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"
43 #include "ofp-print.h"
45 #include "ofproto-sflow.h"
47 #include "openflow/nicira-ext.h"
48 #include "openflow/openflow.h"
49 #include "openvswitch/datapath-protocol.h"
53 #include "poll-loop.h"
57 #include "stream-ssl.h"
65 VLOG_DEFINE_THIS_MODULE(ofproto);
67 COVERAGE_DEFINE(facet_changed_rule);
68 COVERAGE_DEFINE(facet_revalidate);
69 COVERAGE_DEFINE(odp_overflow);
70 COVERAGE_DEFINE(ofproto_agg_request);
71 COVERAGE_DEFINE(ofproto_costly_flags);
72 COVERAGE_DEFINE(ofproto_ctlr_action);
73 COVERAGE_DEFINE(ofproto_del_rule);
74 COVERAGE_DEFINE(ofproto_error);
75 COVERAGE_DEFINE(ofproto_expiration);
76 COVERAGE_DEFINE(ofproto_expired);
77 COVERAGE_DEFINE(ofproto_flows_req);
78 COVERAGE_DEFINE(ofproto_flush);
79 COVERAGE_DEFINE(ofproto_invalidated);
80 COVERAGE_DEFINE(ofproto_no_packet_in);
81 COVERAGE_DEFINE(ofproto_ofconn_stuck);
82 COVERAGE_DEFINE(ofproto_ofp2odp);
83 COVERAGE_DEFINE(ofproto_packet_in);
84 COVERAGE_DEFINE(ofproto_packet_out);
85 COVERAGE_DEFINE(ofproto_queue_req);
86 COVERAGE_DEFINE(ofproto_recv_openflow);
87 COVERAGE_DEFINE(ofproto_reinit_ports);
88 COVERAGE_DEFINE(ofproto_unexpected_rule);
89 COVERAGE_DEFINE(ofproto_uninstallable);
90 COVERAGE_DEFINE(ofproto_update_port);
92 #include "sflow_api.h"
97 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
98 struct netdev *netdev;
99 struct ofp_phy_port opp; /* In host byte order. */
103 static void ofport_free(struct ofport *);
104 static void hton_ofp_phy_port(struct ofp_phy_port *);
106 struct action_xlate_ctx {
107 /* action_xlate_ctx_init() initializes these members. */
110 struct ofproto *ofproto;
112 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
113 * this flow when actions change header fields. */
116 /* The packet corresponding to 'flow', or a null pointer if we are
117 * revalidating without a packet to refer to. */
118 const struct ofpbuf *packet;
120 /* If nonnull, called just before executing a resubmit action.
122 * This is normally null so the client has to set it manually after
123 * calling action_xlate_ctx_init(). */
124 void (*resubmit_hook)(struct action_xlate_ctx *, const struct rule *);
126 /* xlate_actions() initializes and uses these members. The client might want
127 * to look at them after it returns. */
129 struct ofpbuf *odp_actions; /* Datapath actions. */
130 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
131 bool may_set_up_flow; /* True ordinarily; false if the actions must
132 * be reassessed for every packet. */
133 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
135 /* xlate_actions() initializes and uses these members, but the client has no
136 * reason to look at them. */
138 int recurse; /* Recursion level, via xlate_table_action. */
139 int last_pop_priority; /* Offset in 'odp_actions' just past most
140 * recently added ODPAT_SET_PRIORITY. */
143 static void action_xlate_ctx_init(struct action_xlate_ctx *,
144 struct ofproto *, const struct flow *,
145 const struct ofpbuf *);
146 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
147 const union ofp_action *in, size_t n_in);
149 /* An OpenFlow flow. */
151 long long int used; /* Time last used; time created if not used. */
152 long long int created; /* Creation time. */
156 * - Do include packets and bytes from facets that have been deleted or
157 * whose own statistics have been folded into the rule.
159 * - Do include packets and bytes sent "by hand" that were accounted to
160 * the rule without any facet being involved (this is a rare corner
161 * case in rule_execute()).
163 * - Do not include packet or bytes that can be obtained from any facet's
164 * packet_count or byte_count member or that can be obtained from the
165 * datapath by, e.g., dpif_flow_get() for any facet.
167 uint64_t packet_count; /* Number of packets received. */
168 uint64_t byte_count; /* Number of bytes received. */
170 ovs_be64 flow_cookie; /* Controller-issued identifier. */
172 struct cls_rule cr; /* In owning ofproto's classifier. */
173 uint16_t idle_timeout; /* In seconds from time of last use. */
174 uint16_t hard_timeout; /* In seconds from time of creation. */
175 bool send_flow_removed; /* Send a flow removed message? */
176 int n_actions; /* Number of elements in actions[]. */
177 union ofp_action *actions; /* OpenFlow actions. */
178 struct list facets; /* List of "struct facet"s. */
181 static struct rule *rule_from_cls_rule(const struct cls_rule *);
182 static bool rule_is_hidden(const struct rule *);
184 static struct rule *rule_create(const struct cls_rule *,
185 const union ofp_action *, size_t n_actions,
186 uint16_t idle_timeout, uint16_t hard_timeout,
187 ovs_be64 flow_cookie, bool send_flow_removed);
188 static void rule_destroy(struct ofproto *, struct rule *);
189 static void rule_free(struct rule *);
191 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
192 static void rule_insert(struct ofproto *, struct rule *);
193 static void rule_remove(struct ofproto *, struct rule *);
195 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
197 /* An exact-match instantiation of an OpenFlow flow. */
199 long long int used; /* Time last used; time created if not used. */
203 * - Do include packets and bytes sent "by hand", e.g. with
206 * - Do include packets and bytes that were obtained from the datapath
207 * when a flow was deleted (e.g. dpif_flow_del()) or when its
208 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
210 * - Do not include any packets or bytes that can currently be obtained
211 * from the datapath by, e.g., dpif_flow_get().
213 uint64_t packet_count; /* Number of packets received. */
214 uint64_t byte_count; /* Number of bytes received. */
216 /* Number of bytes passed to account_cb. This may include bytes that can
217 * currently obtained from the datapath (thus, it can be greater than
219 uint64_t accounted_bytes;
221 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
222 struct list list_node; /* In owning rule's 'facets' list. */
223 struct rule *rule; /* Owning rule. */
224 struct flow flow; /* Exact-match flow. */
225 bool installed; /* Installed in datapath? */
226 bool may_install; /* True ordinarily; false if actions must
227 * be reassessed for every packet. */
228 size_t actions_len; /* Number of bytes in actions[]. */
229 struct nlattr *actions; /* Datapath actions. */
230 tag_type tags; /* Tags (set only by hooks). */
231 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
234 static struct facet *facet_create(struct ofproto *, struct rule *,
236 const struct ofpbuf *packet);
237 static void facet_remove(struct ofproto *, struct facet *);
238 static void facet_free(struct facet *);
240 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
241 static bool facet_revalidate(struct ofproto *, struct facet *);
243 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
244 static void facet_uninstall(struct ofproto *, struct facet *);
245 static void facet_flush_stats(struct ofproto *, struct facet *);
247 static void facet_make_actions(struct ofproto *, struct facet *,
248 const struct ofpbuf *packet);
249 static void facet_update_stats(struct ofproto *, struct facet *,
250 const struct odp_flow_stats *);
252 /* ofproto supports two kinds of OpenFlow connections:
254 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
255 * maintains persistent connections to these controllers and by default
256 * sends them asynchronous messages such as packet-ins.
258 * - "Service" connections, e.g. from ovs-ofctl. When these connections
259 * drop, it is the other side's responsibility to reconnect them if
260 * necessary. ofproto does not send them asynchronous messages by default.
262 * Currently, active (tcp, ssl, unix) connections are always "primary"
263 * connections and passive (ptcp, pssl, punix) connections are always "service"
264 * connections. There is no inherent reason for this, but it reflects the
268 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
269 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
272 /* A listener for incoming OpenFlow "service" connections. */
274 struct hmap_node node; /* In struct ofproto's "services" hmap. */
275 struct pvconn *pvconn; /* OpenFlow connection listener. */
277 /* These are not used by ofservice directly. They are settings for
278 * accepted "struct ofconn"s from the pvconn. */
279 int probe_interval; /* Max idle time before probing, in seconds. */
280 int rate_limit; /* Max packet-in rate in packets per second. */
281 int burst_limit; /* Limit on accumulating packet credits. */
284 static struct ofservice *ofservice_lookup(struct ofproto *,
286 static int ofservice_create(struct ofproto *,
287 const struct ofproto_controller *);
288 static void ofservice_reconfigure(struct ofservice *,
289 const struct ofproto_controller *);
290 static void ofservice_destroy(struct ofproto *, struct ofservice *);
292 /* An OpenFlow connection. */
294 struct ofproto *ofproto; /* The ofproto that owns this connection. */
295 struct list node; /* In struct ofproto's "all_conns" list. */
296 struct rconn *rconn; /* OpenFlow connection. */
297 enum ofconn_type type; /* Type. */
298 enum nx_flow_format flow_format; /* Currently selected flow format. */
300 /* OFPT_PACKET_IN related data. */
301 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
302 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
303 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
304 int miss_send_len; /* Bytes to send of buffered packets. */
306 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
307 * requests, and the maximum number before we stop reading OpenFlow
309 #define OFCONN_REPLY_MAX 100
310 struct rconn_packet_counter *reply_counter;
312 /* type == OFCONN_PRIMARY only. */
313 enum nx_role role; /* Role. */
314 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
315 struct discovery *discovery; /* Controller discovery object, if enabled. */
316 struct status_category *ss; /* Switch status category. */
317 enum ofproto_band band; /* In-band or out-of-band? */
320 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
321 * "schedulers" array. Their values are 0 and 1, and their meanings and values
322 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
323 * case anything ever changes, check their values here. */
324 #define N_SCHEDULERS 2
325 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
326 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
327 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
328 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
330 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
332 static void ofconn_destroy(struct ofconn *);
333 static void ofconn_run(struct ofconn *);
334 static void ofconn_wait(struct ofconn *);
335 static bool ofconn_receives_async_msgs(const struct ofconn *);
336 static char *ofconn_make_name(const struct ofproto *, const char *target);
337 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
339 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
340 struct rconn_packet_counter *counter);
342 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
343 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
347 uint64_t datapath_id; /* Datapath ID. */
348 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
349 char *mfr_desc; /* Manufacturer. */
350 char *hw_desc; /* Hardware. */
351 char *sw_desc; /* Software version. */
352 char *serial_desc; /* Serial number. */
353 char *dp_desc; /* Datapath description. */
357 struct netdev_monitor *netdev_monitor;
358 struct hmap ports; /* Contains "struct ofport"s. */
359 struct shash port_by_name;
363 struct switch_status *switch_status;
364 struct fail_open *fail_open;
365 struct netflow *netflow;
366 struct ofproto_sflow *sflow;
368 /* In-band control. */
369 struct in_band *in_band;
370 long long int next_in_band_update;
371 struct sockaddr_in *extra_in_band_remotes;
372 size_t n_extra_remotes;
376 struct classifier cls;
377 long long int next_expiration;
381 bool need_revalidate;
382 struct tag_set revalidate_set;
384 /* OpenFlow connections. */
385 struct hmap controllers; /* Controller "struct ofconn"s. */
386 struct list all_conns; /* Contains "struct ofconn"s. */
387 enum ofproto_fail_mode fail_mode;
389 /* OpenFlow listeners. */
390 struct hmap services; /* Contains "struct ofservice"s. */
391 struct pvconn **snoops;
394 /* Hooks for ovs-vswitchd. */
395 const struct ofhooks *ofhooks;
398 /* Used by default ofhooks. */
399 struct mac_learning *ml;
402 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
403 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
405 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
407 static const struct ofhooks default_ofhooks;
409 static uint64_t pick_datapath_id(const struct ofproto *);
410 static uint64_t pick_fallback_dpid(void);
412 static int ofproto_expire(struct ofproto *);
414 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
416 static void handle_openflow(struct ofconn *, struct ofpbuf *);
418 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
419 static void update_port(struct ofproto *, const char *devname);
420 static int init_ports(struct ofproto *);
421 static void reinit_ports(struct ofproto *);
423 static void ofproto_unixctl_init(void);
426 ofproto_create(const char *datapath, const char *datapath_type,
427 const struct ofhooks *ofhooks, void *aux,
428 struct ofproto **ofprotop)
430 struct odp_stats stats;
437 ofproto_unixctl_init();
439 /* Connect to datapath and start listening for messages. */
440 error = dpif_open(datapath, datapath_type, &dpif);
442 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
445 error = dpif_get_dp_stats(dpif, &stats);
447 VLOG_ERR("failed to obtain stats for datapath %s: %s",
448 datapath, strerror(error));
452 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
454 VLOG_ERR("failed to listen on datapath %s: %s",
455 datapath, strerror(error));
459 dpif_flow_flush(dpif);
460 dpif_recv_purge(dpif);
462 /* Initialize settings. */
463 p = xzalloc(sizeof *p);
464 p->fallback_dpid = pick_fallback_dpid();
465 p->datapath_id = p->fallback_dpid;
466 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
467 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
468 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
469 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
470 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
472 /* Initialize datapath. */
474 p->netdev_monitor = netdev_monitor_create();
475 hmap_init(&p->ports);
476 shash_init(&p->port_by_name);
477 p->max_ports = stats.max_ports;
479 /* Initialize submodules. */
480 p->switch_status = switch_status_create(p);
485 /* Initialize in-band control. */
487 p->in_band_queue = -1;
489 /* Initialize flow table. */
490 classifier_init(&p->cls);
491 p->next_expiration = time_msec() + 1000;
493 /* Initialize facet table. */
494 hmap_init(&p->facets);
495 p->need_revalidate = false;
496 tag_set_init(&p->revalidate_set);
498 /* Initialize OpenFlow connections. */
499 list_init(&p->all_conns);
500 hmap_init(&p->controllers);
501 hmap_init(&p->services);
505 /* Initialize hooks. */
507 p->ofhooks = ofhooks;
511 p->ofhooks = &default_ofhooks;
513 p->ml = mac_learning_create();
516 /* Pick final datapath ID. */
517 p->datapath_id = pick_datapath_id(p);
518 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
520 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
527 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
529 uint64_t old_dpid = p->datapath_id;
530 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
531 if (p->datapath_id != old_dpid) {
532 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
534 /* Force all active connections to reconnect, since there is no way to
535 * notify a controller that the datapath ID has changed. */
536 ofproto_reconnect_controllers(p);
541 is_discovery_controller(const struct ofproto_controller *c)
543 return !strcmp(c->target, "discover");
547 is_in_band_controller(const struct ofproto_controller *c)
549 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
552 /* Creates a new controller in 'ofproto'. Some of the settings are initially
553 * drawn from 'c', but update_controller() needs to be called later to finish
554 * the new ofconn's configuration. */
556 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
558 struct discovery *discovery;
559 struct ofconn *ofconn;
561 if (is_discovery_controller(c)) {
562 int error = discovery_create(c->accept_re, c->update_resolv_conf,
563 ofproto->dpif, ofproto->switch_status,
572 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
573 ofconn->pktbuf = pktbuf_create();
574 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
576 ofconn->discovery = discovery;
578 char *name = ofconn_make_name(ofproto, c->target);
579 rconn_connect(ofconn->rconn, c->target, name);
582 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
583 hash_string(c->target, 0));
586 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
587 * target or turn discovery on or off (these are done by creating new ofconns
588 * and deleting old ones), but it can update the rest of an ofconn's
591 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
595 ofconn->band = (is_in_band_controller(c)
596 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
598 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
600 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
601 rconn_set_probe_interval(ofconn->rconn, probe_interval);
603 if (ofconn->discovery) {
604 discovery_set_update_resolv_conf(ofconn->discovery,
605 c->update_resolv_conf);
606 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
609 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
613 ofconn_get_target(const struct ofconn *ofconn)
615 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
618 static struct ofconn *
619 find_controller_by_target(struct ofproto *ofproto, const char *target)
621 struct ofconn *ofconn;
623 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
624 hash_string(target, 0), &ofproto->controllers) {
625 if (!strcmp(ofconn_get_target(ofconn), target)) {
633 update_in_band_remotes(struct ofproto *ofproto)
635 const struct ofconn *ofconn;
636 struct sockaddr_in *addrs;
637 size_t max_addrs, n_addrs;
641 /* Allocate enough memory for as many remotes as we could possibly have. */
642 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
643 addrs = xmalloc(max_addrs * sizeof *addrs);
646 /* Add all the remotes. */
648 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
649 struct sockaddr_in *sin = &addrs[n_addrs];
651 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
655 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
656 if (sin->sin_addr.s_addr) {
657 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
660 if (ofconn->discovery) {
664 for (i = 0; i < ofproto->n_extra_remotes; i++) {
665 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
668 /* Create or update or destroy in-band.
670 * Ordinarily we only enable in-band if there's at least one remote
671 * address, but discovery needs the in-band rules for DHCP to be installed
672 * even before we know any remote addresses. */
673 if (n_addrs || discovery) {
674 if (!ofproto->in_band) {
675 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
678 if (ofproto->in_band) {
679 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
681 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
682 ofproto->next_in_band_update = time_msec() + 1000;
684 in_band_destroy(ofproto->in_band);
685 ofproto->in_band = NULL;
693 update_fail_open(struct ofproto *p)
695 struct ofconn *ofconn;
697 if (!hmap_is_empty(&p->controllers)
698 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
699 struct rconn **rconns;
703 p->fail_open = fail_open_create(p, p->switch_status);
707 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
708 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
709 rconns[n++] = ofconn->rconn;
712 fail_open_set_controllers(p->fail_open, rconns, n);
713 /* p->fail_open takes ownership of 'rconns'. */
715 fail_open_destroy(p->fail_open);
721 ofproto_set_controllers(struct ofproto *p,
722 const struct ofproto_controller *controllers,
723 size_t n_controllers)
725 struct shash new_controllers;
726 struct ofconn *ofconn, *next_ofconn;
727 struct ofservice *ofservice, *next_ofservice;
731 /* Create newly configured controllers and services.
732 * Create a name to ofproto_controller mapping in 'new_controllers'. */
733 shash_init(&new_controllers);
734 for (i = 0; i < n_controllers; i++) {
735 const struct ofproto_controller *c = &controllers[i];
737 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
738 if (!find_controller_by_target(p, c->target)) {
739 add_controller(p, c);
741 } else if (!pvconn_verify_name(c->target)) {
742 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
746 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
747 dpif_name(p->dpif), c->target);
751 shash_add_once(&new_controllers, c->target, &controllers[i]);
754 /* Delete controllers that are no longer configured.
755 * Update configuration of all now-existing controllers. */
757 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
758 struct ofproto_controller *c;
760 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
762 ofconn_destroy(ofconn);
764 update_controller(ofconn, c);
771 /* Delete services that are no longer configured.
772 * Update configuration of all now-existing services. */
773 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
774 struct ofproto_controller *c;
776 c = shash_find_data(&new_controllers,
777 pvconn_get_name(ofservice->pvconn));
779 ofservice_destroy(p, ofservice);
781 ofservice_reconfigure(ofservice, c);
785 shash_destroy(&new_controllers);
787 update_in_band_remotes(p);
790 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
791 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
792 struct ofconn, hmap_node);
793 ofconn->ss = switch_status_register(p->switch_status, "remote",
794 rconn_status_cb, ofconn->rconn);
799 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
801 p->fail_mode = fail_mode;
805 /* Drops the connections between 'ofproto' and all of its controllers, forcing
806 * them to reconnect. */
808 ofproto_reconnect_controllers(struct ofproto *ofproto)
810 struct ofconn *ofconn;
812 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
813 rconn_reconnect(ofconn->rconn);
818 any_extras_changed(const struct ofproto *ofproto,
819 const struct sockaddr_in *extras, size_t n)
823 if (n != ofproto->n_extra_remotes) {
827 for (i = 0; i < n; i++) {
828 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
829 const struct sockaddr_in *new = &extras[i];
831 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
832 old->sin_port != new->sin_port) {
840 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
841 * in-band control should guarantee access, in the same way that in-band
842 * control guarantees access to OpenFlow controllers. */
844 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
845 const struct sockaddr_in *extras, size_t n)
847 if (!any_extras_changed(ofproto, extras, n)) {
851 free(ofproto->extra_in_band_remotes);
852 ofproto->n_extra_remotes = n;
853 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
855 update_in_band_remotes(ofproto);
858 /* Sets the OpenFlow queue used by flows set up by in-band control on
859 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
860 * flows will use the default queue. */
862 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
864 if (queue_id != ofproto->in_band_queue) {
865 ofproto->in_band_queue = queue_id;
866 update_in_band_remotes(ofproto);
871 ofproto_set_desc(struct ofproto *p,
872 const char *mfr_desc, const char *hw_desc,
873 const char *sw_desc, const char *serial_desc,
876 struct ofp_desc_stats *ods;
879 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
880 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
881 sizeof ods->mfr_desc);
884 p->mfr_desc = xstrdup(mfr_desc);
887 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
888 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
889 sizeof ods->hw_desc);
892 p->hw_desc = xstrdup(hw_desc);
895 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
896 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
897 sizeof ods->sw_desc);
900 p->sw_desc = xstrdup(sw_desc);
903 if (strlen(serial_desc) >= sizeof ods->serial_num) {
904 VLOG_WARN("truncating serial_desc, must be less than %zu "
906 sizeof ods->serial_num);
908 free(p->serial_desc);
909 p->serial_desc = xstrdup(serial_desc);
912 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
913 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
914 sizeof ods->dp_desc);
917 p->dp_desc = xstrdup(dp_desc);
922 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
923 const struct svec *svec)
925 struct pvconn **pvconns = *pvconnsp;
926 size_t n_pvconns = *n_pvconnsp;
930 for (i = 0; i < n_pvconns; i++) {
931 pvconn_close(pvconns[i]);
935 pvconns = xmalloc(svec->n * sizeof *pvconns);
937 for (i = 0; i < svec->n; i++) {
938 const char *name = svec->names[i];
939 struct pvconn *pvconn;
942 error = pvconn_open(name, &pvconn);
944 pvconns[n_pvconns++] = pvconn;
946 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
954 *n_pvconnsp = n_pvconns;
960 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
962 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
966 ofproto_set_netflow(struct ofproto *ofproto,
967 const struct netflow_options *nf_options)
969 if (nf_options && nf_options->collectors.n) {
970 if (!ofproto->netflow) {
971 ofproto->netflow = netflow_create();
973 return netflow_set_options(ofproto->netflow, nf_options);
975 netflow_destroy(ofproto->netflow);
976 ofproto->netflow = NULL;
982 ofproto_set_sflow(struct ofproto *ofproto,
983 const struct ofproto_sflow_options *oso)
985 struct ofproto_sflow *os = ofproto->sflow;
988 struct ofport *ofport;
990 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
991 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
992 ofproto_sflow_add_port(os, ofport->odp_port,
993 netdev_get_name(ofport->netdev));
996 ofproto_sflow_set_options(os, oso);
998 ofproto_sflow_destroy(os);
999 ofproto->sflow = NULL;
1004 ofproto_get_datapath_id(const struct ofproto *ofproto)
1006 return ofproto->datapath_id;
1010 ofproto_has_primary_controller(const struct ofproto *ofproto)
1012 return !hmap_is_empty(&ofproto->controllers);
1015 enum ofproto_fail_mode
1016 ofproto_get_fail_mode(const struct ofproto *p)
1018 return p->fail_mode;
1022 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1026 for (i = 0; i < ofproto->n_snoops; i++) {
1027 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1032 ofproto_destroy(struct ofproto *p)
1034 struct ofservice *ofservice, *next_ofservice;
1035 struct ofconn *ofconn, *next_ofconn;
1036 struct ofport *ofport, *next_ofport;
1043 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1045 /* Destroy fail-open and in-band early, since they touch the classifier. */
1046 fail_open_destroy(p->fail_open);
1047 p->fail_open = NULL;
1049 in_band_destroy(p->in_band);
1051 free(p->extra_in_band_remotes);
1053 ofproto_flush_flows(p);
1054 classifier_destroy(&p->cls);
1055 hmap_destroy(&p->facets);
1057 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1058 ofconn_destroy(ofconn);
1060 hmap_destroy(&p->controllers);
1062 dpif_close(p->dpif);
1063 netdev_monitor_destroy(p->netdev_monitor);
1064 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1065 hmap_remove(&p->ports, &ofport->hmap_node);
1066 ofport_free(ofport);
1068 shash_destroy(&p->port_by_name);
1070 switch_status_destroy(p->switch_status);
1071 netflow_destroy(p->netflow);
1072 ofproto_sflow_destroy(p->sflow);
1074 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1075 ofservice_destroy(p, ofservice);
1077 hmap_destroy(&p->services);
1079 for (i = 0; i < p->n_snoops; i++) {
1080 pvconn_close(p->snoops[i]);
1084 mac_learning_destroy(p->ml);
1089 free(p->serial_desc);
1092 hmap_destroy(&p->ports);
1098 ofproto_run(struct ofproto *p)
1100 int error = ofproto_run1(p);
1102 error = ofproto_run2(p, false);
1108 process_port_change(struct ofproto *ofproto, int error, char *devname)
1110 if (error == ENOBUFS) {
1111 reinit_ports(ofproto);
1112 } else if (!error) {
1113 update_port(ofproto, devname);
1118 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1119 * means that 'ofconn' is more interesting for monitoring than a lower return
1122 snoop_preference(const struct ofconn *ofconn)
1124 switch (ofconn->role) {
1125 case NX_ROLE_MASTER:
1132 /* Shouldn't happen. */
1137 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1138 * Connects this vconn to a controller. */
1140 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1142 struct ofconn *ofconn, *best;
1144 /* Pick a controller for monitoring. */
1146 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1147 if (ofconn->type == OFCONN_PRIMARY
1148 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1154 rconn_add_monitor(best->rconn, vconn);
1156 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1162 ofproto_run1(struct ofproto *p)
1164 struct ofconn *ofconn, *next_ofconn;
1165 struct ofservice *ofservice;
1170 if (shash_is_empty(&p->port_by_name)) {
1174 for (i = 0; i < 50; i++) {
1177 error = dpif_recv(p->dpif, &buf);
1179 if (error == ENODEV) {
1180 /* Someone destroyed the datapath behind our back. The caller
1181 * better destroy us and give up, because we're just going to
1182 * spin from here on out. */
1183 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1184 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1185 dpif_name(p->dpif));
1191 handle_odp_msg(p, buf);
1194 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1195 process_port_change(p, error, devname);
1197 while ((error = netdev_monitor_poll(p->netdev_monitor,
1198 &devname)) != EAGAIN) {
1199 process_port_change(p, error, devname);
1203 if (time_msec() >= p->next_in_band_update) {
1204 update_in_band_remotes(p);
1206 in_band_run(p->in_band);
1209 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1213 /* Fail-open maintenance. Do this after processing the ofconns since
1214 * fail-open checks the status of the controller rconn. */
1216 fail_open_run(p->fail_open);
1219 HMAP_FOR_EACH (ofservice, node, &p->services) {
1220 struct vconn *vconn;
1223 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1225 struct rconn *rconn;
1228 rconn = rconn_create(ofservice->probe_interval, 0);
1229 name = ofconn_make_name(p, vconn_get_name(vconn));
1230 rconn_connect_unreliably(rconn, vconn, name);
1233 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1234 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1235 ofservice->burst_limit);
1236 } else if (retval != EAGAIN) {
1237 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1241 for (i = 0; i < p->n_snoops; i++) {
1242 struct vconn *vconn;
1245 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1247 add_snooper(p, vconn);
1248 } else if (retval != EAGAIN) {
1249 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1253 if (time_msec() >= p->next_expiration) {
1254 int delay = ofproto_expire(p);
1255 p->next_expiration = time_msec() + delay;
1256 COVERAGE_INC(ofproto_expiration);
1260 netflow_run(p->netflow);
1263 ofproto_sflow_run(p->sflow);
1270 ofproto_run2(struct ofproto *p, bool revalidate_all)
1272 /* Figure out what we need to revalidate now, if anything. */
1273 struct tag_set revalidate_set = p->revalidate_set;
1274 if (p->need_revalidate) {
1275 revalidate_all = true;
1278 /* Clear the revalidation flags. */
1279 tag_set_init(&p->revalidate_set);
1280 p->need_revalidate = false;
1282 /* Now revalidate if there's anything to do. */
1283 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1284 struct facet *facet, *next;
1286 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1288 || tag_set_intersects(&revalidate_set, facet->tags)) {
1289 facet_revalidate(p, facet);
1298 ofproto_wait(struct ofproto *p)
1300 struct ofservice *ofservice;
1301 struct ofconn *ofconn;
1304 dpif_recv_wait(p->dpif);
1305 dpif_port_poll_wait(p->dpif);
1306 netdev_monitor_poll_wait(p->netdev_monitor);
1307 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1308 ofconn_wait(ofconn);
1311 poll_timer_wait_until(p->next_in_band_update);
1312 in_band_wait(p->in_band);
1315 fail_open_wait(p->fail_open);
1318 ofproto_sflow_wait(p->sflow);
1320 if (!tag_set_is_empty(&p->revalidate_set)) {
1321 poll_immediate_wake();
1323 if (p->need_revalidate) {
1324 /* Shouldn't happen, but if it does just go around again. */
1325 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1326 poll_immediate_wake();
1327 } else if (p->next_expiration != LLONG_MAX) {
1328 poll_timer_wait_until(p->next_expiration);
1330 HMAP_FOR_EACH (ofservice, node, &p->services) {
1331 pvconn_wait(ofservice->pvconn);
1333 for (i = 0; i < p->n_snoops; i++) {
1334 pvconn_wait(p->snoops[i]);
1339 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1341 tag_set_add(&ofproto->revalidate_set, tag);
1345 ofproto_get_revalidate_set(struct ofproto *ofproto)
1347 return &ofproto->revalidate_set;
1351 ofproto_is_alive(const struct ofproto *p)
1353 return !hmap_is_empty(&p->controllers);
1356 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1358 * This is almost the same as calling dpif_port_del() directly on the
1359 * datapath, but it also makes 'ofproto' close its open netdev for the port
1360 * (if any). This makes it possible to create a new netdev of a different
1361 * type under the same name, which otherwise the netdev library would refuse
1362 * to do because of the conflict. (The netdev would eventually get closed on
1363 * the next trip through ofproto_run(), but this interface is more direct.)
1365 * Returns 0 if successful, otherwise a positive errno. */
1367 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1369 struct ofport *ofport = get_port(ofproto, odp_port);
1370 const char *name = ofport ? ofport->opp.name : "<unknown>";
1373 error = dpif_port_del(ofproto->dpif, odp_port);
1375 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1376 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1377 } else if (ofport) {
1378 /* 'name' is ofport->opp.name and update_port() is going to destroy
1379 * 'ofport'. Just in case update_port() refers to 'name' after it
1380 * destroys 'ofport', make a copy of it around the update_port()
1382 char *devname = xstrdup(name);
1383 update_port(ofproto, devname);
1389 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1390 * true if 'odp_port' exists and should be included, false otherwise. */
1392 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1394 struct ofport *ofport = get_port(ofproto, odp_port);
1395 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1399 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1400 const union ofp_action *actions, size_t n_actions,
1401 const struct ofpbuf *packet)
1403 struct action_xlate_ctx ctx;
1404 struct ofpbuf *odp_actions;
1406 action_xlate_ctx_init(&ctx, p, flow, packet);
1407 odp_actions = xlate_actions(&ctx, actions, n_actions);
1409 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1411 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
1413 ofpbuf_delete(odp_actions);
1418 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1419 * performs the 'n_actions' actions in 'actions'. The new flow will not
1422 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1423 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1424 * controllers; otherwise, it will be hidden.
1426 * The caller retains ownership of 'cls_rule' and 'actions'. */
1428 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1429 const union ofp_action *actions, size_t n_actions)
1432 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1433 rule_insert(p, rule);
1437 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1441 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1444 rule_remove(ofproto, rule);
1449 ofproto_flush_flows(struct ofproto *ofproto)
1451 struct facet *facet, *next_facet;
1452 struct rule *rule, *next_rule;
1453 struct cls_cursor cursor;
1455 COVERAGE_INC(ofproto_flush);
1457 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1458 /* Mark the facet as not installed so that facet_remove() doesn't
1459 * bother trying to uninstall it. There is no point in uninstalling it
1460 * individually since we are about to blow away all the facets with
1461 * dpif_flow_flush(). */
1462 facet->installed = false;
1463 facet_remove(ofproto, facet);
1466 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1467 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1468 rule_remove(ofproto, rule);
1471 dpif_flow_flush(ofproto->dpif);
1472 if (ofproto->in_band) {
1473 in_band_flushed(ofproto->in_band);
1475 if (ofproto->fail_open) {
1476 fail_open_flushed(ofproto->fail_open);
1481 reinit_ports(struct ofproto *p)
1483 struct svec devnames;
1484 struct ofport *ofport;
1485 struct odp_port *odp_ports;
1489 COVERAGE_INC(ofproto_reinit_ports);
1491 svec_init(&devnames);
1492 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1493 svec_add (&devnames, ofport->opp.name);
1495 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1496 for (i = 0; i < n_odp_ports; i++) {
1497 svec_add (&devnames, odp_ports[i].devname);
1501 svec_sort_unique(&devnames);
1502 for (i = 0; i < devnames.n; i++) {
1503 update_port(p, devnames.names[i]);
1505 svec_destroy(&devnames);
1508 static struct ofport *
1509 make_ofport(const struct odp_port *odp_port)
1511 struct netdev_options netdev_options;
1512 enum netdev_flags flags;
1513 struct ofport *ofport;
1514 struct netdev *netdev;
1517 memset(&netdev_options, 0, sizeof netdev_options);
1518 netdev_options.name = odp_port->devname;
1519 netdev_options.type = odp_port->type;
1520 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1522 error = netdev_open(&netdev_options, &netdev);
1524 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1525 "cannot be opened (%s)",
1526 odp_port->devname, odp_port->port,
1527 odp_port->devname, strerror(error));
1531 ofport = xmalloc(sizeof *ofport);
1532 ofport->netdev = netdev;
1533 ofport->odp_port = odp_port->port;
1534 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1535 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1536 memcpy(ofport->opp.name, odp_port->devname,
1537 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1538 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1540 netdev_get_flags(netdev, &flags);
1541 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1543 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1545 netdev_get_features(netdev,
1546 &ofport->opp.curr, &ofport->opp.advertised,
1547 &ofport->opp.supported, &ofport->opp.peer);
1552 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1554 if (get_port(p, odp_port->port)) {
1555 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1558 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1559 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1568 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1570 const struct ofp_phy_port *a = &a_->opp;
1571 const struct ofp_phy_port *b = &b_->opp;
1573 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1574 return (a->port_no == b->port_no
1575 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1576 && !strcmp(a->name, b->name)
1577 && a->state == b->state
1578 && a->config == b->config
1579 && a->curr == b->curr
1580 && a->advertised == b->advertised
1581 && a->supported == b->supported
1582 && a->peer == b->peer);
1586 send_port_status(struct ofproto *p, const struct ofport *ofport,
1589 /* XXX Should limit the number of queued port status change messages. */
1590 struct ofconn *ofconn;
1591 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1592 struct ofp_port_status *ops;
1595 /* Primary controllers, even slaves, should always get port status
1596 updates. Otherwise obey ofconn_receives_async_msgs(). */
1597 if (ofconn->type != OFCONN_PRIMARY
1598 && !ofconn_receives_async_msgs(ofconn)) {
1602 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1603 ops->reason = reason;
1604 ops->desc = ofport->opp;
1605 hton_ofp_phy_port(&ops->desc);
1606 queue_tx(b, ofconn, NULL);
1611 ofport_install(struct ofproto *p, struct ofport *ofport)
1613 const char *netdev_name = ofport->opp.name;
1615 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1616 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1617 shash_add(&p->port_by_name, netdev_name, ofport);
1619 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1624 ofport_remove(struct ofproto *p, struct ofport *ofport)
1626 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1627 hmap_remove(&p->ports, &ofport->hmap_node);
1628 shash_delete(&p->port_by_name,
1629 shash_find(&p->port_by_name, ofport->opp.name));
1631 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1636 ofport_free(struct ofport *ofport)
1639 netdev_close(ofport->netdev);
1644 static struct ofport *
1645 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1647 struct ofport *port;
1649 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1650 hash_int(odp_port, 0), &ofproto->ports) {
1651 if (port->odp_port == odp_port) {
1659 update_port(struct ofproto *p, const char *devname)
1661 struct odp_port odp_port;
1662 struct ofport *old_ofport;
1663 struct ofport *new_ofport;
1666 COVERAGE_INC(ofproto_update_port);
1668 /* Query the datapath for port information. */
1669 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1671 /* Find the old ofport. */
1672 old_ofport = shash_find_data(&p->port_by_name, devname);
1675 /* There's no port named 'devname' but there might be a port with
1676 * the same port number. This could happen if a port is deleted
1677 * and then a new one added in its place very quickly, or if a port
1678 * is renamed. In the former case we want to send an OFPPR_DELETE
1679 * and an OFPPR_ADD, and in the latter case we want to send a
1680 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1681 * the old port's ifindex against the new port, or perhaps less
1682 * reliably but more portably by comparing the old port's MAC
1683 * against the new port's MAC. However, this code isn't that smart
1684 * and always sends an OFPPR_MODIFY (XXX). */
1685 old_ofport = get_port(p, odp_port.port);
1687 } else if (error != ENOENT && error != ENODEV) {
1688 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1689 "%s", strerror(error));
1693 /* Create a new ofport. */
1694 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1696 /* Eliminate a few pathological cases. */
1697 if (!old_ofport && !new_ofport) {
1699 } else if (old_ofport && new_ofport) {
1700 /* Most of the 'config' bits are OpenFlow soft state, but
1701 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1702 * OpenFlow bits from old_ofport. (make_ofport() only sets
1703 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1704 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1706 if (ofport_equal(old_ofport, new_ofport)) {
1707 /* False alarm--no change. */
1708 ofport_free(new_ofport);
1713 /* Now deal with the normal cases. */
1715 ofport_remove(p, old_ofport);
1718 ofport_install(p, new_ofport);
1720 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1721 (!old_ofport ? OFPPR_ADD
1722 : !new_ofport ? OFPPR_DELETE
1724 ofport_free(old_ofport);
1728 init_ports(struct ofproto *p)
1730 struct odp_port *ports;
1735 error = dpif_port_list(p->dpif, &ports, &n_ports);
1740 for (i = 0; i < n_ports; i++) {
1741 const struct odp_port *odp_port = &ports[i];
1742 if (!ofport_conflicts(p, odp_port)) {
1743 struct ofport *ofport = make_ofport(odp_port);
1745 ofport_install(p, ofport);
1753 static struct ofconn *
1754 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1756 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1757 ofconn->ofproto = p;
1758 list_push_back(&p->all_conns, &ofconn->node);
1759 ofconn->rconn = rconn;
1760 ofconn->type = type;
1761 ofconn->flow_format = NXFF_OPENFLOW10;
1762 ofconn->role = NX_ROLE_OTHER;
1763 ofconn->packet_in_counter = rconn_packet_counter_create ();
1764 ofconn->pktbuf = NULL;
1765 ofconn->miss_send_len = 0;
1766 ofconn->reply_counter = rconn_packet_counter_create ();
1771 ofconn_destroy(struct ofconn *ofconn)
1773 if (ofconn->type == OFCONN_PRIMARY) {
1774 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1776 discovery_destroy(ofconn->discovery);
1778 list_remove(&ofconn->node);
1779 switch_status_unregister(ofconn->ss);
1780 rconn_destroy(ofconn->rconn);
1781 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1782 rconn_packet_counter_destroy(ofconn->reply_counter);
1783 pktbuf_destroy(ofconn->pktbuf);
1788 ofconn_run(struct ofconn *ofconn)
1790 struct ofproto *p = ofconn->ofproto;
1794 if (ofconn->discovery) {
1795 char *controller_name;
1796 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1797 discovery_question_connectivity(ofconn->discovery);
1799 if (discovery_run(ofconn->discovery, &controller_name)) {
1800 if (controller_name) {
1801 char *ofconn_name = ofconn_make_name(p, controller_name);
1802 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1805 rconn_disconnect(ofconn->rconn);
1810 for (i = 0; i < N_SCHEDULERS; i++) {
1811 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1814 rconn_run(ofconn->rconn);
1816 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1817 /* Limit the number of iterations to prevent other tasks from
1819 for (iteration = 0; iteration < 50; iteration++) {
1820 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1825 fail_open_maybe_recover(p->fail_open);
1827 handle_openflow(ofconn, of_msg);
1828 ofpbuf_delete(of_msg);
1832 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1833 ofconn_destroy(ofconn);
1838 ofconn_wait(struct ofconn *ofconn)
1842 if (ofconn->discovery) {
1843 discovery_wait(ofconn->discovery);
1845 for (i = 0; i < N_SCHEDULERS; i++) {
1846 pinsched_wait(ofconn->schedulers[i]);
1848 rconn_run_wait(ofconn->rconn);
1849 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1850 rconn_recv_wait(ofconn->rconn);
1852 COVERAGE_INC(ofproto_ofconn_stuck);
1856 /* Returns true if 'ofconn' should receive asynchronous messages. */
1858 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1860 if (ofconn->type == OFCONN_PRIMARY) {
1861 /* Primary controllers always get asynchronous messages unless they
1862 * have configured themselves as "slaves". */
1863 return ofconn->role != NX_ROLE_SLAVE;
1865 /* Service connections don't get asynchronous messages unless they have
1866 * explicitly asked for them by setting a nonzero miss send length. */
1867 return ofconn->miss_send_len > 0;
1871 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1872 * and 'target', suitable for use in log messages for identifying the
1875 * The name is dynamically allocated. The caller should free it (with free())
1876 * when it is no longer needed. */
1878 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1880 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1884 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1888 for (i = 0; i < N_SCHEDULERS; i++) {
1889 struct pinsched **s = &ofconn->schedulers[i];
1893 *s = pinsched_create(rate, burst,
1894 ofconn->ofproto->switch_status);
1896 pinsched_set_limits(*s, rate, burst);
1899 pinsched_destroy(*s);
1906 ofservice_reconfigure(struct ofservice *ofservice,
1907 const struct ofproto_controller *c)
1909 ofservice->probe_interval = c->probe_interval;
1910 ofservice->rate_limit = c->rate_limit;
1911 ofservice->burst_limit = c->burst_limit;
1914 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1915 * positive errno value. */
1917 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1919 struct ofservice *ofservice;
1920 struct pvconn *pvconn;
1923 error = pvconn_open(c->target, &pvconn);
1928 ofservice = xzalloc(sizeof *ofservice);
1929 hmap_insert(&ofproto->services, &ofservice->node,
1930 hash_string(c->target, 0));
1931 ofservice->pvconn = pvconn;
1933 ofservice_reconfigure(ofservice, c);
1939 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1941 hmap_remove(&ofproto->services, &ofservice->node);
1942 pvconn_close(ofservice->pvconn);
1946 /* Finds and returns the ofservice within 'ofproto' that has the given
1947 * 'target', or a null pointer if none exists. */
1948 static struct ofservice *
1949 ofservice_lookup(struct ofproto *ofproto, const char *target)
1951 struct ofservice *ofservice;
1953 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1954 &ofproto->services) {
1955 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1962 /* Returns true if 'rule' should be hidden from the controller.
1964 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1965 * (e.g. by in-band control) and are intentionally hidden from the
1968 rule_is_hidden(const struct rule *rule)
1970 return rule->cr.priority > UINT16_MAX;
1973 /* Creates and returns a new rule initialized as specified.
1975 * The caller is responsible for inserting the rule into the classifier (with
1976 * rule_insert()). */
1977 static struct rule *
1978 rule_create(const struct cls_rule *cls_rule,
1979 const union ofp_action *actions, size_t n_actions,
1980 uint16_t idle_timeout, uint16_t hard_timeout,
1981 ovs_be64 flow_cookie, bool send_flow_removed)
1983 struct rule *rule = xzalloc(sizeof *rule);
1984 rule->cr = *cls_rule;
1985 rule->idle_timeout = idle_timeout;
1986 rule->hard_timeout = hard_timeout;
1987 rule->flow_cookie = flow_cookie;
1988 rule->used = rule->created = time_msec();
1989 rule->send_flow_removed = send_flow_removed;
1990 list_init(&rule->facets);
1991 if (n_actions > 0) {
1992 rule->n_actions = n_actions;
1993 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1999 static struct rule *
2000 rule_from_cls_rule(const struct cls_rule *cls_rule)
2002 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2006 rule_free(struct rule *rule)
2008 free(rule->actions);
2012 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2013 * destroying any that no longer has a rule (which is probably all of them).
2015 * The caller must have already removed 'rule' from the classifier. */
2017 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2019 struct facet *facet, *next_facet;
2020 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2021 facet_revalidate(ofproto, facet);
2026 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2027 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2030 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2032 const union ofp_action *oa;
2033 struct actions_iterator i;
2035 if (out_port == htons(OFPP_NONE)) {
2038 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2039 oa = actions_next(&i)) {
2040 if (action_outputs_to_port(oa, out_port)) {
2047 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2048 * 'packet', which arrived on 'in_port'.
2050 * Takes ownership of 'packet'. */
2052 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
2053 const struct nlattr *odp_actions, size_t actions_len,
2054 struct ofpbuf *packet)
2056 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2057 && odp_actions->nla_type == ODPAT_CONTROLLER) {
2058 /* As an optimization, avoid a round-trip from userspace to kernel to
2059 * userspace. This also avoids possibly filling up kernel packet
2060 * buffers along the way. */
2061 struct odp_msg *msg;
2063 msg = ofpbuf_push_uninit(packet, sizeof *msg);
2064 msg->type = _ODPL_ACTION_NR;
2065 msg->length = sizeof(struct odp_msg) + packet->size;
2066 msg->port = in_port;
2067 msg->arg = nl_attr_get_u64(odp_actions);
2069 send_packet_in(ofproto, packet);
2075 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2076 ofpbuf_delete(packet);
2081 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2082 * statistics appropriately. 'packet' must have at least sizeof(struct
2083 * ofp_packet_in) bytes of headroom.
2085 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2086 * applying flow_extract() to 'packet' would yield the same flow as
2089 * 'facet' must have accurately composed ODP actions; that is, it must not be
2090 * in need of revalidation.
2092 * Takes ownership of 'packet'. */
2094 facet_execute(struct ofproto *ofproto, struct facet *facet,
2095 struct ofpbuf *packet)
2097 struct odp_flow_stats stats;
2099 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2101 flow_extract_stats(&facet->flow, packet, &stats);
2102 if (execute_odp_actions(ofproto, facet->flow.in_port,
2103 facet->actions, facet->actions_len, packet)) {
2104 facet_update_stats(ofproto, facet, &stats);
2105 facet->used = time_msec();
2106 netflow_flow_update_time(ofproto->netflow,
2107 &facet->nf_flow, facet->used);
2111 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2112 * statistics (or the statistics for one of its facets) appropriately.
2113 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2115 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2116 * with statistics for 'packet' either way.
2118 * Takes ownership of 'packet'. */
2120 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2121 struct ofpbuf *packet)
2123 struct action_xlate_ctx ctx;
2124 struct ofpbuf *odp_actions;
2125 struct facet *facet;
2129 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2131 flow_extract(packet, 0, in_port, &flow);
2133 /* First look for a related facet. If we find one, account it to that. */
2134 facet = facet_lookup_valid(ofproto, &flow);
2135 if (facet && facet->rule == rule) {
2136 facet_execute(ofproto, facet, packet);
2140 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2141 * create a new facet for it and use that. */
2142 if (rule_lookup(ofproto, &flow) == rule) {
2143 facet = facet_create(ofproto, rule, &flow, packet);
2144 facet_execute(ofproto, facet, packet);
2145 facet_install(ofproto, facet, true);
2149 /* We can't account anything to a facet. If we were to try, then that
2150 * facet would have a non-matching rule, busting our invariants. */
2151 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2152 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2153 size = packet->size;
2154 if (execute_odp_actions(ofproto, in_port, odp_actions->data,
2155 odp_actions->size, packet)) {
2156 rule->used = time_msec();
2157 rule->packet_count++;
2158 rule->byte_count += size;
2160 ofpbuf_delete(odp_actions);
2163 /* Inserts 'rule' into 'p''s flow table. */
2165 rule_insert(struct ofproto *p, struct rule *rule)
2167 struct rule *displaced_rule;
2169 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2170 if (displaced_rule) {
2171 rule_destroy(p, displaced_rule);
2173 p->need_revalidate = true;
2176 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2177 * 'flow' and an example 'packet' within that flow.
2179 * The caller must already have determined that no facet with an identical
2180 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2181 * 'ofproto''s classifier table. */
2182 static struct facet *
2183 facet_create(struct ofproto *ofproto, struct rule *rule,
2184 const struct flow *flow, const struct ofpbuf *packet)
2186 struct facet *facet;
2188 facet = xzalloc(sizeof *facet);
2189 facet->used = time_msec();
2190 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2191 list_push_back(&rule->facets, &facet->list_node);
2193 facet->flow = *flow;
2194 netflow_flow_init(&facet->nf_flow);
2195 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2197 facet_make_actions(ofproto, facet, packet);
2203 facet_free(struct facet *facet)
2205 free(facet->actions);
2209 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2211 * - Removes 'rule' from the classifier.
2213 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2214 * destroys them), via rule_destroy().
2217 rule_remove(struct ofproto *ofproto, struct rule *rule)
2219 COVERAGE_INC(ofproto_del_rule);
2220 ofproto->need_revalidate = true;
2221 classifier_remove(&ofproto->cls, &rule->cr);
2222 rule_destroy(ofproto, rule);
2225 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2227 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2228 * rule's statistics, via facet_uninstall().
2230 * - Removes 'facet' from its rule and from ofproto->facets.
2233 facet_remove(struct ofproto *ofproto, struct facet *facet)
2235 facet_uninstall(ofproto, facet);
2236 facet_flush_stats(ofproto, facet);
2237 hmap_remove(&ofproto->facets, &facet->hmap_node);
2238 list_remove(&facet->list_node);
2242 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2244 facet_make_actions(struct ofproto *p, struct facet *facet,
2245 const struct ofpbuf *packet)
2247 const struct rule *rule = facet->rule;
2248 struct ofpbuf *odp_actions;
2249 struct action_xlate_ctx ctx;
2251 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2252 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2253 facet->tags = ctx.tags;
2254 facet->may_install = ctx.may_set_up_flow;
2255 facet->nf_flow.output_iface = ctx.nf_output_iface;
2257 if (facet->actions_len != odp_actions->size
2258 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2259 free(facet->actions);
2260 facet->actions_len = odp_actions->size;
2261 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2264 ofpbuf_delete(odp_actions);
2268 facet_put__(struct ofproto *ofproto, struct facet *facet, int flags,
2269 struct odp_flow_put *put)
2271 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2272 odp_flow_key_from_flow(&put->flow.key, &facet->flow);
2273 put->flow.actions = facet->actions;
2274 put->flow.actions_len = facet->actions_len;
2275 put->flow.flags = 0;
2277 return dpif_flow_put(ofproto->dpif, put);
2280 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2281 * 'zero_stats' is true, clears any existing statistics from the datapath for
2284 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2286 if (facet->may_install) {
2287 struct odp_flow_put put;
2290 flags = ODPPF_CREATE | ODPPF_MODIFY;
2292 flags |= ODPPF_ZERO_STATS;
2294 if (!facet_put__(p, facet, flags, &put)) {
2295 facet->installed = true;
2300 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2301 * to the accounting hook function in the ofhooks structure. */
2303 facet_account(struct ofproto *ofproto,
2304 struct facet *facet, uint64_t extra_bytes)
2306 uint64_t total_bytes = facet->byte_count + extra_bytes;
2308 if (ofproto->ofhooks->account_flow_cb
2309 && total_bytes > facet->accounted_bytes)
2311 ofproto->ofhooks->account_flow_cb(
2312 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2313 total_bytes - facet->accounted_bytes, ofproto->aux);
2314 facet->accounted_bytes = total_bytes;
2318 /* If 'rule' is installed in the datapath, uninstalls it. */
2320 facet_uninstall(struct ofproto *p, struct facet *facet)
2322 if (facet->installed) {
2323 struct odp_flow odp_flow;
2325 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
2326 odp_flow.actions = NULL;
2327 odp_flow.actions_len = 0;
2329 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2330 facet_update_stats(p, facet, &odp_flow.stats);
2332 facet->installed = false;
2336 /* Returns true if the only action for 'facet' is to send to the controller.
2337 * (We don't report NetFlow expiration messages for such facets because they
2338 * are just part of the control logic for the network, not real traffic). */
2340 facet_is_controller_flow(struct facet *facet)
2343 && facet->rule->n_actions == 1
2344 && action_outputs_to_port(&facet->rule->actions[0],
2345 htons(OFPP_CONTROLLER)));
2348 /* Folds all of 'facet''s statistics into its rule. Also updates the
2349 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2351 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2353 facet_account(ofproto, facet, 0);
2355 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2356 struct ofexpired expired;
2357 expired.flow = facet->flow;
2358 expired.packet_count = facet->packet_count;
2359 expired.byte_count = facet->byte_count;
2360 expired.used = facet->used;
2361 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2364 facet->rule->packet_count += facet->packet_count;
2365 facet->rule->byte_count += facet->byte_count;
2367 /* Reset counters to prevent double counting if 'facet' ever gets
2369 facet->packet_count = 0;
2370 facet->byte_count = 0;
2371 facet->accounted_bytes = 0;
2373 netflow_flow_clear(&facet->nf_flow);
2376 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2377 * Returns it if found, otherwise a null pointer.
2379 * The returned facet might need revalidation; use facet_lookup_valid()
2380 * instead if that is important. */
2381 static struct facet *
2382 facet_find(struct ofproto *ofproto, const struct flow *flow)
2384 struct facet *facet;
2386 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2388 if (flow_equal(flow, &facet->flow)) {
2396 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2397 * Returns it if found, otherwise a null pointer.
2399 * The returned facet is guaranteed to be valid. */
2400 static struct facet *
2401 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2403 struct facet *facet = facet_find(ofproto, flow);
2405 /* The facet we found might not be valid, since we could be in need of
2406 * revalidation. If it is not valid, don't return it. */
2408 && ofproto->need_revalidate
2409 && !facet_revalidate(ofproto, facet)) {
2410 COVERAGE_INC(ofproto_invalidated);
2417 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2419 * - If the rule found is different from 'facet''s current rule, moves
2420 * 'facet' to the new rule and recompiles its actions.
2422 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2423 * where it is and recompiles its actions anyway.
2425 * - If there is none, destroys 'facet'.
2427 * Returns true if 'facet' still exists, false if it has been destroyed. */
2429 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2431 struct action_xlate_ctx ctx;
2432 struct ofpbuf *odp_actions;
2433 struct rule *new_rule;
2434 bool actions_changed;
2436 COVERAGE_INC(facet_revalidate);
2438 /* Determine the new rule. */
2439 new_rule = rule_lookup(ofproto, &facet->flow);
2441 /* No new rule, so delete the facet. */
2442 facet_remove(ofproto, facet);
2446 /* Calculate new ODP actions.
2448 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2449 * emit a NetFlow expiration and, if so, we need to have the old state
2450 * around to properly compose it. */
2451 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2452 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2453 actions_changed = (facet->actions_len != odp_actions->size
2454 || memcmp(facet->actions, odp_actions->data,
2455 facet->actions_len));
2457 /* If the ODP actions changed or the installability changed, then we need
2458 * to talk to the datapath. */
2459 if (actions_changed || facet->may_install != facet->installed) {
2460 if (facet->may_install) {
2461 struct odp_flow_put put;
2463 memset(&put.flow.stats, 0, sizeof put.flow.stats);
2464 odp_flow_key_from_flow(&put.flow.key, &facet->flow);
2465 put.flow.actions = odp_actions->data;
2466 put.flow.actions_len = odp_actions->size;
2468 put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS;
2469 dpif_flow_put(ofproto->dpif, &put);
2471 facet_update_stats(ofproto, facet, &put.flow.stats);
2473 facet_uninstall(ofproto, facet);
2476 /* The datapath flow is gone or has zeroed stats, so push stats out of
2477 * 'facet' into 'rule'. */
2478 facet_flush_stats(ofproto, facet);
2481 /* Update 'facet' now that we've taken care of all the old state. */
2482 facet->tags = ctx.tags;
2483 facet->nf_flow.output_iface = ctx.nf_output_iface;
2484 facet->may_install = ctx.may_set_up_flow;
2485 if (actions_changed) {
2486 free(facet->actions);
2487 facet->actions_len = odp_actions->size;
2488 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2490 if (facet->rule != new_rule) {
2491 COVERAGE_INC(facet_changed_rule);
2492 list_remove(&facet->list_node);
2493 list_push_back(&new_rule->facets, &facet->list_node);
2494 facet->rule = new_rule;
2495 facet->used = new_rule->created;
2498 ofpbuf_delete(odp_actions);
2504 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2505 struct rconn_packet_counter *counter)
2507 update_openflow_length(msg);
2508 if (rconn_send(ofconn->rconn, msg, counter)) {
2514 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2517 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2519 COVERAGE_INC(ofproto_error);
2520 queue_tx(buf, ofconn, ofconn->reply_counter);
2525 hton_ofp_phy_port(struct ofp_phy_port *opp)
2527 opp->port_no = htons(opp->port_no);
2528 opp->config = htonl(opp->config);
2529 opp->state = htonl(opp->state);
2530 opp->curr = htonl(opp->curr);
2531 opp->advertised = htonl(opp->advertised);
2532 opp->supported = htonl(opp->supported);
2533 opp->peer = htonl(opp->peer);
2537 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2539 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2544 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2546 struct ofp_switch_features *osf;
2548 struct ofport *port;
2550 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2551 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2552 osf->n_buffers = htonl(pktbuf_capacity());
2554 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2555 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2556 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2557 (1u << OFPAT_SET_VLAN_VID) |
2558 (1u << OFPAT_SET_VLAN_PCP) |
2559 (1u << OFPAT_STRIP_VLAN) |
2560 (1u << OFPAT_SET_DL_SRC) |
2561 (1u << OFPAT_SET_DL_DST) |
2562 (1u << OFPAT_SET_NW_SRC) |
2563 (1u << OFPAT_SET_NW_DST) |
2564 (1u << OFPAT_SET_NW_TOS) |
2565 (1u << OFPAT_SET_TP_SRC) |
2566 (1u << OFPAT_SET_TP_DST) |
2567 (1u << OFPAT_ENQUEUE));
2569 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2570 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2573 queue_tx(buf, ofconn, ofconn->reply_counter);
2578 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2581 struct ofp_switch_config *osc;
2585 /* Figure out flags. */
2586 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2587 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2590 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2591 osc->flags = htons(flags);
2592 osc->miss_send_len = htons(ofconn->miss_send_len);
2593 queue_tx(buf, ofconn, ofconn->reply_counter);
2599 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2601 uint16_t flags = ntohs(osc->flags);
2603 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2604 switch (flags & OFPC_FRAG_MASK) {
2605 case OFPC_FRAG_NORMAL:
2606 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2608 case OFPC_FRAG_DROP:
2609 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2612 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2618 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2623 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2624 * flow translation. */
2625 #define MAX_RESUBMIT_RECURSION 8
2627 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2628 struct action_xlate_ctx *ctx);
2631 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2633 const struct ofport *ofport = get_port(ctx->ofproto, port);
2636 if (ofport->opp.config & OFPPC_NO_FWD) {
2637 /* Forwarding disabled on port. */
2642 * We don't have an ofport record for this port, but it doesn't hurt to
2643 * allow forwarding to it anyhow. Maybe such a port will appear later
2644 * and we're pre-populating the flow table.
2648 nl_msg_put_u32(ctx->odp_actions, ODPAT_OUTPUT, port);
2649 ctx->nf_output_iface = port;
2652 static struct rule *
2653 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2655 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2659 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2661 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2662 uint16_t old_in_port;
2665 /* Look up a flow with 'in_port' as the input port. Then restore the
2666 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2667 * have surprising behavior). */
2668 old_in_port = ctx->flow.in_port;
2669 ctx->flow.in_port = in_port;
2670 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2671 ctx->flow.in_port = old_in_port;
2673 if (ctx->resubmit_hook) {
2674 ctx->resubmit_hook(ctx, rule);
2679 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2683 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2685 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2686 MAX_RESUBMIT_RECURSION);
2691 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2692 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2694 struct ofport *ofport;
2696 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2697 uint16_t odp_port = ofport->odp_port;
2698 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2699 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, odp_port);
2702 *nf_output_iface = NF_OUT_FLOOD;
2706 xlate_output_action__(struct action_xlate_ctx *ctx,
2707 uint16_t port, uint16_t max_len)
2710 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2712 ctx->nf_output_iface = NF_OUT_DROP;
2716 add_output_action(ctx, ctx->flow.in_port);
2719 xlate_table_action(ctx, ctx->flow.in_port);
2722 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2723 ctx->odp_actions, &ctx->tags,
2724 &ctx->nf_output_iface,
2725 ctx->ofproto->aux)) {
2726 COVERAGE_INC(ofproto_uninstallable);
2727 ctx->may_set_up_flow = false;
2731 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2732 &ctx->nf_output_iface, ctx->odp_actions);
2735 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2736 &ctx->nf_output_iface, ctx->odp_actions);
2738 case OFPP_CONTROLLER:
2739 nl_msg_put_u64(ctx->odp_actions, ODPAT_CONTROLLER, max_len);
2742 add_output_action(ctx, ODPP_LOCAL);
2745 odp_port = ofp_port_to_odp_port(port);
2746 if (odp_port != ctx->flow.in_port) {
2747 add_output_action(ctx, odp_port);
2752 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2753 ctx->nf_output_iface = NF_OUT_FLOOD;
2754 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2755 ctx->nf_output_iface = prev_nf_output_iface;
2756 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2757 ctx->nf_output_iface != NF_OUT_FLOOD) {
2758 ctx->nf_output_iface = NF_OUT_MULTI;
2763 xlate_output_action(struct action_xlate_ctx *ctx,
2764 const struct ofp_action_output *oao)
2766 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2769 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2770 * optimization, because we're going to add another action that sets the
2771 * priority immediately after, or because there are no actions following the
2774 remove_pop_action(struct action_xlate_ctx *ctx)
2776 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2777 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2778 ctx->last_pop_priority = -1;
2783 add_pop_action(struct action_xlate_ctx *ctx)
2785 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2786 nl_msg_put_flag(ctx->odp_actions, ODPAT_POP_PRIORITY);
2787 ctx->last_pop_priority = ctx->odp_actions->size;
2792 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2793 const struct ofp_action_enqueue *oae)
2795 uint16_t ofp_port, odp_port;
2799 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2802 /* Fall back to ordinary output action. */
2803 xlate_output_action__(ctx, ntohs(oae->port), 0);
2807 /* Figure out ODP output port. */
2808 ofp_port = ntohs(oae->port);
2809 if (ofp_port != OFPP_IN_PORT) {
2810 odp_port = ofp_port_to_odp_port(ofp_port);
2812 odp_port = ctx->flow.in_port;
2815 /* Add ODP actions. */
2816 remove_pop_action(ctx);
2817 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2818 add_output_action(ctx, odp_port);
2819 add_pop_action(ctx);
2821 /* Update NetFlow output port. */
2822 if (ctx->nf_output_iface == NF_OUT_DROP) {
2823 ctx->nf_output_iface = odp_port;
2824 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2825 ctx->nf_output_iface = NF_OUT_MULTI;
2830 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2831 const struct nx_action_set_queue *nasq)
2836 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2839 /* Couldn't translate queue to a priority, so ignore. A warning
2840 * has already been logged. */
2844 remove_pop_action(ctx);
2845 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2849 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2851 ovs_be16 tci = ctx->flow.vlan_tci;
2852 if (!(tci & htons(VLAN_CFI))) {
2853 nl_msg_put_flag(ctx->odp_actions, ODPAT_STRIP_VLAN);
2855 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_DL_TCI,
2856 tci & ~htons(VLAN_CFI));
2861 xlate_reg_move_action(struct action_xlate_ctx *ctx,
2862 const struct nx_action_reg_move *narm)
2864 ovs_be16 old_tci = ctx->flow.vlan_tci;
2866 nxm_execute_reg_move(narm, &ctx->flow);
2868 if (ctx->flow.vlan_tci != old_tci) {
2869 xlate_set_dl_tci(ctx);
2874 xlate_nicira_action(struct action_xlate_ctx *ctx,
2875 const struct nx_action_header *nah)
2877 const struct nx_action_resubmit *nar;
2878 const struct nx_action_set_tunnel *nast;
2879 const struct nx_action_set_queue *nasq;
2880 enum nx_action_subtype subtype = ntohs(nah->subtype);
2883 assert(nah->vendor == htonl(NX_VENDOR_ID));
2885 case NXAST_RESUBMIT:
2886 nar = (const struct nx_action_resubmit *) nah;
2887 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2890 case NXAST_SET_TUNNEL:
2891 nast = (const struct nx_action_set_tunnel *) nah;
2892 tun_id = htonll(ntohl(nast->tun_id));
2893 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id);
2894 ctx->flow.tun_id = tun_id;
2897 case NXAST_DROP_SPOOFED_ARP:
2898 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2899 nl_msg_put_flag(ctx->odp_actions, ODPAT_DROP_SPOOFED_ARP);
2903 case NXAST_SET_QUEUE:
2904 nasq = (const struct nx_action_set_queue *) nah;
2905 xlate_set_queue_action(ctx, nasq);
2908 case NXAST_POP_QUEUE:
2909 add_pop_action(ctx);
2912 case NXAST_REG_MOVE:
2913 xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah);
2916 case NXAST_REG_LOAD:
2917 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2921 /* Nothing to do. */
2924 case NXAST_SET_TUNNEL64:
2925 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2926 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id);
2927 ctx->flow.tun_id = tun_id;
2930 /* If you add a new action here that modifies flow data, don't forget to
2931 * update the flow key in ctx->flow at the same time. */
2933 case NXAST_SNAT__OBSOLETE:
2935 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2941 do_xlate_actions(const union ofp_action *in, size_t n_in,
2942 struct action_xlate_ctx *ctx)
2944 struct actions_iterator iter;
2945 const union ofp_action *ia;
2946 const struct ofport *port;
2948 port = get_port(ctx->ofproto, ctx->flow.in_port);
2949 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2950 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2951 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2952 /* Drop this flow. */
2956 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2957 enum ofp_action_type type = ntohs(ia->type);
2958 const struct ofp_action_dl_addr *oada;
2962 xlate_output_action(ctx, &ia->output);
2965 case OFPAT_SET_VLAN_VID:
2966 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2967 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2968 xlate_set_dl_tci(ctx);
2971 case OFPAT_SET_VLAN_PCP:
2972 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2973 ctx->flow.vlan_tci |= htons(
2974 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2975 xlate_set_dl_tci(ctx);
2978 case OFPAT_STRIP_VLAN:
2979 ctx->flow.vlan_tci = htons(0);
2980 xlate_set_dl_tci(ctx);
2983 case OFPAT_SET_DL_SRC:
2984 oada = ((struct ofp_action_dl_addr *) ia);
2985 nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_SRC,
2986 oada->dl_addr, ETH_ADDR_LEN);
2987 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2990 case OFPAT_SET_DL_DST:
2991 oada = ((struct ofp_action_dl_addr *) ia);
2992 nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_DST,
2993 oada->dl_addr, ETH_ADDR_LEN);
2994 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2997 case OFPAT_SET_NW_SRC:
2998 nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_SRC,
2999 ia->nw_addr.nw_addr);
3000 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3003 case OFPAT_SET_NW_DST:
3004 nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_DST,
3005 ia->nw_addr.nw_addr);
3006 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3009 case OFPAT_SET_NW_TOS:
3010 nl_msg_put_u8(ctx->odp_actions, ODPAT_SET_NW_TOS,
3012 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3015 case OFPAT_SET_TP_SRC:
3016 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_SRC,
3017 ia->tp_port.tp_port);
3018 ctx->flow.tp_src = ia->tp_port.tp_port;
3021 case OFPAT_SET_TP_DST:
3022 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_DST,
3023 ia->tp_port.tp_port);
3024 ctx->flow.tp_dst = ia->tp_port.tp_port;
3028 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3032 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3036 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3043 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3044 struct ofproto *ofproto, const struct flow *flow,
3045 const struct ofpbuf *packet)
3047 ctx->ofproto = ofproto;
3049 ctx->packet = packet;
3050 ctx->resubmit_hook = NULL;
3053 static struct ofpbuf *
3054 xlate_actions(struct action_xlate_ctx *ctx,
3055 const union ofp_action *in, size_t n_in)
3057 COVERAGE_INC(ofproto_ofp2odp);
3059 ctx->odp_actions = ofpbuf_new(512);
3061 ctx->may_set_up_flow = true;
3062 ctx->nf_output_iface = NF_OUT_DROP;
3064 ctx->last_pop_priority = -1;
3065 do_xlate_actions(in, n_in, ctx);
3066 remove_pop_action(ctx);
3068 /* Check with in-band control to see if we're allowed to set up this
3070 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3071 ctx->odp_actions->data, ctx->odp_actions->size)) {
3072 ctx->may_set_up_flow = false;
3075 return ctx->odp_actions;
3078 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3079 * error message code (composed with ofp_mkerr()) for the caller to propagate
3080 * upward. Otherwise, returns 0.
3082 * The log message mentions 'msg_type'. */
3084 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3086 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3087 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3088 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3091 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3098 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3100 struct ofproto *p = ofconn->ofproto;
3101 struct ofp_packet_out *opo;
3102 struct ofpbuf payload, *buffer;
3103 union ofp_action *ofp_actions;
3104 struct action_xlate_ctx ctx;
3105 struct ofpbuf *odp_actions;
3106 struct ofpbuf request;
3108 size_t n_ofp_actions;
3112 COVERAGE_INC(ofproto_packet_out);
3114 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3119 /* Get ofp_packet_out. */
3120 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3121 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3124 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3125 &ofp_actions, &n_ofp_actions);
3131 if (opo->buffer_id != htonl(UINT32_MAX)) {
3132 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3134 if (error || !buffer) {
3143 /* Extract flow, check actions. */
3144 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3146 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3152 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3153 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3154 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3155 ofpbuf_delete(odp_actions);
3158 ofpbuf_delete(buffer);
3163 update_port_config(struct ofproto *p, struct ofport *port,
3164 uint32_t config, uint32_t mask)
3166 mask &= config ^ port->opp.config;
3167 if (mask & OFPPC_PORT_DOWN) {
3168 if (config & OFPPC_PORT_DOWN) {
3169 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3171 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3174 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3175 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3176 if (mask & REVALIDATE_BITS) {
3177 COVERAGE_INC(ofproto_costly_flags);
3178 port->opp.config ^= mask & REVALIDATE_BITS;
3179 p->need_revalidate = true;
3181 #undef REVALIDATE_BITS
3182 if (mask & OFPPC_NO_PACKET_IN) {
3183 port->opp.config ^= OFPPC_NO_PACKET_IN;
3188 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3190 struct ofproto *p = ofconn->ofproto;
3191 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3192 struct ofport *port;
3195 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3200 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3202 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3203 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3204 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3206 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3207 if (opm->advertise) {
3208 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3214 static struct ofpbuf *
3215 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3217 struct ofp_stats_reply *osr;
3220 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3221 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3223 osr->flags = htons(0);
3227 static struct ofpbuf *
3228 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3230 const struct ofp_stats_request *osr
3231 = (const struct ofp_stats_request *) request;
3232 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3236 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3237 struct ofpbuf **msgp)
3239 struct ofpbuf *msg = *msgp;
3240 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3241 if (nbytes + msg->size > UINT16_MAX) {
3242 struct ofp_stats_reply *reply = msg->data;
3243 reply->flags = htons(OFPSF_REPLY_MORE);
3244 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3245 queue_tx(msg, ofconn, ofconn->reply_counter);
3247 return ofpbuf_put_uninit(*msgp, nbytes);
3250 static struct ofpbuf *
3251 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3253 struct nicira_stats_msg *nsm;
3256 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3257 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3258 nsm->type = htons(OFPST_VENDOR);
3259 nsm->flags = htons(0);
3260 nsm->vendor = htonl(NX_VENDOR_ID);
3261 nsm->subtype = htonl(subtype);
3265 static struct ofpbuf *
3266 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3268 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3272 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3273 struct ofpbuf **msgp)
3275 struct ofpbuf *msg = *msgp;
3276 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3277 if (nbytes + msg->size > UINT16_MAX) {
3278 struct nicira_stats_msg *reply = msg->data;
3279 reply->flags = htons(OFPSF_REPLY_MORE);
3280 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3281 queue_tx(msg, ofconn, ofconn->reply_counter);
3283 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3287 handle_desc_stats_request(struct ofconn *ofconn,
3288 const struct ofp_header *request)
3290 struct ofproto *p = ofconn->ofproto;
3291 struct ofp_desc_stats *ods;
3294 msg = start_ofp_stats_reply(request, sizeof *ods);
3295 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3296 memset(ods, 0, sizeof *ods);
3297 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3298 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3299 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3300 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3301 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3302 queue_tx(msg, ofconn, ofconn->reply_counter);
3308 handle_table_stats_request(struct ofconn *ofconn,
3309 const struct ofp_header *request)
3311 struct ofproto *p = ofconn->ofproto;
3312 struct ofp_table_stats *ots;
3315 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3317 /* Classifier table. */
3318 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3319 memset(ots, 0, sizeof *ots);
3320 strcpy(ots->name, "classifier");
3321 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3322 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3323 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3324 ots->active_count = htonl(classifier_count(&p->cls));
3325 ots->lookup_count = htonll(0); /* XXX */
3326 ots->matched_count = htonll(0); /* XXX */
3328 queue_tx(msg, ofconn, ofconn->reply_counter);
3333 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3334 struct ofpbuf **msgp)
3336 struct netdev_stats stats;
3337 struct ofp_port_stats *ops;
3339 /* Intentionally ignore return value, since errors will set
3340 * 'stats' to all-1s, which is correct for OpenFlow, and
3341 * netdev_get_stats() will log errors. */
3342 netdev_get_stats(port->netdev, &stats);
3344 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3345 ops->port_no = htons(port->opp.port_no);
3346 memset(ops->pad, 0, sizeof ops->pad);
3347 ops->rx_packets = htonll(stats.rx_packets);
3348 ops->tx_packets = htonll(stats.tx_packets);
3349 ops->rx_bytes = htonll(stats.rx_bytes);
3350 ops->tx_bytes = htonll(stats.tx_bytes);
3351 ops->rx_dropped = htonll(stats.rx_dropped);
3352 ops->tx_dropped = htonll(stats.tx_dropped);
3353 ops->rx_errors = htonll(stats.rx_errors);
3354 ops->tx_errors = htonll(stats.tx_errors);
3355 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3356 ops->rx_over_err = htonll(stats.rx_over_errors);
3357 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3358 ops->collisions = htonll(stats.collisions);
3362 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3364 struct ofproto *p = ofconn->ofproto;
3365 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3366 struct ofp_port_stats *ops;
3368 struct ofport *port;
3370 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3371 if (psr->port_no != htons(OFPP_NONE)) {
3372 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3374 append_port_stat(port, ofconn, &msg);
3377 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3378 append_port_stat(port, ofconn, &msg);
3382 queue_tx(msg, ofconn, ofconn->reply_counter);
3386 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3387 * '*packet_countp' and '*byte_countp'. The returned statistics include
3388 * statistics for all of 'rule''s facets. */
3390 query_stats(struct ofproto *p, struct rule *rule,
3391 uint64_t *packet_countp, uint64_t *byte_countp)
3393 uint64_t packet_count, byte_count;
3394 struct facet *facet;
3395 struct odp_flow *odp_flows;
3398 /* Start from historical data for 'rule' itself that are no longer tracked
3399 * by the datapath. This counts, for example, facets that have expired. */
3400 packet_count = rule->packet_count;
3401 byte_count = rule->byte_count;
3403 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3405 * Also, add any statistics that are not tracked by the datapath for each
3406 * facet. This includes, for example, statistics for packets that were
3407 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3409 odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows);
3411 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3412 struct odp_flow *odp_flow = &odp_flows[n_odp_flows++];
3413 odp_flow_key_from_flow(&odp_flow->key, &facet->flow);
3414 packet_count += facet->packet_count;
3415 byte_count += facet->byte_count;
3418 /* Fetch up-to-date statistics from the datapath and add them in. */
3419 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3422 for (i = 0; i < n_odp_flows; i++) {
3423 struct odp_flow *odp_flow = &odp_flows[i];
3424 packet_count += odp_flow->stats.n_packets;
3425 byte_count += odp_flow->stats.n_bytes;
3430 /* Return the stats to the caller. */
3431 *packet_countp = packet_count;
3432 *byte_countp = byte_count;
3436 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3438 long long int msecs = time_msec() - start;
3439 *sec = htonl(msecs / 1000);
3440 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3444 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3445 ovs_be16 out_port, struct ofpbuf **replyp)
3447 struct ofp_flow_stats *ofs;
3448 uint64_t packet_count, byte_count;
3449 size_t act_len, len;
3451 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3455 act_len = sizeof *rule->actions * rule->n_actions;
3456 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3458 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3460 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3461 ofs->length = htons(len);
3464 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3465 rule->flow_cookie, &ofs->cookie);
3466 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3467 ofs->priority = htons(rule->cr.priority);
3468 ofs->idle_timeout = htons(rule->idle_timeout);
3469 ofs->hard_timeout = htons(rule->hard_timeout);
3470 memset(ofs->pad2, 0, sizeof ofs->pad2);
3471 ofs->packet_count = htonll(packet_count);
3472 ofs->byte_count = htonll(byte_count);
3473 if (rule->n_actions > 0) {
3474 memcpy(ofs->actions, rule->actions, act_len);
3479 is_valid_table(uint8_t table_id)
3481 return table_id == 0 || table_id == 0xff;
3485 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3487 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3488 struct ofpbuf *reply;
3490 COVERAGE_INC(ofproto_flows_req);
3491 reply = start_ofp_stats_reply(oh, 1024);
3492 if (is_valid_table(fsr->table_id)) {
3493 struct cls_cursor cursor;
3494 struct cls_rule target;
3497 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3499 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3500 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3501 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3504 queue_tx(reply, ofconn, ofconn->reply_counter);
3510 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3511 ovs_be16 out_port, struct ofpbuf **replyp)
3513 struct nx_flow_stats *nfs;
3514 uint64_t packet_count, byte_count;
3515 size_t act_len, start_len;
3516 struct ofpbuf *reply;
3518 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3522 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3524 act_len = sizeof *rule->actions * rule->n_actions;
3526 start_len = (*replyp)->size;
3527 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3530 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3533 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3534 nfs->cookie = rule->flow_cookie;
3535 nfs->priority = htons(rule->cr.priority);
3536 nfs->idle_timeout = htons(rule->idle_timeout);
3537 nfs->hard_timeout = htons(rule->hard_timeout);
3538 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3539 memset(nfs->pad2, 0, sizeof nfs->pad2);
3540 nfs->packet_count = htonll(packet_count);
3541 nfs->byte_count = htonll(byte_count);
3542 if (rule->n_actions > 0) {
3543 ofpbuf_put(reply, rule->actions, act_len);
3545 nfs->length = htons(reply->size - start_len);
3549 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3551 struct nx_flow_stats_request *nfsr;
3552 struct cls_rule target;
3553 struct ofpbuf *reply;
3557 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3559 /* Dissect the message. */
3560 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3561 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3566 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3569 COVERAGE_INC(ofproto_flows_req);
3570 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3571 if (is_valid_table(nfsr->table_id)) {
3572 struct cls_cursor cursor;
3575 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3576 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3577 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3580 queue_tx(reply, ofconn, ofconn->reply_counter);
3586 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3588 uint64_t packet_count, byte_count;
3589 size_t act_len = sizeof *rule->actions * rule->n_actions;
3591 query_stats(ofproto, rule, &packet_count, &byte_count);
3593 ds_put_format(results, "duration=%llds, ",
3594 (time_msec() - rule->created) / 1000);
3595 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3596 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3597 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3598 cls_rule_format(&rule->cr, results);
3600 ofp_print_actions(results, &rule->actions->header, act_len);
3602 ds_put_cstr(results, "drop");
3604 ds_put_cstr(results, "\n");
3607 /* Adds a pretty-printed description of all flows to 'results', including
3608 * those marked hidden by secchan (e.g., by in-band control). */
3610 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3612 struct cls_cursor cursor;
3615 cls_cursor_init(&cursor, &p->cls, NULL);
3616 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3617 flow_stats_ds(p, rule, results);
3622 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3623 ovs_be16 out_port, uint8_t table_id,
3624 struct ofp_aggregate_stats_reply *oasr)
3626 uint64_t total_packets = 0;
3627 uint64_t total_bytes = 0;
3630 COVERAGE_INC(ofproto_agg_request);
3632 if (is_valid_table(table_id)) {
3633 struct cls_cursor cursor;
3636 cls_cursor_init(&cursor, &ofproto->cls, target);
3637 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3638 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3639 uint64_t packet_count;
3640 uint64_t byte_count;
3642 query_stats(ofproto, rule, &packet_count, &byte_count);
3644 total_packets += packet_count;
3645 total_bytes += byte_count;
3651 oasr->flow_count = htonl(n_flows);
3652 oasr->packet_count = htonll(total_packets);
3653 oasr->byte_count = htonll(total_bytes);
3654 memset(oasr->pad, 0, sizeof oasr->pad);
3658 handle_aggregate_stats_request(struct ofconn *ofconn,
3659 const struct ofp_header *oh)
3661 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3662 struct ofp_aggregate_stats_reply *reply;
3663 struct cls_rule target;
3666 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3669 msg = start_ofp_stats_reply(oh, sizeof *reply);
3670 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3671 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3672 request->table_id, reply);
3673 queue_tx(msg, ofconn, ofconn->reply_counter);
3678 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3680 struct nx_aggregate_stats_request *request;
3681 struct ofp_aggregate_stats_reply *reply;
3682 struct cls_rule target;
3687 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3689 /* Dissect the message. */
3690 request = ofpbuf_pull(&b, sizeof *request);
3691 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3696 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3700 COVERAGE_INC(ofproto_flows_req);
3701 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3702 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3703 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3704 request->table_id, reply);
3705 queue_tx(buf, ofconn, ofconn->reply_counter);
3710 struct queue_stats_cbdata {
3711 struct ofconn *ofconn;
3712 struct ofport *ofport;
3717 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3718 const struct netdev_queue_stats *stats)
3720 struct ofp_queue_stats *reply;
3722 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3723 reply->port_no = htons(cbdata->ofport->opp.port_no);
3724 memset(reply->pad, 0, sizeof reply->pad);
3725 reply->queue_id = htonl(queue_id);
3726 reply->tx_bytes = htonll(stats->tx_bytes);
3727 reply->tx_packets = htonll(stats->tx_packets);
3728 reply->tx_errors = htonll(stats->tx_errors);
3732 handle_queue_stats_dump_cb(uint32_t queue_id,
3733 struct netdev_queue_stats *stats,
3736 struct queue_stats_cbdata *cbdata = cbdata_;
3738 put_queue_stats(cbdata, queue_id, stats);
3742 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3743 struct queue_stats_cbdata *cbdata)
3745 cbdata->ofport = port;
3746 if (queue_id == OFPQ_ALL) {
3747 netdev_dump_queue_stats(port->netdev,
3748 handle_queue_stats_dump_cb, cbdata);
3750 struct netdev_queue_stats stats;
3752 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3753 put_queue_stats(cbdata, queue_id, &stats);
3759 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3761 struct ofproto *ofproto = ofconn->ofproto;
3762 const struct ofp_queue_stats_request *qsr;
3763 struct queue_stats_cbdata cbdata;
3764 struct ofport *port;
3765 unsigned int port_no;
3768 qsr = ofputil_stats_body(oh);
3770 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3773 COVERAGE_INC(ofproto_queue_req);
3775 cbdata.ofconn = ofconn;
3776 cbdata.msg = start_ofp_stats_reply(oh, 128);
3778 port_no = ntohs(qsr->port_no);
3779 queue_id = ntohl(qsr->queue_id);
3780 if (port_no == OFPP_ALL) {
3781 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3782 handle_queue_stats_for_port(port, queue_id, &cbdata);
3784 } else if (port_no < ofproto->max_ports) {
3785 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3787 handle_queue_stats_for_port(port, queue_id, &cbdata);
3790 ofpbuf_delete(cbdata.msg);
3791 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3793 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3798 static long long int
3799 msec_from_nsec(uint64_t sec, uint32_t nsec)
3801 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3805 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3806 const struct odp_flow_stats *stats)
3808 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3809 if (used > facet->used) {
3811 if (used > facet->rule->used) {
3812 facet->rule->used = used;
3814 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3818 /* Folds the statistics from 'stats' into the counters in 'facet'.
3820 * Because of the meaning of a facet's counters, it only makes sense to do this
3821 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3822 * packet that was sent by hand or if it represents statistics that have been
3823 * cleared out of the datapath. */
3825 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3826 const struct odp_flow_stats *stats)
3828 if (stats->n_packets) {
3829 facet_update_time(ofproto, facet, stats);
3830 facet->packet_count += stats->n_packets;
3831 facet->byte_count += stats->n_bytes;
3832 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3836 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3837 * in which no matching flow already exists in the flow table.
3839 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3840 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3841 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3843 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3846 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3848 struct ofproto *p = ofconn->ofproto;
3849 struct ofpbuf *packet;
3854 if (fm->flags & OFPFF_CHECK_OVERLAP
3855 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3856 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3860 if (fm->buffer_id != UINT32_MAX) {
3861 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3865 in_port = UINT16_MAX;
3868 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3869 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3870 fm->flags & OFPFF_SEND_FLOW_REM);
3871 rule_insert(p, rule);
3873 rule_execute(p, rule, in_port, packet);
3878 static struct rule *
3879 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3881 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3885 send_buffered_packet(struct ofconn *ofconn,
3886 struct rule *rule, uint32_t buffer_id)
3888 struct ofpbuf *packet;
3892 if (buffer_id == UINT32_MAX) {
3896 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3901 rule_execute(ofconn->ofproto, rule, in_port, packet);
3906 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3908 struct modify_flows_cbdata {
3909 struct ofproto *ofproto;
3910 const struct flow_mod *fm;
3914 static int modify_flow(struct ofproto *, const struct flow_mod *,
3917 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3918 * encoded by ofp_mkerr() on failure.
3920 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3923 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3925 struct ofproto *p = ofconn->ofproto;
3926 struct rule *match = NULL;
3927 struct cls_cursor cursor;
3930 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3931 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3932 if (!rule_is_hidden(rule)) {
3934 modify_flow(p, fm, rule);
3939 /* This credits the packet to whichever flow happened to match last.
3940 * That's weird. Maybe we should do a lookup for the flow that
3941 * actually matches the packet? Who knows. */
3942 send_buffered_packet(ofconn, match, fm->buffer_id);
3945 return add_flow(ofconn, fm);
3949 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3950 * code as encoded by ofp_mkerr() on failure.
3952 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3955 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3957 struct ofproto *p = ofconn->ofproto;
3958 struct rule *rule = find_flow_strict(p, fm);
3959 if (rule && !rule_is_hidden(rule)) {
3960 modify_flow(p, fm, rule);
3961 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3963 return add_flow(ofconn, fm);
3967 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3968 * been identified as a flow in 'p''s flow table to be modified, by changing
3969 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3970 * ofp_action[] structures). */
3972 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3974 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3976 rule->flow_cookie = fm->cookie;
3978 /* If the actions are the same, do nothing. */
3979 if (fm->n_actions == rule->n_actions
3981 || !memcmp(fm->actions, rule->actions, actions_len))) {
3985 /* Replace actions. */
3986 free(rule->actions);
3987 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3988 rule->n_actions = fm->n_actions;
3990 p->need_revalidate = true;
3995 /* OFPFC_DELETE implementation. */
3997 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3999 /* Implements OFPFC_DELETE. */
4001 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4003 struct rule *rule, *next_rule;
4004 struct cls_cursor cursor;
4006 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4007 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4008 delete_flow(p, rule, htons(fm->out_port));
4012 /* Implements OFPFC_DELETE_STRICT. */
4014 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4016 struct rule *rule = find_flow_strict(p, fm);
4018 delete_flow(p, rule, htons(fm->out_port));
4022 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4023 * been identified as a flow to delete from 'p''s flow table, by deleting the
4024 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4027 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4028 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4029 * specified 'out_port'. */
4031 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4033 if (rule_is_hidden(rule)) {
4037 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4041 rule_send_removed(p, rule, OFPRR_DELETE);
4042 rule_remove(p, rule);
4046 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4048 struct ofproto *p = ofconn->ofproto;
4052 error = reject_slave_controller(ofconn, "flow_mod");
4057 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4062 /* We do not support the emergency flow cache. It will hopefully get
4063 * dropped from OpenFlow in the near future. */
4064 if (fm.flags & OFPFF_EMERG) {
4065 /* There isn't a good fit for an error code, so just state that the
4066 * flow table is full. */
4067 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4070 error = validate_actions(fm.actions, fm.n_actions,
4071 &fm.cr.flow, p->max_ports);
4076 switch (fm.command) {
4078 return add_flow(ofconn, &fm);
4081 return modify_flows_loose(ofconn, &fm);
4083 case OFPFC_MODIFY_STRICT:
4084 return modify_flow_strict(ofconn, &fm);
4087 delete_flows_loose(p, &fm);
4090 case OFPFC_DELETE_STRICT:
4091 delete_flow_strict(p, &fm);
4095 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4100 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4102 const struct nxt_tun_id_cookie *msg
4103 = (const struct nxt_tun_id_cookie *) oh;
4105 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4110 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4112 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4113 struct nx_role_request *reply;
4117 if (ofconn->type != OFCONN_PRIMARY) {
4118 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4120 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4123 role = ntohl(nrr->role);
4124 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4125 && role != NX_ROLE_SLAVE) {
4126 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4128 /* There's no good error code for this. */
4129 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4132 if (role == NX_ROLE_MASTER) {
4133 struct ofconn *other;
4135 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4136 if (other->role == NX_ROLE_MASTER) {
4137 other->role = NX_ROLE_SLAVE;
4141 ofconn->role = role;
4143 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4144 reply->role = htonl(role);
4145 queue_tx(buf, ofconn, ofconn->reply_counter);
4151 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4153 const struct nxt_set_flow_format *msg
4154 = (const struct nxt_set_flow_format *) oh;
4157 format = ntohl(msg->format);
4158 if (format == NXFF_OPENFLOW10
4159 || format == NXFF_TUN_ID_FROM_COOKIE
4160 || format == NXFF_NXM) {
4161 ofconn->flow_format = format;
4164 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4169 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4171 struct ofp_header *ob;
4174 /* Currently, everything executes synchronously, so we can just
4175 * immediately send the barrier reply. */
4176 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4177 queue_tx(buf, ofconn, ofconn->reply_counter);
4182 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4184 const struct ofp_header *oh = msg->data;
4185 const struct ofputil_msg_type *type;
4188 error = ofputil_decode_msg_type(oh, &type);
4193 switch (ofputil_msg_type_code(type)) {
4194 /* OpenFlow requests. */
4195 case OFPUTIL_OFPT_ECHO_REQUEST:
4196 return handle_echo_request(ofconn, oh);
4198 case OFPUTIL_OFPT_FEATURES_REQUEST:
4199 return handle_features_request(ofconn, oh);
4201 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4202 return handle_get_config_request(ofconn, oh);
4204 case OFPUTIL_OFPT_SET_CONFIG:
4205 return handle_set_config(ofconn, msg->data);
4207 case OFPUTIL_OFPT_PACKET_OUT:
4208 return handle_packet_out(ofconn, oh);
4210 case OFPUTIL_OFPT_PORT_MOD:
4211 return handle_port_mod(ofconn, oh);
4213 case OFPUTIL_OFPT_FLOW_MOD:
4214 return handle_flow_mod(ofconn, oh);
4216 case OFPUTIL_OFPT_BARRIER_REQUEST:
4217 return handle_barrier_request(ofconn, oh);
4219 /* OpenFlow replies. */
4220 case OFPUTIL_OFPT_ECHO_REPLY:
4223 /* Nicira extension requests. */
4224 case OFPUTIL_NXT_STATUS_REQUEST:
4225 return switch_status_handle_request(
4226 ofconn->ofproto->switch_status, ofconn->rconn, oh);
4228 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4229 return handle_tun_id_from_cookie(ofconn, oh);
4231 case OFPUTIL_NXT_ROLE_REQUEST:
4232 return handle_role_request(ofconn, oh);
4234 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4235 return handle_nxt_set_flow_format(ofconn, oh);
4237 case OFPUTIL_NXT_FLOW_MOD:
4238 return handle_flow_mod(ofconn, oh);
4240 /* OpenFlow statistics requests. */
4241 case OFPUTIL_OFPST_DESC_REQUEST:
4242 return handle_desc_stats_request(ofconn, oh);
4244 case OFPUTIL_OFPST_FLOW_REQUEST:
4245 return handle_flow_stats_request(ofconn, oh);
4247 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4248 return handle_aggregate_stats_request(ofconn, oh);
4250 case OFPUTIL_OFPST_TABLE_REQUEST:
4251 return handle_table_stats_request(ofconn, oh);
4253 case OFPUTIL_OFPST_PORT_REQUEST:
4254 return handle_port_stats_request(ofconn, oh);
4256 case OFPUTIL_OFPST_QUEUE_REQUEST:
4257 return handle_queue_stats_request(ofconn, oh);
4259 /* Nicira extension statistics requests. */
4260 case OFPUTIL_NXST_FLOW_REQUEST:
4261 return handle_nxst_flow(ofconn, oh);
4263 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4264 return handle_nxst_aggregate(ofconn, oh);
4266 case OFPUTIL_INVALID:
4267 case OFPUTIL_OFPT_HELLO:
4268 case OFPUTIL_OFPT_ERROR:
4269 case OFPUTIL_OFPT_FEATURES_REPLY:
4270 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4271 case OFPUTIL_OFPT_PACKET_IN:
4272 case OFPUTIL_OFPT_FLOW_REMOVED:
4273 case OFPUTIL_OFPT_PORT_STATUS:
4274 case OFPUTIL_OFPT_BARRIER_REPLY:
4275 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4276 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4277 case OFPUTIL_OFPST_DESC_REPLY:
4278 case OFPUTIL_OFPST_FLOW_REPLY:
4279 case OFPUTIL_OFPST_QUEUE_REPLY:
4280 case OFPUTIL_OFPST_PORT_REPLY:
4281 case OFPUTIL_OFPST_TABLE_REPLY:
4282 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4283 case OFPUTIL_NXT_STATUS_REPLY:
4284 case OFPUTIL_NXT_ROLE_REPLY:
4285 case OFPUTIL_NXT_FLOW_REMOVED:
4286 case OFPUTIL_NXST_FLOW_REPLY:
4287 case OFPUTIL_NXST_AGGREGATE_REPLY:
4289 if (VLOG_IS_WARN_ENABLED()) {
4290 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4291 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4294 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4295 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4297 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4303 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4305 int error = handle_openflow__(ofconn, ofp_msg);
4307 send_error_oh(ofconn, ofp_msg->data, error);
4309 COVERAGE_INC(ofproto_recv_openflow);
4313 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4315 struct odp_msg *msg = packet->data;
4316 struct ofpbuf payload;
4317 struct facet *facet;
4320 ofpbuf_use_const(&payload, msg + 1, msg->length - sizeof *msg);
4321 flow_extract(&payload, msg->arg, msg->port, &flow);
4323 packet->l2 = payload.l2;
4324 packet->l3 = payload.l3;
4325 packet->l4 = payload.l4;
4326 packet->l7 = payload.l7;
4328 /* Check with in-band control to see if this packet should be sent
4329 * to the local port regardless of the flow table. */
4330 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4331 struct ofpbuf odp_actions;
4333 ofpbuf_init(&odp_actions, 32);
4334 nl_msg_put_u32(&odp_actions, ODPAT_OUTPUT, ODPP_LOCAL);
4335 dpif_execute(p->dpif, odp_actions.data, odp_actions.size, &payload);
4336 ofpbuf_uninit(&odp_actions);
4339 facet = facet_lookup_valid(p, &flow);
4341 struct rule *rule = rule_lookup(p, &flow);
4343 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4344 struct ofport *port = get_port(p, msg->port);
4346 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4347 COVERAGE_INC(ofproto_no_packet_in);
4348 /* XXX install 'drop' flow entry */
4349 ofpbuf_delete(packet);
4353 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4357 COVERAGE_INC(ofproto_packet_in);
4358 send_packet_in(p, packet);
4362 facet = facet_create(p, rule, &flow, packet);
4363 } else if (!facet->may_install) {
4364 /* The facet is not installable, that is, we need to process every
4365 * packet, so process the current packet's actions into 'facet'. */
4366 facet_make_actions(p, facet, packet);
4369 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4371 * Extra-special case for fail-open mode.
4373 * We are in fail-open mode and the packet matched the fail-open rule,
4374 * but we are connected to a controller too. We should send the packet
4375 * up to the controller in the hope that it will try to set up a flow
4376 * and thereby allow us to exit fail-open.
4378 * See the top-level comment in fail-open.c for more information.
4380 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4381 DPIF_RECV_MSG_PADDING));
4384 ofpbuf_pull(packet, sizeof *msg);
4385 facet_execute(p, facet, packet);
4386 facet_install(p, facet, false);
4390 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4392 struct odp_msg *msg = packet->data;
4394 switch (msg->type) {
4395 case _ODPL_ACTION_NR:
4396 COVERAGE_INC(ofproto_ctlr_action);
4397 send_packet_in(p, packet);
4400 case _ODPL_SFLOW_NR:
4402 ofproto_sflow_received(p->sflow, msg);
4404 ofpbuf_delete(packet);
4408 handle_odp_miss_msg(p, packet);
4412 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4418 /* Flow expiration. */
4420 static int ofproto_dp_max_idle(const struct ofproto *);
4421 static void ofproto_update_used(struct ofproto *);
4422 static void rule_expire(struct ofproto *, struct rule *);
4423 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4425 /* This function is called periodically by ofproto_run(). Its job is to
4426 * collect updates for the flows that have been installed into the datapath,
4427 * most importantly when they last were used, and then use that information to
4428 * expire flows that have not been used recently.
4430 * Returns the number of milliseconds after which it should be called again. */
4432 ofproto_expire(struct ofproto *ofproto)
4434 struct rule *rule, *next_rule;
4435 struct cls_cursor cursor;
4438 /* Update 'used' for each flow in the datapath. */
4439 ofproto_update_used(ofproto);
4441 /* Expire facets that have been idle too long. */
4442 dp_max_idle = ofproto_dp_max_idle(ofproto);
4443 ofproto_expire_facets(ofproto, dp_max_idle);
4445 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4446 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4447 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4448 rule_expire(ofproto, rule);
4451 /* Let the hook know that we're at a stable point: all outstanding data
4452 * in existing flows has been accounted to the account_cb. Thus, the
4453 * hook can now reasonably do operations that depend on having accurate
4454 * flow volume accounting (currently, that's just bond rebalancing). */
4455 if (ofproto->ofhooks->account_checkpoint_cb) {
4456 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4459 return MIN(dp_max_idle, 1000);
4462 /* Update 'used' member of installed facets. */
4464 ofproto_update_used(struct ofproto *p)
4466 struct odp_flow *flows;
4471 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4476 for (i = 0; i < n_flows; i++) {
4477 struct odp_flow *f = &flows[i];
4478 struct facet *facet;
4481 odp_flow_key_to_flow(&f->key, &flow);
4482 facet = facet_find(p, &flow);
4484 if (facet && facet->installed) {
4485 facet_update_time(p, facet, &f->stats);
4486 facet_account(p, facet, f->stats.n_bytes);
4488 /* There's a flow in the datapath that we know nothing about.
4490 COVERAGE_INC(ofproto_unexpected_rule);
4491 dpif_flow_del(p->dpif, f);
4498 /* Calculates and returns the number of milliseconds of idle time after which
4499 * facets should expire from the datapath and we should fold their statistics
4500 * into their parent rules in userspace. */
4502 ofproto_dp_max_idle(const struct ofproto *ofproto)
4505 * Idle time histogram.
4507 * Most of the time a switch has a relatively small number of facets. When
4508 * this is the case we might as well keep statistics for all of them in
4509 * userspace and to cache them in the kernel datapath for performance as
4512 * As the number of facets increases, the memory required to maintain
4513 * statistics about them in userspace and in the kernel becomes
4514 * significant. However, with a large number of facets it is likely that
4515 * only a few of them are "heavy hitters" that consume a large amount of
4516 * bandwidth. At this point, only heavy hitters are worth caching in the
4517 * kernel and maintaining in userspaces; other facets we can discard.
4519 * The technique used to compute the idle time is to build a histogram with
4520 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4521 * that is installed in the kernel gets dropped in the appropriate bucket.
4522 * After the histogram has been built, we compute the cutoff so that only
4523 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4524 * cached. At least the most-recently-used bucket of facets is kept, so
4525 * actually an arbitrary number of facets can be kept in any given
4526 * expiration run (though the next run will delete most of those unless
4527 * they receive additional data).
4529 * This requires a second pass through the facets, in addition to the pass
4530 * made by ofproto_update_used(), because the former function never looks
4531 * at uninstallable facets.
4533 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4534 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4535 int buckets[N_BUCKETS] = { 0 };
4536 struct facet *facet;
4541 total = hmap_count(&ofproto->facets);
4542 if (total <= 1000) {
4543 return N_BUCKETS * BUCKET_WIDTH;
4546 /* Build histogram. */
4548 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4549 long long int idle = now - facet->used;
4550 int bucket = (idle <= 0 ? 0
4551 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4552 : (unsigned int) idle / BUCKET_WIDTH);
4556 /* Find the first bucket whose flows should be expired. */
4557 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4558 if (buckets[bucket]) {
4561 subtotal += buckets[bucket++];
4562 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4567 if (VLOG_IS_DBG_ENABLED()) {
4571 ds_put_cstr(&s, "keep");
4572 for (i = 0; i < N_BUCKETS; i++) {
4574 ds_put_cstr(&s, ", drop");
4577 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4580 VLOG_INFO("%s: %s (msec:count)",
4581 dpif_name(ofproto->dpif), ds_cstr(&s));
4585 return bucket * BUCKET_WIDTH;
4589 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4591 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4592 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4593 struct ofexpired expired;
4594 struct odp_flow odp_flow;
4596 /* Get updated flow stats.
4598 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4599 * updated TCP flags and (2) the dpif_flow_list_all() in
4600 * ofproto_update_used() zeroed TCP flags. */
4601 memset(&odp_flow, 0, sizeof odp_flow);
4602 if (facet->installed) {
4603 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
4604 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4605 dpif_flow_get(ofproto->dpif, &odp_flow);
4607 if (odp_flow.stats.n_packets) {
4608 facet_update_time(ofproto, facet, &odp_flow.stats);
4609 netflow_flow_update_flags(&facet->nf_flow,
4610 odp_flow.stats.tcp_flags);
4614 expired.flow = facet->flow;
4615 expired.packet_count = facet->packet_count +
4616 odp_flow.stats.n_packets;
4617 expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes;
4618 expired.used = facet->used;
4620 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4625 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4627 long long int cutoff = time_msec() - dp_max_idle;
4628 struct facet *facet, *next_facet;
4630 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4631 facet_active_timeout(ofproto, facet);
4632 if (facet->used < cutoff) {
4633 facet_remove(ofproto, facet);
4638 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4639 * then delete it entirely. */
4641 rule_expire(struct ofproto *ofproto, struct rule *rule)
4643 struct facet *facet, *next_facet;
4647 /* Has 'rule' expired? */
4649 if (rule->hard_timeout
4650 && now > rule->created + rule->hard_timeout * 1000) {
4651 reason = OFPRR_HARD_TIMEOUT;
4652 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4653 && now >rule->used + rule->idle_timeout * 1000) {
4654 reason = OFPRR_IDLE_TIMEOUT;
4659 COVERAGE_INC(ofproto_expired);
4661 /* Update stats. (This is a no-op if the rule expired due to an idle
4662 * timeout, because that only happens when the rule has no facets left.) */
4663 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4664 facet_remove(ofproto, facet);
4667 /* Get rid of the rule. */
4668 if (!rule_is_hidden(rule)) {
4669 rule_send_removed(ofproto, rule, reason);
4671 rule_remove(ofproto, rule);
4674 static struct ofpbuf *
4675 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4678 struct ofp_flow_removed *ofr;
4681 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4682 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4683 rule->flow_cookie, &ofr->cookie);
4684 ofr->priority = htons(rule->cr.priority);
4685 ofr->reason = reason;
4686 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4687 ofr->idle_timeout = htons(rule->idle_timeout);
4688 ofr->packet_count = htonll(rule->packet_count);
4689 ofr->byte_count = htonll(rule->byte_count);
4694 static struct ofpbuf *
4695 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4697 struct nx_flow_removed *nfr;
4701 nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf);
4703 match_len = nx_put_match(buf, &rule->cr);
4705 nfr->cookie = rule->flow_cookie;
4706 nfr->priority = htons(rule->cr.priority);
4707 nfr->reason = reason;
4708 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4709 nfr->idle_timeout = htons(rule->idle_timeout);
4710 nfr->match_len = htons(match_len);
4711 nfr->packet_count = htonll(rule->packet_count);
4712 nfr->byte_count = htonll(rule->byte_count);
4718 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4720 struct ofconn *ofconn;
4722 if (!rule->send_flow_removed) {
4726 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4729 if (!rconn_is_connected(ofconn->rconn)
4730 || !ofconn_receives_async_msgs(ofconn)) {
4734 msg = (ofconn->flow_format == NXFF_NXM
4735 ? compose_nx_flow_removed(rule, reason)
4736 : compose_ofp_flow_removed(ofconn, rule, reason));
4738 /* Account flow expirations under ofconn->reply_counter, the counter
4739 * for replies to OpenFlow requests. That works because preventing
4740 * OpenFlow requests from being processed also prevents new flows from
4741 * being added (and expiring). (It also prevents processing OpenFlow
4742 * requests that would not add new flows, so it is imperfect.) */
4743 queue_tx(msg, ofconn, ofconn->reply_counter);
4747 /* pinsched callback for sending 'packet' on 'ofconn'. */
4749 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4751 struct ofconn *ofconn = ofconn_;
4753 rconn_send_with_limit(ofconn->rconn, packet,
4754 ofconn->packet_in_counter, 100);
4757 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4758 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4759 * packet scheduler for sending.
4761 * 'max_len' specifies the maximum number of bytes of the packet to send on
4762 * 'ofconn' (INT_MAX specifies no limit).
4764 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4765 * ownership is transferred to this function. */
4767 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4770 struct ofproto *ofproto = ofconn->ofproto;
4771 struct ofp_packet_in *opi = packet->data;
4772 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4773 int send_len, trim_size;
4777 if (opi->reason == OFPR_ACTION) {
4778 buffer_id = UINT32_MAX;
4779 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4780 buffer_id = pktbuf_get_null();
4781 } else if (!ofconn->pktbuf) {
4782 buffer_id = UINT32_MAX;
4784 struct ofpbuf payload;
4786 ofpbuf_use_const(&payload, opi->data,
4787 packet->size - offsetof(struct ofp_packet_in, data));
4788 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4791 /* Figure out how much of the packet to send. */
4792 send_len = ntohs(opi->total_len);
4793 if (buffer_id != UINT32_MAX) {
4794 send_len = MIN(send_len, ofconn->miss_send_len);
4796 send_len = MIN(send_len, max_len);
4798 /* Adjust packet length and clone if necessary. */
4799 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4801 packet = ofpbuf_clone_data(packet->data, trim_size);
4804 packet->size = trim_size;
4807 /* Update packet headers. */
4808 opi->buffer_id = htonl(buffer_id);
4809 update_openflow_length(packet);
4811 /* Hand over to packet scheduler. It might immediately call into
4812 * do_send_packet_in() or it might buffer it for a while (until a later
4813 * call to pinsched_run()). */
4814 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4815 packet, do_send_packet_in, ofconn);
4818 /* Replace struct odp_msg header in 'packet' by equivalent struct
4819 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4820 * returned by dpif_recv()).
4822 * The conversion is not complete: the caller still needs to trim any unneeded
4823 * payload off the end of the buffer, set the length in the OpenFlow header,
4824 * and set buffer_id. Those require us to know the controller settings and so
4825 * must be done on a per-controller basis.
4827 * Returns the maximum number of bytes of the packet that should be sent to
4828 * the controller (INT_MAX if no limit). */
4830 do_convert_to_packet_in(struct ofpbuf *packet)
4832 struct odp_msg *msg = packet->data;
4833 struct ofp_packet_in *opi;
4839 /* Extract relevant header fields */
4840 if (msg->type == _ODPL_ACTION_NR) {
4841 reason = OFPR_ACTION;
4844 reason = OFPR_NO_MATCH;
4847 total_len = msg->length - sizeof *msg;
4848 in_port = odp_port_to_ofp_port(msg->port);
4850 /* Repurpose packet buffer by overwriting header. */
4851 ofpbuf_pull(packet, sizeof(struct odp_msg));
4852 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4853 opi->header.version = OFP_VERSION;
4854 opi->header.type = OFPT_PACKET_IN;
4855 opi->total_len = htons(total_len);
4856 opi->in_port = htons(in_port);
4857 opi->reason = reason;
4862 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4863 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4864 * as necessary according to their individual configurations.
4866 * 'packet' must have sufficient headroom to convert it into a struct
4867 * ofp_packet_in (e.g. as returned by dpif_recv()).
4869 * Takes ownership of 'packet'. */
4871 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4873 struct ofconn *ofconn, *prev;
4876 max_len = do_convert_to_packet_in(packet);
4879 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4880 if (ofconn_receives_async_msgs(ofconn)) {
4882 schedule_packet_in(prev, packet, max_len, true);
4888 schedule_packet_in(prev, packet, max_len, false);
4890 ofpbuf_delete(packet);
4895 pick_datapath_id(const struct ofproto *ofproto)
4897 const struct ofport *port;
4899 port = get_port(ofproto, ODPP_LOCAL);
4901 uint8_t ea[ETH_ADDR_LEN];
4904 error = netdev_get_etheraddr(port->netdev, ea);
4906 return eth_addr_to_uint64(ea);
4908 VLOG_WARN("could not get MAC address for %s (%s)",
4909 netdev_get_name(port->netdev), strerror(error));
4911 return ofproto->fallback_dpid;
4915 pick_fallback_dpid(void)
4917 uint8_t ea[ETH_ADDR_LEN];
4918 eth_addr_nicira_random(ea);
4919 return eth_addr_to_uint64(ea);
4923 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4924 void *aux OVS_UNUSED)
4926 const struct shash_node *node;
4930 SHASH_FOR_EACH (node, &all_ofprotos) {
4931 ds_put_format(&results, "%s\n", node->name);
4933 unixctl_command_reply(conn, 200, ds_cstr(&results));
4934 ds_destroy(&results);
4937 struct ofproto_trace {
4938 struct action_xlate_ctx ctx;
4944 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4946 ds_put_char_multiple(result, '\t', level);
4948 ds_put_cstr(result, "No match\n");
4952 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4953 ntohll(rule->flow_cookie));
4954 cls_rule_format(&rule->cr, result);
4955 ds_put_char(result, '\n');
4957 ds_put_char_multiple(result, '\t', level);
4958 ds_put_cstr(result, "OpenFlow ");
4959 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4960 rule->n_actions * sizeof *rule->actions);
4961 ds_put_char(result, '\n');
4965 trace_format_flow(struct ds *result, int level, const char *title,
4966 struct ofproto_trace *trace)
4968 ds_put_char_multiple(result, '\t', level);
4969 ds_put_format(result, "%s: ", title);
4970 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4971 ds_put_cstr(result, "unchanged");
4973 flow_format(result, &trace->ctx.flow);
4974 trace->flow = trace->ctx.flow;
4976 ds_put_char(result, '\n');
4980 trace_resubmit(struct action_xlate_ctx *ctx, const struct rule *rule)
4982 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4983 struct ds *result = trace->result;
4985 ds_put_char(result, '\n');
4986 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4987 trace_format_rule(result, ctx->recurse + 1, rule);
4991 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4992 void *aux OVS_UNUSED)
4994 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4995 char *args = xstrdup(args_);
4996 char *save_ptr = NULL;
4997 struct ofproto *ofproto;
4998 struct ofpbuf packet;
5006 ofpbuf_init(&packet, strlen(args) / 2);
5009 dpname = strtok_r(args, " ", &save_ptr);
5010 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5011 in_port_s = strtok_r(NULL, " ", &save_ptr);
5012 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5013 if (!dpname || !in_port_s || !packet_s) {
5014 unixctl_command_reply(conn, 501, "Bad command syntax");
5018 ofproto = shash_find_data(&all_ofprotos, dpname);
5020 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5025 tun_id = ntohl(strtoul(tun_id_s, NULL, 10));
5026 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5028 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5029 packet_s += strspn(packet_s, " ");
5030 if (*packet_s != '\0') {
5031 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5034 if (packet.size < ETH_HEADER_LEN) {
5035 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5039 ds_put_cstr(&result, "Packet: ");
5040 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5041 ds_put_cstr(&result, s);
5044 flow_extract(&packet, tun_id, in_port, &flow);
5045 ds_put_cstr(&result, "Flow: ");
5046 flow_format(&result, &flow);
5047 ds_put_char(&result, '\n');
5049 rule = rule_lookup(ofproto, &flow);
5050 trace_format_rule(&result, 0, rule);
5052 struct ofproto_trace trace;
5053 struct ofpbuf *odp_actions;
5055 trace.result = &result;
5057 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5058 trace.ctx.resubmit_hook = trace_resubmit;
5059 odp_actions = xlate_actions(&trace.ctx,
5060 rule->actions, rule->n_actions);
5062 ds_put_char(&result, '\n');
5063 trace_format_flow(&result, 0, "Final flow", &trace);
5064 ds_put_cstr(&result, "Datapath actions: ");
5065 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5066 ofpbuf_delete(odp_actions);
5069 unixctl_command_reply(conn, 200, ds_cstr(&result));
5072 ds_destroy(&result);
5073 ofpbuf_uninit(&packet);
5078 ofproto_unixctl_init(void)
5080 static bool registered;
5086 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5087 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5091 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5092 struct ofpbuf *odp_actions, tag_type *tags,
5093 uint16_t *nf_output_iface, void *ofproto_)
5095 struct ofproto *ofproto = ofproto_;
5098 /* Drop frames for reserved multicast addresses. */
5099 if (eth_addr_is_reserved(flow->dl_dst)) {
5103 /* Learn source MAC (but don't try to learn from revalidation). */
5104 if (packet != NULL) {
5105 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5107 GRAT_ARP_LOCK_NONE);
5109 /* The log messages here could actually be useful in debugging,
5110 * so keep the rate limit relatively high. */
5111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5112 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5113 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5114 ofproto_revalidate(ofproto, rev_tag);
5118 /* Determine output port. */
5119 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5122 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5123 nf_output_iface, odp_actions);
5124 } else if (out_port != flow->in_port) {
5125 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, out_port);
5126 *nf_output_iface = out_port;
5134 static const struct ofhooks default_ofhooks = {
5135 default_normal_ofhook_cb,