = (facet->n_actions != ctx.out.n_actions
2438 || memcmp(facet->actions, ctx.out.actions,
2441 /* If the ODP actions changed or the installability changed, then we need
2442 * to talk to the datapath. */
2443 if (actions_changed || facet->may_install != facet->installed) {
2444 if (facet->may_install) {
2445 struct odp_flow_put put;
2447 memset(&put.flow.stats, 0, sizeof put.flow.stats);
2448 odp_flow_key_from_flow(&put.flow.key, &facet->flow);
2449 put.flow.actions = ctx.out.actions;
2450 put.flow.n_actions = ctx.out.n_actions;
2452 put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS;
2453 dpif_flow_put(ofproto->dpif, &put);
2455 facet_update_stats(ofproto, facet, &put.flow.stats);
2457 facet_uninstall(ofproto, facet);
2460 /* The datapath flow is gone or has zeroed stats, so push stats out of
2461 * 'facet' into 'rule'. */
2462 facet_flush_stats(ofproto, facet);
2465 /* Update 'facet' now that we've taken care of all the old state. */
2466 facet->tags = ctx.tags;
2467 facet->nf_flow.output_iface = ctx.nf_output_iface;
2468 facet->may_install = ctx.may_set_up_flow;
2469 if (actions_changed) {
2470 free(facet->actions);
2471 facet->n_actions = ctx.out.n_actions;
2472 facet->actions = xmemdup(ctx.out.actions, actions_len);
2474 if (facet->rule != new_rule) {
2475 COVERAGE_INC(facet_changed_rule);
2476 list_remove(&facet->list_node);
2477 list_push_back(&new_rule->facets, &facet->list_node);
2478 facet->rule = new_rule;
2479 facet->used = new_rule->created;
2486 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2487 struct rconn_packet_counter *counter)
2489 update_openflow_length(msg);
2490 if (rconn_send(ofconn->rconn, msg, counter)) {
2496 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2499 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2501 COVERAGE_INC(ofproto_error);
2502 queue_tx(buf, ofconn, ofconn->reply_counter);
2507 hton_ofp_phy_port(struct ofp_phy_port *opp)
2509 opp->port_no = htons(opp->port_no);
2510 opp->config = htonl(opp->config);
2511 opp->state = htonl(opp->state);
2512 opp->curr = htonl(opp->curr);
2513 opp->advertised = htonl(opp->advertised);
2514 opp->supported = htonl(opp->supported);
2515 opp->peer = htonl(opp->peer);
2519 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2521 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2526 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2528 struct ofp_switch_features *osf;
2530 struct ofport *port;
2532 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2533 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2534 osf->n_buffers = htonl(pktbuf_capacity());
2536 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2537 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2538 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2539 (1u << OFPAT_SET_VLAN_VID) |
2540 (1u << OFPAT_SET_VLAN_PCP) |
2541 (1u << OFPAT_STRIP_VLAN) |
2542 (1u << OFPAT_SET_DL_SRC) |
2543 (1u << OFPAT_SET_DL_DST) |
2544 (1u << OFPAT_SET_NW_SRC) |
2545 (1u << OFPAT_SET_NW_DST) |
2546 (1u << OFPAT_SET_NW_TOS) |
2547 (1u << OFPAT_SET_TP_SRC) |
2548 (1u << OFPAT_SET_TP_DST) |
2549 (1u << OFPAT_ENQUEUE));
2551 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2552 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2555 queue_tx(buf, ofconn, ofconn->reply_counter);
2560 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2563 struct ofp_switch_config *osc;
2567 /* Figure out flags. */
2568 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2569 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2572 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2573 osc->flags = htons(flags);
2574 osc->miss_send_len = htons(ofconn->miss_send_len);
2575 queue_tx(buf, ofconn, ofconn->reply_counter);
2581 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2583 uint16_t flags = ntohs(osc->flags);
2585 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2586 switch (flags & OFPC_FRAG_MASK) {
2587 case OFPC_FRAG_NORMAL:
2588 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2590 case OFPC_FRAG_DROP:
2591 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2594 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2600 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2606 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2608 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2609 a->controller.arg = max_len;
2612 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2613 * flow translation. */
2614 #define MAX_RESUBMIT_RECURSION 8
2616 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2617 struct action_xlate_ctx *ctx);
2620 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2622 const struct ofport *ofport = get_port(ctx->ofproto, port);
2625 if (ofport->opp.config & OFPPC_NO_FWD) {
2626 /* Forwarding disabled on port. */
2631 * We don't have an ofport record for this port, but it doesn't hurt to
2632 * allow forwarding to it anyhow. Maybe such a port will appear later
2633 * and we're pre-populating the flow table.
2637 odp_actions_add(&ctx->out, ODPAT_OUTPUT)->output.port = port;
2638 ctx->nf_output_iface = port;
2641 static struct rule *
2642 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2644 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2648 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2650 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2651 uint16_t old_in_port;
2654 /* Look up a flow with 'in_port' as the input port. Then restore the
2655 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2656 * have surprising behavior). */
2657 old_in_port = ctx->flow.in_port;
2658 ctx->flow.in_port = in_port;
2659 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2660 ctx->flow.in_port = old_in_port;
2664 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2668 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2670 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2671 MAX_RESUBMIT_RECURSION);
2676 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2677 uint16_t *nf_output_iface, struct odp_actions *actions)
2679 struct ofport *ofport;
2681 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2682 uint16_t odp_port = ofport->odp_port;
2683 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2684 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2687 *nf_output_iface = NF_OUT_FLOOD;
2691 xlate_output_action__(struct action_xlate_ctx *ctx,
2692 uint16_t port, uint16_t max_len)
2695 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2697 ctx->nf_output_iface = NF_OUT_DROP;
2701 add_output_action(ctx, ctx->flow.in_port);
2704 xlate_table_action(ctx, ctx->flow.in_port);
2707 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2708 &ctx->out, &ctx->tags,
2709 &ctx->nf_output_iface,
2710 ctx->ofproto->aux)) {
2711 COVERAGE_INC(ofproto_uninstallable);
2712 ctx->may_set_up_flow = false;
2716 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2717 &ctx->nf_output_iface, &ctx->out);
2720 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2721 &ctx->nf_output_iface, &ctx->out);
2723 case OFPP_CONTROLLER:
2724 add_controller_action(&ctx->out, max_len);
2727 add_output_action(ctx, ODPP_LOCAL);
2730 odp_port = ofp_port_to_odp_port(port);
2731 if (odp_port != ctx->flow.in_port) {
2732 add_output_action(ctx, odp_port);
2737 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2738 ctx->nf_output_iface = NF_OUT_FLOOD;
2739 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2740 ctx->nf_output_iface = prev_nf_output_iface;
2741 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2742 ctx->nf_output_iface != NF_OUT_FLOOD) {
2743 ctx->nf_output_iface = NF_OUT_MULTI;
2748 xlate_output_action(struct action_xlate_ctx *ctx,
2749 const struct ofp_action_output *oao)
2751 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2754 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2755 * optimization, because we're going to add another action that sets the
2756 * priority immediately after, or because there are no actions following the
2759 remove_pop_action(struct action_xlate_ctx *ctx)
2761 size_t n = ctx->out.n_actions;
2762 if (n > 0 && ctx->out.actions[n - 1].type == ODPAT_POP_PRIORITY) {
2763 ctx->out.n_actions--;
2768 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2769 const struct ofp_action_enqueue *oae)
2771 uint16_t ofp_port, odp_port;
2775 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2778 /* Fall back to ordinary output action. */
2779 xlate_output_action__(ctx, ntohs(oae->port), 0);
2783 /* Figure out ODP output port. */
2784 ofp_port = ntohs(oae->port);
2785 if (ofp_port != OFPP_IN_PORT) {
2786 odp_port = ofp_port_to_odp_port(ofp_port);
2788 odp_port = ctx->flow.in_port;
2791 /* Add ODP actions. */
2792 remove_pop_action(ctx);
2793 odp_actions_add(&ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2795 add_output_action(ctx, odp_port);
2796 odp_actions_add(&ctx->out, ODPAT_POP_PRIORITY);
2798 /* Update NetFlow output port. */
2799 if (ctx->nf_output_iface == NF_OUT_DROP) {
2800 ctx->nf_output_iface = odp_port;
2801 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2802 ctx->nf_output_iface = NF_OUT_MULTI;
2807 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2808 const struct nx_action_set_queue *nasq)
2813 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2816 /* Couldn't translate queue to a priority, so ignore. A warning
2817 * has already been logged. */
2821 remove_pop_action(ctx);
2822 odp_actions_add(&ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2827 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2829 ovs_be16 tci = ctx->flow.vlan_tci;
2830 if (!(tci & htons(VLAN_CFI))) {
2831 odp_actions_add(&ctx->out, ODPAT_STRIP_VLAN);
2833 union odp_action *oa = odp_actions_add(&ctx->out, ODPAT_SET_DL_TCI);
2834 oa->dl_tci.tci = tci & ~htons(VLAN_CFI);
2839 xlate_reg_move_action(struct action_xlate_ctx *ctx,
2840 const struct nx_action_reg_move *narm)
2842 ovs_be16 old_tci = ctx->flow.vlan_tci;
2844 nxm_execute_reg_move(narm, &ctx->flow);
2846 if (ctx->flow.vlan_tci != old_tci) {
2847 xlate_set_dl_tci(ctx);
2852 xlate_nicira_action(struct action_xlate_ctx *ctx,
2853 const struct nx_action_header *nah)
2855 const struct nx_action_resubmit *nar;
2856 const struct nx_action_set_tunnel *nast;
2857 const struct nx_action_set_queue *nasq;
2858 union odp_action *oa;
2859 enum nx_action_subtype subtype = ntohs(nah->subtype);
2861 assert(nah->vendor == htonl(NX_VENDOR_ID));
2863 case NXAST_RESUBMIT:
2864 nar = (const struct nx_action_resubmit *) nah;
2865 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2868 case NXAST_SET_TUNNEL:
2869 nast = (const struct nx_action_set_tunnel *) nah;
2870 oa = odp_actions_add(&ctx->out, ODPAT_SET_TUNNEL);
2871 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2874 case NXAST_DROP_SPOOFED_ARP:
2875 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2876 odp_actions_add(&ctx->out, ODPAT_DROP_SPOOFED_ARP);
2880 case NXAST_SET_QUEUE:
2881 nasq = (const struct nx_action_set_queue *) nah;
2882 xlate_set_queue_action(ctx, nasq);
2885 case NXAST_POP_QUEUE:
2886 odp_actions_add(&ctx->out, ODPAT_POP_PRIORITY);
2889 case NXAST_REG_MOVE:
2890 xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah);
2893 case NXAST_REG_LOAD:
2894 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2898 /* Nothing to do. */
2901 /* If you add a new action here that modifies flow data, don't forget to
2902 * update the flow key in ctx->flow at the same time. */
2904 case NXAST_SNAT__OBSOLETE:
2906 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2912 do_xlate_actions(const union ofp_action *in, size_t n_in,
2913 struct action_xlate_ctx *ctx)
2915 struct actions_iterator iter;
2916 const union ofp_action *ia;
2917 const struct ofport *port;
2919 port = get_port(ctx->ofproto, ctx->flow.in_port);
2920 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2921 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2922 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2923 /* Drop this flow. */
2927 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2928 enum ofp_action_type type = ntohs(ia->type);
2929 union odp_action *oa;
2933 xlate_output_action(ctx, &ia->output);
2936 case OFPAT_SET_VLAN_VID:
2937 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2938 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2939 xlate_set_dl_tci(ctx);
2942 case OFPAT_SET_VLAN_PCP:
2943 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2944 ctx->flow.vlan_tci |= htons(
2945 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2946 xlate_set_dl_tci(ctx);
2949 case OFPAT_STRIP_VLAN:
2950 ctx->flow.vlan_tci = htons(0);
2951 xlate_set_dl_tci(ctx);
2954 case OFPAT_SET_DL_SRC:
2955 oa = odp_actions_add(&ctx->out, ODPAT_SET_DL_SRC);
2956 memcpy(oa->dl_addr.dl_addr,
2957 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2958 memcpy(ctx->flow.dl_src,
2959 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2962 case OFPAT_SET_DL_DST:
2963 oa = odp_actions_add(&ctx->out, ODPAT_SET_DL_DST);
2964 memcpy(oa->dl_addr.dl_addr,
2965 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2966 memcpy(ctx->flow.dl_dst,
2967 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2970 case OFPAT_SET_NW_SRC:
2971 oa = odp_actions_add(&ctx->out, ODPAT_SET_NW_SRC);
2972 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2975 case OFPAT_SET_NW_DST:
2976 oa = odp_actions_add(&ctx->out, ODPAT_SET_NW_DST);
2977 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2980 case OFPAT_SET_NW_TOS:
2981 oa = odp_actions_add(&ctx->out, ODPAT_SET_NW_TOS);
2982 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2985 case OFPAT_SET_TP_SRC:
2986 oa = odp_actions_add(&ctx->out, ODPAT_SET_TP_SRC);
2987 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2990 case OFPAT_SET_TP_DST:
2991 oa = odp_actions_add(&ctx->out, ODPAT_SET_TP_DST);
2992 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2996 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3000 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3004 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3011 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3012 struct ofproto *ofproto, const struct flow *flow,
3013 const struct ofpbuf *packet)
3015 ctx->ofproto = ofproto;
3017 ctx->packet = packet;
3021 xlate_actions(struct action_xlate_ctx *ctx,
3022 const union ofp_action *in, size_t n_in)
3024 COVERAGE_INC(ofproto_ofp2odp);
3025 odp_actions_init(&ctx->out);
3027 ctx->may_set_up_flow = true;
3028 ctx->nf_output_iface = NF_OUT_DROP;
3030 do_xlate_actions(in, n_in, ctx);
3031 remove_pop_action(ctx);
3033 /* Check with in-band control to see if we're allowed to set up this
3035 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow, &ctx->out)) {
3036 ctx->may_set_up_flow = false;
3039 if (odp_actions_overflow(&ctx->out)) {
3040 COVERAGE_INC(odp_overflow);
3041 odp_actions_init(&ctx->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, const 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 action_xlate_ctx ctx;
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. */
3088 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3089 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3092 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3093 &ofp_actions, &n_ofp_actions);
3099 if (opo->buffer_id != htonl(UINT32_MAX)) {
3100 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3102 if (error || !buffer) {
3111 /* Extract flow, check actions. */
3112 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3114 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3120 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3121 error = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3123 dpif_execute(p->dpif, ctx.out.actions, ctx.out.n_actions, &payload);
3127 ofpbuf_delete(buffer);
3132 update_port_config(struct ofproto *p, struct ofport *port,
3133 uint32_t config, uint32_t mask)
3135 mask &= config ^ port->opp.config;
3136 if (mask & OFPPC_PORT_DOWN) {
3137 if (config & OFPPC_PORT_DOWN) {
3138 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3140 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3143 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3144 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3145 if (mask & REVALIDATE_BITS) {
3146 COVERAGE_INC(ofproto_costly_flags);
3147 port->opp.config ^= mask & REVALIDATE_BITS;
3148 p->need_revalidate = true;
3150 #undef REVALIDATE_BITS
3151 if (mask & OFPPC_NO_PACKET_IN) {
3152 port->opp.config ^= OFPPC_NO_PACKET_IN;
3157 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3159 struct ofproto *p = ofconn->ofproto;
3160 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3161 struct ofport *port;
3164 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3169 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3171 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3172 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3173 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3175 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3176 if (opm->advertise) {
3177 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3183 static struct ofpbuf *
3184 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3186 struct ofp_stats_reply *osr;
3189 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3190 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3192 osr->flags = htons(0);
3196 static struct ofpbuf *
3197 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3199 const struct ofp_stats_request *osr
3200 = (const struct ofp_stats_request *) request;
3201 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3205 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3206 struct ofpbuf **msgp)
3208 struct ofpbuf *msg = *msgp;
3209 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3210 if (nbytes + msg->size > UINT16_MAX) {
3211 struct ofp_stats_reply *reply = msg->data;
3212 reply->flags = htons(OFPSF_REPLY_MORE);
3213 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3214 queue_tx(msg, ofconn, ofconn->reply_counter);
3216 return ofpbuf_put_uninit(*msgp, nbytes);
3219 static struct ofpbuf *
3220 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3222 struct nicira_stats_msg *nsm;
3225 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3226 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3227 nsm->type = htons(OFPST_VENDOR);
3228 nsm->flags = htons(0);
3229 nsm->vendor = htonl(NX_VENDOR_ID);
3230 nsm->subtype = htonl(subtype);
3234 static struct ofpbuf *
3235 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3237 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3241 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3242 struct ofpbuf **msgp)
3244 struct ofpbuf *msg = *msgp;
3245 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3246 if (nbytes + msg->size > UINT16_MAX) {
3247 struct nicira_stats_msg *reply = msg->data;
3248 reply->flags = htons(OFPSF_REPLY_MORE);
3249 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3250 queue_tx(msg, ofconn, ofconn->reply_counter);
3252 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3256 handle_desc_stats_request(struct ofconn *ofconn,
3257 const struct ofp_header *request)
3259 struct ofproto *p = ofconn->ofproto;
3260 struct ofp_desc_stats *ods;
3263 msg = start_ofp_stats_reply(request, sizeof *ods);
3264 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3265 memset(ods, 0, sizeof *ods);
3266 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3267 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3268 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3269 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3270 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3271 queue_tx(msg, ofconn, ofconn->reply_counter);
3277 handle_table_stats_request(struct ofconn *ofconn,
3278 const struct ofp_header *request)
3280 struct ofproto *p = ofconn->ofproto;
3281 struct ofp_table_stats *ots;
3284 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3286 /* Classifier table. */
3287 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3288 memset(ots, 0, sizeof *ots);
3289 strcpy(ots->name, "classifier");
3290 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3291 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3292 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3293 ots->active_count = htonl(classifier_count(&p->cls));
3294 ots->lookup_count = htonll(0); /* XXX */
3295 ots->matched_count = htonll(0); /* XXX */
3297 queue_tx(msg, ofconn, ofconn->reply_counter);
3302 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3303 struct ofpbuf **msgp)
3305 struct netdev_stats stats;
3306 struct ofp_port_stats *ops;
3308 /* Intentionally ignore return value, since errors will set
3309 * 'stats' to all-1s, which is correct for OpenFlow, and
3310 * netdev_get_stats() will log errors. */
3311 netdev_get_stats(port->netdev, &stats);
3313 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3314 ops->port_no = htons(port->opp.port_no);
3315 memset(ops->pad, 0, sizeof ops->pad);
3316 ops->rx_packets = htonll(stats.rx_packets);
3317 ops->tx_packets = htonll(stats.tx_packets);
3318 ops->rx_bytes = htonll(stats.rx_bytes);
3319 ops->tx_bytes = htonll(stats.tx_bytes);
3320 ops->rx_dropped = htonll(stats.rx_dropped);
3321 ops->tx_dropped = htonll(stats.tx_dropped);
3322 ops->rx_errors = htonll(stats.rx_errors);
3323 ops->tx_errors = htonll(stats.tx_errors);
3324 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3325 ops->rx_over_err = htonll(stats.rx_over_errors);
3326 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3327 ops->collisions = htonll(stats.collisions);
3331 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3333 struct ofproto *p = ofconn->ofproto;
3334 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3335 struct ofp_port_stats *ops;
3337 struct ofport *port;
3339 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3340 if (psr->port_no != htons(OFPP_NONE)) {
3341 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3343 append_port_stat(port, ofconn, &msg);
3346 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3347 append_port_stat(port, ofconn, &msg);
3351 queue_tx(msg, ofconn, ofconn->reply_counter);
3355 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3356 * '*packet_countp' and '*byte_countp'. The returned statistics include
3357 * statistics for all of 'rule''s facets. */
3359 query_stats(struct ofproto *p, struct rule *rule,
3360 uint64_t *packet_countp, uint64_t *byte_countp)
3362 uint64_t packet_count, byte_count;
3363 struct facet *facet;
3364 struct odp_flow *odp_flows;
3367 /* Start from historical data for 'rule' itself that are no longer tracked
3368 * by the datapath. This counts, for example, facets that have expired. */
3369 packet_count = rule->packet_count;
3370 byte_count = rule->byte_count;
3372 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3374 * Also, add any statistics that are not tracked by the datapath for each
3375 * facet. This includes, for example, statistics for packets that were
3376 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3378 odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows);
3380 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3381 struct odp_flow *odp_flow = &odp_flows[n_odp_flows++];
3382 odp_flow_key_from_flow(&odp_flow->key, &facet->flow);
3383 packet_count += facet->packet_count;
3384 byte_count += facet->byte_count;
3387 /* Fetch up-to-date statistics from the datapath and add them in. */
3388 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3391 for (i = 0; i < n_odp_flows; i++) {
3392 struct odp_flow *odp_flow = &odp_flows[i];
3393 packet_count += odp_flow->stats.n_packets;
3394 byte_count += odp_flow->stats.n_bytes;
3399 /* Return the stats to the caller. */
3400 *packet_countp = packet_count;
3401 *byte_countp = byte_count;
3405 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3407 long long int msecs = time_msec() - start;
3408 *sec = htonl(msecs / 1000);
3409 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3413 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3414 ovs_be16 out_port, struct ofpbuf **replyp)
3416 struct ofp_flow_stats *ofs;
3417 uint64_t packet_count, byte_count;
3418 size_t act_len, len;
3420 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3424 act_len = sizeof *rule->actions * rule->n_actions;
3425 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3427 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3429 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3430 ofs->length = htons(len);
3433 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match);
3434 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3435 ofs->cookie = rule->flow_cookie;
3436 ofs->priority = htons(rule->cr.priority);
3437 ofs->idle_timeout = htons(rule->idle_timeout);
3438 ofs->hard_timeout = htons(rule->hard_timeout);
3439 memset(ofs->pad2, 0, sizeof ofs->pad2);
3440 ofs->packet_count = htonll(packet_count);
3441 ofs->byte_count = htonll(byte_count);
3442 if (rule->n_actions > 0) {
3443 memcpy(ofs->actions, rule->actions, act_len);
3448 is_valid_table(uint8_t table_id)
3450 return table_id == 0 || table_id == 0xff;
3454 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3456 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3457 struct ofpbuf *reply;
3459 COVERAGE_INC(ofproto_flows_req);
3460 reply = start_ofp_stats_reply(oh, 1024);
3461 if (is_valid_table(fsr->table_id)) {
3462 struct cls_cursor cursor;
3463 struct cls_rule target;
3466 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3468 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3469 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3470 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3473 queue_tx(reply, ofconn, ofconn->reply_counter);
3479 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3480 ovs_be16 out_port, struct ofpbuf **replyp)
3482 struct nx_flow_stats *nfs;
3483 uint64_t packet_count, byte_count;
3484 size_t act_len, start_len;
3485 struct ofpbuf *reply;
3487 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3491 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3493 act_len = sizeof *rule->actions * rule->n_actions;
3495 start_len = (*replyp)->size;
3496 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3499 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3502 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3503 nfs->cookie = rule->flow_cookie;
3504 nfs->priority = htons(rule->cr.priority);
3505 nfs->idle_timeout = htons(rule->idle_timeout);
3506 nfs->hard_timeout = htons(rule->hard_timeout);
3507 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3508 memset(nfs->pad2, 0, sizeof nfs->pad2);
3509 nfs->packet_count = htonll(packet_count);
3510 nfs->byte_count = htonll(byte_count);
3511 if (rule->n_actions > 0) {
3512 ofpbuf_put(reply, rule->actions, act_len);
3514 nfs->length = htons(reply->size - start_len);
3518 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3520 struct nx_flow_stats_request *nfsr;
3521 struct cls_rule target;
3522 struct ofpbuf *reply;
3526 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3528 /* Dissect the message. */
3529 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3530 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3535 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3538 COVERAGE_INC(ofproto_flows_req);
3539 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3540 if (is_valid_table(nfsr->table_id)) {
3541 struct cls_cursor cursor;
3544 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3545 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3546 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3549 queue_tx(reply, ofconn, ofconn->reply_counter);
3555 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3557 struct ofp_match match;
3558 uint64_t packet_count, byte_count;
3559 size_t act_len = sizeof *rule->actions * rule->n_actions;
3561 query_stats(ofproto, rule, &packet_count, &byte_count);
3562 ofputil_cls_rule_to_match(&rule->cr, NXFF_OPENFLOW10, &match);
3564 ds_put_format(results, "duration=%llds, ",
3565 (time_msec() - rule->created) / 1000);
3566 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3567 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3568 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3569 ofp_print_match(results, &match, true);
3571 ofp_print_actions(results, &rule->actions->header, act_len);
3573 ds_put_cstr(results, "drop");
3575 ds_put_cstr(results, "\n");
3578 /* Adds a pretty-printed description of all flows to 'results', including
3579 * those marked hidden by secchan (e.g., by in-band control). */
3581 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3583 struct cls_cursor cursor;
3586 cls_cursor_init(&cursor, &p->cls, NULL);
3587 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3588 flow_stats_ds(p, rule, results);
3593 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3594 ovs_be16 out_port, uint8_t table_id,
3595 struct ofp_aggregate_stats_reply *oasr)
3597 uint64_t total_packets = 0;
3598 uint64_t total_bytes = 0;
3601 COVERAGE_INC(ofproto_agg_request);
3603 if (is_valid_table(table_id)) {
3604 struct cls_cursor cursor;
3607 cls_cursor_init(&cursor, &ofproto->cls, target);
3608 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3609 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3610 uint64_t packet_count;
3611 uint64_t byte_count;
3613 query_stats(ofproto, rule, &packet_count, &byte_count);
3615 total_packets += packet_count;
3616 total_bytes += byte_count;
3622 oasr->flow_count = htonl(n_flows);
3623 oasr->packet_count = htonll(total_packets);
3624 oasr->byte_count = htonll(total_bytes);
3625 memset(oasr->pad, 0, sizeof oasr->pad);
3629 handle_aggregate_stats_request(struct ofconn *ofconn,
3630 const struct ofp_header *oh)
3632 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3633 struct ofp_aggregate_stats_reply *reply;
3634 struct cls_rule target;
3637 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3640 msg = start_ofp_stats_reply(oh, sizeof *reply);
3641 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3642 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3643 request->table_id, reply);
3644 queue_tx(msg, ofconn, ofconn->reply_counter);
3649 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3651 struct nx_aggregate_stats_request *request;
3652 struct ofp_aggregate_stats_reply *reply;
3653 struct cls_rule target;
3658 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3660 /* Dissect the message. */
3661 request = ofpbuf_pull(&b, sizeof *request);
3662 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3667 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3671 COVERAGE_INC(ofproto_flows_req);
3672 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3673 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3674 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3675 request->table_id, reply);
3676 queue_tx(buf, ofconn, ofconn->reply_counter);
3681 struct queue_stats_cbdata {
3682 struct ofconn *ofconn;
3683 struct ofport *ofport;
3688 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3689 const struct netdev_queue_stats *stats)
3691 struct ofp_queue_stats *reply;
3693 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3694 reply->port_no = htons(cbdata->ofport->opp.port_no);
3695 memset(reply->pad, 0, sizeof reply->pad);
3696 reply->queue_id = htonl(queue_id);
3697 reply->tx_bytes = htonll(stats->tx_bytes);
3698 reply->tx_packets = htonll(stats->tx_packets);
3699 reply->tx_errors = htonll(stats->tx_errors);
3703 handle_queue_stats_dump_cb(uint32_t queue_id,
3704 struct netdev_queue_stats *stats,
3707 struct queue_stats_cbdata *cbdata = cbdata_;
3709 put_queue_stats(cbdata, queue_id, stats);
3713 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3714 struct queue_stats_cbdata *cbdata)
3716 cbdata->ofport = port;
3717 if (queue_id == OFPQ_ALL) {
3718 netdev_dump_queue_stats(port->netdev,
3719 handle_queue_stats_dump_cb, cbdata);
3721 struct netdev_queue_stats stats;
3723 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3724 put_queue_stats(cbdata, queue_id, &stats);
3730 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3732 struct ofproto *ofproto = ofconn->ofproto;
3733 const struct ofp_queue_stats_request *qsr;
3734 struct queue_stats_cbdata cbdata;
3735 struct ofport *port;
3736 unsigned int port_no;
3739 qsr = ofputil_stats_body(oh);
3741 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3744 COVERAGE_INC(ofproto_queue_req);
3746 cbdata.ofconn = ofconn;
3747 cbdata.msg = start_ofp_stats_reply(oh, 128);
3749 port_no = ntohs(qsr->port_no);
3750 queue_id = ntohl(qsr->queue_id);
3751 if (port_no == OFPP_ALL) {
3752 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3753 handle_queue_stats_for_port(port, queue_id, &cbdata);
3755 } else if (port_no < ofproto->max_ports) {
3756 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3758 handle_queue_stats_for_port(port, queue_id, &cbdata);
3761 ofpbuf_delete(cbdata.msg);
3762 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3764 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3769 static long long int
3770 msec_from_nsec(uint64_t sec, uint32_t nsec)
3772 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3776 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3777 const struct odp_flow_stats *stats)
3779 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3780 if (used > facet->used) {
3782 if (used > facet->rule->used) {
3783 facet->rule->used = used;
3785 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3789 /* Folds the statistics from 'stats' into the counters in 'facet'.
3791 * Because of the meaning of a facet's counters, it only makes sense to do this
3792 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3793 * packet that was sent by hand or if it represents statistics that have been
3794 * cleared out of the datapath. */
3796 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3797 const struct odp_flow_stats *stats)
3799 if (stats->n_packets) {
3800 facet_update_time(ofproto, facet, stats);
3801 facet->packet_count += stats->n_packets;
3802 facet->byte_count += stats->n_bytes;
3803 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3807 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3808 * in which no matching flow already exists in the flow table.
3810 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3811 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3812 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3814 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3817 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3819 struct ofproto *p = ofconn->ofproto;
3820 struct ofpbuf *packet;
3825 if (fm->flags & OFPFF_CHECK_OVERLAP
3826 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3827 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3831 if (fm->buffer_id != UINT32_MAX) {
3832 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3836 in_port = UINT16_MAX;
3839 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3840 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3841 fm->flags & OFPFF_SEND_FLOW_REM);
3842 rule_insert(p, rule);
3844 rule_execute(p, rule, in_port, packet);
3849 static struct rule *
3850 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3852 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3856 send_buffered_packet(struct ofconn *ofconn,
3857 struct rule *rule, uint32_t buffer_id)
3859 struct ofpbuf *packet;
3863 if (buffer_id == UINT32_MAX) {
3867 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3872 rule_execute(ofconn->ofproto, rule, in_port, packet);
3877 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3879 struct modify_flows_cbdata {
3880 struct ofproto *ofproto;
3881 const struct flow_mod *fm;
3885 static int modify_flow(struct ofproto *, const struct flow_mod *,
3888 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3889 * encoded by ofp_mkerr() on failure.
3891 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3894 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3896 struct ofproto *p = ofconn->ofproto;
3897 struct rule *match = NULL;
3898 struct cls_cursor cursor;
3901 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3902 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3903 if (!rule_is_hidden(rule)) {
3905 modify_flow(p, fm, rule);
3910 /* This credits the packet to whichever flow happened to match last.
3911 * That's weird. Maybe we should do a lookup for the flow that
3912 * actually matches the packet? Who knows. */
3913 send_buffered_packet(ofconn, match, fm->buffer_id);
3916 return add_flow(ofconn, fm);
3920 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3921 * code as encoded by ofp_mkerr() on failure.
3923 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3926 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3928 struct ofproto *p = ofconn->ofproto;
3929 struct rule *rule = find_flow_strict(p, fm);
3930 if (rule && !rule_is_hidden(rule)) {
3931 modify_flow(p, fm, rule);
3932 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3934 return add_flow(ofconn, fm);
3938 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3939 * been identified as a flow in 'p''s flow table to be modified, by changing
3940 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3941 * ofp_action[] structures). */
3943 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3945 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3947 rule->flow_cookie = fm->cookie;
3949 /* If the actions are the same, do nothing. */
3950 if (fm->n_actions == rule->n_actions
3952 || !memcmp(fm->actions, rule->actions, actions_len))) {
3956 /* Replace actions. */
3957 free(rule->actions);
3958 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3959 rule->n_actions = fm->n_actions;
3961 p->need_revalidate = true;
3966 /* OFPFC_DELETE implementation. */
3968 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3970 /* Implements OFPFC_DELETE. */
3972 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3974 struct rule *rule, *next_rule;
3975 struct cls_cursor cursor;
3977 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3978 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3979 delete_flow(p, rule, htons(fm->out_port));
3983 /* Implements OFPFC_DELETE_STRICT. */
3985 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3987 struct rule *rule = find_flow_strict(p, fm);
3989 delete_flow(p, rule, htons(fm->out_port));
3993 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3994 * been identified as a flow to delete from 'p''s flow table, by deleting the
3995 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3998 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3999 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4000 * specified 'out_port'. */
4002 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4004 if (rule_is_hidden(rule)) {
4008 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4012 rule_send_removed(p, rule, OFPRR_DELETE);
4013 rule_remove(p, rule);
4017 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4019 struct ofproto *p = ofconn->ofproto;
4023 error = reject_slave_controller(ofconn, "flow_mod");
4028 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4033 /* We do not support the emergency flow cache. It will hopefully get
4034 * dropped from OpenFlow in the near future. */
4035 if (fm.flags & OFPFF_EMERG) {
4036 /* There isn't a good fit for an error code, so just state that the
4037 * flow table is full. */
4038 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4041 error = validate_actions(fm.actions, fm.n_actions,
4042 &fm.cr.flow, p->max_ports);
4047 switch (fm.command) {
4049 return add_flow(ofconn, &fm);
4052 return modify_flows_loose(ofconn, &fm);
4054 case OFPFC_MODIFY_STRICT:
4055 return modify_flow_strict(ofconn, &fm);
4058 delete_flows_loose(p, &fm);
4061 case OFPFC_DELETE_STRICT:
4062 delete_flow_strict(p, &fm);
4066 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4071 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4073 const struct nxt_tun_id_cookie *msg
4074 = (const struct nxt_tun_id_cookie *) oh;
4076 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4081 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4083 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4084 struct nx_role_request *reply;
4088 if (ofconn->type != OFCONN_PRIMARY) {
4089 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4091 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4094 role = ntohl(nrr->role);
4095 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4096 && role != NX_ROLE_SLAVE) {
4097 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4099 /* There's no good error code for this. */
4100 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4103 if (role == NX_ROLE_MASTER) {
4104 struct ofconn *other;
4106 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4107 if (other->role == NX_ROLE_MASTER) {
4108 other->role = NX_ROLE_SLAVE;
4112 ofconn->role = role;
4114 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4115 reply->role = htonl(role);
4116 queue_tx(buf, ofconn, ofconn->reply_counter);
4122 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4124 const struct nxt_set_flow_format *msg
4125 = (const struct nxt_set_flow_format *) oh;
4128 format = ntohl(msg->format);
4129 if (format == NXFF_OPENFLOW10
4130 || format == NXFF_TUN_ID_FROM_COOKIE
4131 || format == NXFF_NXM) {
4132 ofconn->flow_format = format;
4135 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4140 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4142 struct ofp_header *ob;
4145 /* Currently, everything executes synchronously, so we can just
4146 * immediately send the barrier reply. */
4147 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4148 queue_tx(buf, ofconn, ofconn->reply_counter);
4153 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4155 const struct ofp_header *oh = msg->data;
4156 const struct ofputil_msg_type *type;
4159 error = ofputil_decode_msg_type(oh, &type);
4164 switch (ofputil_msg_type_code(type)) {
4165 /* OpenFlow requests. */
4166 case OFPUTIL_OFPT_ECHO_REQUEST:
4167 return handle_echo_request(ofconn, oh);
4169 case OFPUTIL_OFPT_FEATURES_REQUEST:
4170 return handle_features_request(ofconn, oh);
4172 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4173 return handle_get_config_request(ofconn, oh);
4175 case OFPUTIL_OFPT_SET_CONFIG:
4176 return handle_set_config(ofconn, msg->data);
4178 case OFPUTIL_OFPT_PACKET_OUT:
4179 return handle_packet_out(ofconn, oh);
4181 case OFPUTIL_OFPT_PORT_MOD:
4182 return handle_port_mod(ofconn, oh);
4184 case OFPUTIL_OFPT_FLOW_MOD:
4185 return handle_flow_mod(ofconn, oh);
4187 case OFPUTIL_OFPT_BARRIER_REQUEST:
4188 return handle_barrier_request(ofconn, oh);
4190 /* OpenFlow replies. */
4191 case OFPUTIL_OFPT_ECHO_REPLY:
4194 /* Nicira extension requests. */
4195 case OFPUTIL_NXT_STATUS_REQUEST:
4196 return switch_status_handle_request(
4197 ofconn->ofproto->switch_status, ofconn->rconn, oh);
4199 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4200 return handle_tun_id_from_cookie(ofconn, oh);
4202 case OFPUTIL_NXT_ROLE_REQUEST:
4203 return handle_role_request(ofconn, oh);
4205 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4206 return handle_nxt_set_flow_format(ofconn, oh);
4208 case OFPUTIL_NXT_FLOW_MOD:
4209 return handle_flow_mod(ofconn, oh);
4211 /* OpenFlow statistics requests. */
4212 case OFPUTIL_OFPST_DESC_REQUEST:
4213 return handle_desc_stats_request(ofconn, oh);
4215 case OFPUTIL_OFPST_FLOW_REQUEST:
4216 return handle_flow_stats_request(ofconn, oh);
4218 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4219 return handle_aggregate_stats_request(ofconn, oh);
4221 case OFPUTIL_OFPST_TABLE_REQUEST:
4222 return handle_table_stats_request(ofconn, oh);
4224 case OFPUTIL_OFPST_PORT_REQUEST:
4225 return handle_port_stats_request(ofconn, oh);
4227 case OFPUTIL_OFPST_QUEUE_REQUEST:
4228 return handle_queue_stats_request(ofconn, oh);
4230 /* Nicira extension statistics requests. */
4231 case OFPUTIL_NXST_FLOW_REQUEST:
4232 return handle_nxst_flow(ofconn, oh);
4234 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4235 return handle_nxst_aggregate(ofconn, oh);
4237 case OFPUTIL_INVALID:
4238 case OFPUTIL_OFPT_HELLO:
4239 case OFPUTIL_OFPT_ERROR:
4240 case OFPUTIL_OFPT_FEATURES_REPLY:
4241 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4242 case OFPUTIL_OFPT_PACKET_IN:
4243 case OFPUTIL_OFPT_FLOW_REMOVED:
4244 case OFPUTIL_OFPT_PORT_STATUS:
4245 case OFPUTIL_OFPT_BARRIER_REPLY:
4246 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4247 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4248 case OFPUTIL_OFPST_DESC_REPLY:
4249 case OFPUTIL_OFPST_FLOW_REPLY:
4250 case OFPUTIL_OFPST_QUEUE_REPLY:
4251 case OFPUTIL_OFPST_PORT_REPLY:
4252 case OFPUTIL_OFPST_TABLE_REPLY:
4253 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4254 case OFPUTIL_NXT_STATUS_REPLY:
4255 case OFPUTIL_NXT_ROLE_REPLY:
4256 case OFPUTIL_NXT_FLOW_REMOVED:
4257 case OFPUTIL_NXST_FLOW_REPLY:
4258 case OFPUTIL_NXST_AGGREGATE_REPLY:
4260 if (VLOG_IS_WARN_ENABLED()) {
4261 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4262 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4265 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4266 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4268 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4274 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4276 int error = handle_openflow__(ofconn, ofp_msg);
4278 send_error_oh(ofconn, ofp_msg->data, error);
4280 COVERAGE_INC(ofproto_recv_openflow);
4284 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4286 struct odp_msg *msg = packet->data;
4287 struct ofpbuf payload;
4288 struct facet *facet;
4291 ofpbuf_use_const(&payload, msg + 1, msg->length - sizeof *msg);
4292 flow_extract(&payload, msg->arg, msg->port, &flow);
4294 packet->l2 = payload.l2;
4295 packet->l3 = payload.l3;
4296 packet->l4 = payload.l4;
4297 packet->l7 = payload.l7;
4299 /* Check with in-band control to see if this packet should be sent
4300 * to the local port regardless of the flow table. */
4301 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4302 union odp_action action;
4304 memset(&action, 0, sizeof(action));
4305 action.output.type = ODPAT_OUTPUT;
4306 action.output.port = ODPP_LOCAL;
4307 dpif_execute(p->dpif, &action, 1, &payload);
4310 facet = facet_lookup_valid(p, &flow);
4312 struct rule *rule = rule_lookup(p, &flow);
4314 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4315 struct ofport *port = get_port(p, msg->port);
4317 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4318 COVERAGE_INC(ofproto_no_packet_in);
4319 /* XXX install 'drop' flow entry */
4320 ofpbuf_delete(packet);
4324 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4328 COVERAGE_INC(ofproto_packet_in);
4329 send_packet_in(p, packet);
4333 facet = facet_create(p, rule, &flow, packet);
4334 } else if (!facet->may_install) {
4335 /* The facet is not installable, that is, we need to process every
4336 * packet, so process the current packet's actions into 'facet'. */
4337 facet_make_actions(p, facet, packet);
4340 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4342 * Extra-special case for fail-open mode.
4344 * We are in fail-open mode and the packet matched the fail-open rule,
4345 * but we are connected to a controller too. We should send the packet
4346 * up to the controller in the hope that it will try to set up a flow
4347 * and thereby allow us to exit fail-open.
4349 * See the top-level comment in fail-open.c for more information.
4351 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4352 DPIF_RECV_MSG_PADDING));
4355 ofpbuf_pull(packet, sizeof *msg);
4356 facet_execute(p, facet, packet);
4357 facet_install(p, facet, false);
4361 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4363 struct odp_msg *msg = packet->data;
4365 switch (msg->type) {
4366 case _ODPL_ACTION_NR:
4367 COVERAGE_INC(ofproto_ctlr_action);
4368 send_packet_in(p, packet);
4371 case _ODPL_SFLOW_NR:
4373 ofproto_sflow_received(p->sflow, msg);
4375 ofpbuf_delete(packet);
4379 handle_odp_miss_msg(p, packet);
4383 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4389 /* Flow expiration. */
4391 static int ofproto_dp_max_idle(const struct ofproto *);
4392 static void ofproto_update_used(struct ofproto *);
4393 static void rule_expire(struct ofproto *, struct rule *);
4394 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4396 /* This function is called periodically by ofproto_run(). Its job is to
4397 * collect updates for the flows that have been installed into the datapath,
4398 * most importantly when they last were used, and then use that information to
4399 * expire flows that have not been used recently.
4401 * Returns the number of milliseconds after which it should be called again. */
4403 ofproto_expire(struct ofproto *ofproto)
4405 struct rule *rule, *next_rule;
4406 struct cls_cursor cursor;
4409 /* Update 'used' for each flow in the datapath. */
4410 ofproto_update_used(ofproto);
4412 /* Expire facets that have been idle too long. */
4413 dp_max_idle = ofproto_dp_max_idle(ofproto);
4414 ofproto_expire_facets(ofproto, dp_max_idle);
4416 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4417 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4418 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4419 rule_expire(ofproto, rule);
4422 /* Let the hook know that we're at a stable point: all outstanding data
4423 * in existing flows has been accounted to the account_cb. Thus, the
4424 * hook can now reasonably do operations that depend on having accurate
4425 * flow volume accounting (currently, that's just bond rebalancing). */
4426 if (ofproto->ofhooks->account_checkpoint_cb) {
4427 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4430 return MIN(dp_max_idle, 1000);
4433 /* Update 'used' member of installed facets. */
4435 ofproto_update_used(struct ofproto *p)
4437 struct odp_flow *flows;
4442 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4447 for (i = 0; i < n_flows; i++) {
4448 struct odp_flow *f = &flows[i];
4449 struct facet *facet;
4452 odp_flow_key_to_flow(&f->key, &flow);
4453 facet = facet_find(p, &flow);
4455 if (facet && facet->installed) {
4456 facet_update_time(p, facet, &f->stats);
4457 facet_account(p, facet, f->stats.n_bytes);
4459 /* There's a flow in the datapath that we know nothing about.
4461 COVERAGE_INC(ofproto_unexpected_rule);
4462 dpif_flow_del(p->dpif, f);
4469 /* Calculates and returns the number of milliseconds of idle time after which
4470 * facets should expire from the datapath and we should fold their statistics
4471 * into their parent rules in userspace. */
4473 ofproto_dp_max_idle(const struct ofproto *ofproto)
4476 * Idle time histogram.
4478 * Most of the time a switch has a relatively small number of facets. When
4479 * this is the case we might as well keep statistics for all of them in
4480 * userspace and to cache them in the kernel datapath for performance as
4483 * As the number of facets increases, the memory required to maintain
4484 * statistics about them in userspace and in the kernel becomes
4485 * significant. However, with a large number of facets it is likely that
4486 * only a few of them are "heavy hitters" that consume a large amount of
4487 * bandwidth. At this point, only heavy hitters are worth caching in the
4488 * kernel and maintaining in userspaces; other facets we can discard.
4490 * The technique used to compute the idle time is to build a histogram with
4491 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4492 * that is installed in the kernel gets dropped in the appropriate bucket.
4493 * After the histogram has been built, we compute the cutoff so that only
4494 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4495 * cached. At least the most-recently-used bucket of facets is kept, so
4496 * actually an arbitrary number of facets can be kept in any given
4497 * expiration run (though the next run will delete most of those unless
4498 * they receive additional data).
4500 * This requires a second pass through the facets, in addition to the pass
4501 * made by ofproto_update_used(), because the former function never looks
4502 * at uninstallable facets.
4504 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4505 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4506 int buckets[N_BUCKETS] = { 0 };
4507 struct facet *facet;
4512 total = hmap_count(&ofproto->facets);
4513 if (total <= 1000) {
4514 return N_BUCKETS * BUCKET_WIDTH;
4517 /* Build histogram. */
4519 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4520 long long int idle = now - facet->used;
4521 int bucket = (idle <= 0 ? 0
4522 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4523 : (unsigned int) idle / BUCKET_WIDTH);
4527 /* Find the first bucket whose flows should be expired. */
4528 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4529 if (buckets[bucket]) {
4532 subtotal += buckets[bucket++];
4533 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4538 if (VLOG_IS_DBG_ENABLED()) {
4542 ds_put_cstr(&s, "keep");
4543 for (i = 0; i < N_BUCKETS; i++) {
4545 ds_put_cstr(&s, ", drop");
4548 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4551 VLOG_INFO("%s: %s (msec:count)",
4552 dpif_name(ofproto->dpif), ds_cstr(&s));
4556 return bucket * BUCKET_WIDTH;
4560 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4562 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4563 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4564 struct ofexpired expired;
4565 struct odp_flow odp_flow;
4567 /* Get updated flow stats.
4569 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4570 * updated TCP flags and (2) the dpif_flow_list_all() in
4571 * ofproto_update_used() zeroed TCP flags. */
4572 memset(&odp_flow, 0, sizeof odp_flow);
4573 if (facet->installed) {
4574 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
4575 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4576 dpif_flow_get(ofproto->dpif, &odp_flow);
4578 if (odp_flow.stats.n_packets) {
4579 facet_update_time(ofproto, facet, &odp_flow.stats);
4580 netflow_flow_update_flags(&facet->nf_flow,
4581 odp_flow.stats.tcp_flags);
4585 expired.flow = facet->flow;
4586 expired.packet_count = facet->packet_count +
4587 odp_flow.stats.n_packets;
4588 expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes;
4589 expired.used = facet->used;
4591 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4596 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4598 long long int cutoff = time_msec() - dp_max_idle;
4599 struct facet *facet, *next_facet;
4601 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4602 facet_active_timeout(ofproto, facet);
4603 if (facet->used < cutoff) {
4604 facet_remove(ofproto, facet);
4609 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4610 * then delete it entirely. */
4612 rule_expire(struct ofproto *ofproto, struct rule *rule)
4614 struct facet *facet, *next_facet;
4618 /* Has 'rule' expired? */
4620 if (rule->hard_timeout
4621 && now > rule->created + rule->hard_timeout * 1000) {
4622 reason = OFPRR_HARD_TIMEOUT;
4623 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4624 && now >rule->used + rule->idle_timeout * 1000) {
4625 reason = OFPRR_IDLE_TIMEOUT;
4630 COVERAGE_INC(ofproto_expired);
4632 /* Update stats. (This is a no-op if the rule expired due to an idle
4633 * timeout, because that only happens when the rule has no facets left.) */
4634 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4635 facet_remove(ofproto, facet);
4638 /* Get rid of the rule. */
4639 if (!rule_is_hidden(rule)) {
4640 rule_send_removed(ofproto, rule, reason);
4642 rule_remove(ofproto, rule);
4645 static struct ofpbuf *
4646 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4649 struct ofp_flow_removed *ofr;
4652 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4653 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match);
4654 ofr->cookie = rule->flow_cookie;
4655 ofr->priority = htons(rule->cr.priority);
4656 ofr->reason = reason;
4657 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4658 ofr->idle_timeout = htons(rule->idle_timeout);
4659 ofr->packet_count = htonll(rule->packet_count);
4660 ofr->byte_count = htonll(rule->byte_count);
4665 static struct ofpbuf *
4666 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4668 struct nx_flow_removed *nfr;
4672 nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf);
4674 match_len = nx_put_match(buf, &rule->cr);
4676 nfr->cookie = rule->flow_cookie;
4677 nfr->priority = htons(rule->cr.priority);
4678 nfr->reason = reason;
4679 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4680 nfr->idle_timeout = htons(rule->idle_timeout);
4681 nfr->match_len = htons(match_len);
4682 nfr->packet_count = htonll(rule->packet_count);
4683 nfr->byte_count = htonll(rule->byte_count);
4689 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4691 struct ofconn *ofconn;
4693 if (!rule->send_flow_removed) {
4697 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4700 if (!rconn_is_connected(ofconn->rconn)
4701 || !ofconn_receives_async_msgs(ofconn)) {
4705 msg = (ofconn->flow_format == NXFF_NXM
4706 ? compose_nx_flow_removed(rule, reason)
4707 : compose_ofp_flow_removed(ofconn, rule, reason));
4709 /* Account flow expirations under ofconn->reply_counter, the counter
4710 * for replies to OpenFlow requests. That works because preventing
4711 * OpenFlow requests from being processed also prevents new flows from
4712 * being added (and expiring). (It also prevents processing OpenFlow
4713 * requests that would not add new flows, so it is imperfect.) */
4714 queue_tx(msg, ofconn, ofconn->reply_counter);
4718 /* pinsched callback for sending 'packet' on 'ofconn'. */
4720 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4722 struct ofconn *ofconn = ofconn_;
4724 rconn_send_with_limit(ofconn->rconn, packet,
4725 ofconn->packet_in_counter, 100);
4728 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4729 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4730 * packet scheduler for sending.
4732 * 'max_len' specifies the maximum number of bytes of the packet to send on
4733 * 'ofconn' (INT_MAX specifies no limit).
4735 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4736 * ownership is transferred to this function. */
4738 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4741 struct ofproto *ofproto = ofconn->ofproto;
4742 struct ofp_packet_in *opi = packet->data;
4743 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4744 int send_len, trim_size;
4748 if (opi->reason == OFPR_ACTION) {
4749 buffer_id = UINT32_MAX;
4750 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4751 buffer_id = pktbuf_get_null();
4752 } else if (!ofconn->pktbuf) {
4753 buffer_id = UINT32_MAX;
4755 struct ofpbuf payload;
4757 ofpbuf_use_const(&payload, opi->data,
4758 packet->size - offsetof(struct ofp_packet_in, data));
4759 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4762 /* Figure out how much of the packet to send. */
4763 send_len = ntohs(opi->total_len);
4764 if (buffer_id != UINT32_MAX) {
4765 send_len = MIN(send_len, ofconn->miss_send_len);
4767 send_len = MIN(send_len, max_len);
4769 /* Adjust packet length and clone if necessary. */
4770 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4772 packet = ofpbuf_clone_data(packet->data, trim_size);
4775 packet->size = trim_size;
4778 /* Update packet headers. */
4779 opi->buffer_id = htonl(buffer_id);
4780 update_openflow_length(packet);
4782 /* Hand over to packet scheduler. It might immediately call into
4783 * do_send_packet_in() or it might buffer it for a while (until a later
4784 * call to pinsched_run()). */
4785 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4786 packet, do_send_packet_in, ofconn);
4789 /* Replace struct odp_msg header in 'packet' by equivalent struct
4790 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4791 * returned by dpif_recv()).
4793 * The conversion is not complete: the caller still needs to trim any unneeded
4794 * payload off the end of the buffer, set the length in the OpenFlow header,
4795 * and set buffer_id. Those require us to know the controller settings and so
4796 * must be done on a per-controller basis.
4798 * Returns the maximum number of bytes of the packet that should be sent to
4799 * the controller (INT_MAX if no limit). */
4801 do_convert_to_packet_in(struct ofpbuf *packet)
4803 struct odp_msg *msg = packet->data;
4804 struct ofp_packet_in *opi;
4810 /* Extract relevant header fields */
4811 if (msg->type == _ODPL_ACTION_NR) {
4812 reason = OFPR_ACTION;
4815 reason = OFPR_NO_MATCH;
4818 total_len = msg->length - sizeof *msg;
4819 in_port = odp_port_to_ofp_port(msg->port);
4821 /* Repurpose packet buffer by overwriting header. */
4822 ofpbuf_pull(packet, sizeof(struct odp_msg));
4823 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4824 opi->header.version = OFP_VERSION;
4825 opi->header.type = OFPT_PACKET_IN;
4826 opi->total_len = htons(total_len);
4827 opi->in_port = htons(in_port);
4828 opi->reason = reason;
4833 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4834 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4835 * as necessary according to their individual configurations.
4837 * 'packet' must have sufficient headroom to convert it into a struct
4838 * ofp_packet_in (e.g. as returned by dpif_recv()).
4840 * Takes ownership of 'packet'. */
4842 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4844 struct ofconn *ofconn, *prev;
4847 max_len = do_convert_to_packet_in(packet);
4850 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4851 if (ofconn_receives_async_msgs(ofconn)) {
4853 schedule_packet_in(prev, packet, max_len, true);
4859 schedule_packet_in(prev, packet, max_len, false);
4861 ofpbuf_delete(packet);
4866 pick_datapath_id(const struct ofproto *ofproto)
4868 const struct ofport *port;
4870 port = get_port(ofproto, ODPP_LOCAL);
4872 uint8_t ea[ETH_ADDR_LEN];
4875 error = netdev_get_etheraddr(port->netdev, ea);
4877 return eth_addr_to_uint64(ea);
4879 VLOG_WARN("could not get MAC address for %s (%s)",
4880 netdev_get_name(port->netdev), strerror(error));
4882 return ofproto->fallback_dpid;
4886 pick_fallback_dpid(void)
4888 uint8_t ea[ETH_ADDR_LEN];
4889 eth_addr_nicira_random(ea);
4890 return eth_addr_to_uint64(ea);
4894 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4895 struct odp_actions *actions, tag_type *tags,
4896 uint16_t *nf_output_iface, void *ofproto_)
4898 struct ofproto *ofproto = ofproto_;
4901 /* Drop frames for reserved multicast addresses. */
4902 if (eth_addr_is_reserved(flow->dl_dst)) {
4906 /* Learn source MAC (but don't try to learn from revalidation). */
4907 if (packet != NULL) {
4908 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4910 GRAT_ARP_LOCK_NONE);
4912 /* The log messages here could actually be useful in debugging,
4913 * so keep the rate limit relatively high. */
4914 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4915 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4916 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4917 ofproto_revalidate(ofproto, rev_tag);
4921 /* Determine output port. */
4922 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4925 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4926 nf_output_iface, actions);
4927 } else if (out_port != flow->in_port) {
4928 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4929 *nf_output_iface = out_port;
4937 static const struct ofhooks default_ofhooks = {
4938 default_normal_ofhook_cb,