2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
28 #include "classifier.h"
30 #include "discovery.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
42 #include "ofp-print.h"
44 #include "ofproto-sflow.h"
46 #include "openflow/nicira-ext.h"
47 #include "openflow/openflow.h"
48 #include "openvswitch/datapath-protocol.h"
52 #include "poll-loop.h"
56 #include "stream-ssl.h"
64 VLOG_DEFINE_THIS_MODULE(ofproto);
66 #include "sflow_api.h"
69 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
70 struct netdev *netdev;
71 struct ofp_phy_port opp; /* In host byte order. */
75 static void ofport_free(struct ofport *);
76 static void hton_ofp_phy_port(struct ofp_phy_port *);
78 static int xlate_actions(const union ofp_action *in, size_t n_in,
79 const struct flow *, struct ofproto *,
80 const struct ofpbuf *packet,
81 struct odp_actions *out, tag_type *tags,
82 bool *may_set_up_flow, uint16_t *nf_output_iface);
84 /* An OpenFlow flow. */
86 long long int used; /* Time last used; time created if not used. */
87 long long int created; /* Creation time. */
91 * - Do include packets and bytes from facets that have been deleted or
92 * whose own statistics have been folded into the rule.
94 * - Do include packets and bytes sent "by hand" that were accounted to
95 * the rule without any facet being involved (this is a rare corner
96 * case in rule_execute()).
98 * - Do not include packet or bytes that can be obtained from any facet's
99 * packet_count or byte_count member or that can be obtained from the
100 * datapath by, e.g., dpif_flow_get() for any facet.
102 uint64_t packet_count; /* Number of packets received. */
103 uint64_t byte_count; /* Number of bytes received. */
105 ovs_be64 flow_cookie; /* Controller-issued identifier. */
107 struct cls_rule cr; /* In owning ofproto's classifier. */
108 uint16_t idle_timeout; /* In seconds from time of last use. */
109 uint16_t hard_timeout; /* In seconds from time of creation. */
110 bool send_flow_removed; /* Send a flow removed message? */
111 int n_actions; /* Number of elements in actions[]. */
112 union ofp_action *actions; /* OpenFlow actions. */
113 struct list facets; /* List of "struct facet"s. */
116 static struct rule *rule_from_cls_rule(const struct cls_rule *);
117 static bool rule_is_hidden(const struct rule *);
119 static struct rule *rule_create(const struct cls_rule *,
120 const union ofp_action *, size_t n_actions,
121 uint16_t idle_timeout, uint16_t hard_timeout,
122 ovs_be64 flow_cookie, bool send_flow_removed);
123 static void rule_destroy(struct ofproto *, struct rule *);
124 static void rule_free(struct rule *);
126 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
127 static void rule_insert(struct ofproto *, struct rule *);
128 static void rule_remove(struct ofproto *, struct rule *);
130 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
132 /* An exact-match instantiation of an OpenFlow flow. */
134 long long int used; /* Time last used; time created if not used. */
138 * - Do include packets and bytes sent "by hand", e.g. with
141 * - Do include packets and bytes that were obtained from the datapath
142 * when a flow was deleted (e.g. dpif_flow_del()) or when its
143 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
145 * - Do not include any packets or bytes that can currently be obtained
146 * from the datapath by, e.g., dpif_flow_get().
148 uint64_t packet_count; /* Number of packets received. */
149 uint64_t byte_count; /* Number of bytes received. */
151 /* Number of bytes passed to account_cb. This may include bytes that can
152 * currently obtained from the datapath (thus, it can be greater than
154 uint64_t accounted_bytes;
156 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
157 struct list list_node; /* In owning rule's 'facets' list. */
158 struct rule *rule; /* Owning rule. */
159 struct flow flow; /* Exact-match flow. */
160 bool installed; /* Installed in datapath? */
161 bool may_install; /* True ordinarily; false if actions must
162 * be reassessed for every packet. */
163 int n_actions; /* Number of elements in actions[]. */
164 union odp_action *actions; /* Datapath actions. */
165 tag_type tags; /* Tags (set only by hooks). */
166 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
169 static struct facet *facet_create(struct ofproto *, struct rule *,
171 const struct ofpbuf *packet);
172 static void facet_remove(struct ofproto *, struct facet *);
173 static void facet_free(struct facet *);
175 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
176 static bool facet_revalidate(struct ofproto *, struct facet *);
178 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
179 static void facet_uninstall(struct ofproto *, struct facet *);
180 static void facet_flush_stats(struct ofproto *, struct facet *);
182 static void facet_make_actions(struct ofproto *, struct facet *,
183 const struct ofpbuf *packet);
184 static void facet_update_stats(struct ofproto *, struct facet *,
185 const struct odp_flow_stats *);
187 /* ofproto supports two kinds of OpenFlow connections:
189 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
190 * maintains persistent connections to these controllers and by default
191 * sends them asynchronous messages such as packet-ins.
193 * - "Service" connections, e.g. from ovs-ofctl. When these connections
194 * drop, it is the other side's responsibility to reconnect them if
195 * necessary. ofproto does not send them asynchronous messages by default.
197 * Currently, active (tcp, ssl, unix) connections are always "primary"
198 * connections and passive (ptcp, pssl, punix) connections are always "service"
199 * connections. There is no inherent reason for this, but it reflects the
203 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
204 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
207 /* A listener for incoming OpenFlow "service" connections. */
209 struct hmap_node node; /* In struct ofproto's "services" hmap. */
210 struct pvconn *pvconn; /* OpenFlow connection listener. */
212 /* These are not used by ofservice directly. They are settings for
213 * accepted "struct ofconn"s from the pvconn. */
214 int probe_interval; /* Max idle time before probing, in seconds. */
215 int rate_limit; /* Max packet-in rate in packets per second. */
216 int burst_limit; /* Limit on accumulating packet credits. */
219 static struct ofservice *ofservice_lookup(struct ofproto *,
221 static int ofservice_create(struct ofproto *,
222 const struct ofproto_controller *);
223 static void ofservice_reconfigure(struct ofservice *,
224 const struct ofproto_controller *);
225 static void ofservice_destroy(struct ofproto *, struct ofservice *);
227 /* An OpenFlow connection. */
229 struct ofproto *ofproto; /* The ofproto that owns this connection. */
230 struct list node; /* In struct ofproto's "all_conns" list. */
231 struct rconn *rconn; /* OpenFlow connection. */
232 enum ofconn_type type; /* Type. */
233 int flow_format; /* One of NXFF_*. */
235 /* OFPT_PACKET_IN related data. */
236 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
237 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
238 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
239 int miss_send_len; /* Bytes to send of buffered packets. */
241 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
242 * requests, and the maximum number before we stop reading OpenFlow
244 #define OFCONN_REPLY_MAX 100
245 struct rconn_packet_counter *reply_counter;
247 /* type == OFCONN_PRIMARY only. */
248 enum nx_role role; /* Role. */
249 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
250 struct discovery *discovery; /* Controller discovery object, if enabled. */
251 struct status_category *ss; /* Switch status category. */
252 enum ofproto_band band; /* In-band or out-of-band? */
255 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
256 * "schedulers" array. Their values are 0 and 1, and their meanings and values
257 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
258 * case anything ever changes, check their values here. */
259 #define N_SCHEDULERS 2
260 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
261 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
262 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
263 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
265 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
267 static void ofconn_destroy(struct ofconn *);
268 static void ofconn_run(struct ofconn *);
269 static void ofconn_wait(struct ofconn *);
270 static bool ofconn_receives_async_msgs(const struct ofconn *);
271 static char *ofconn_make_name(const struct ofproto *, const char *target);
272 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
274 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
275 struct rconn_packet_counter *counter);
277 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
278 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
282 uint64_t datapath_id; /* Datapath ID. */
283 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
284 char *mfr_desc; /* Manufacturer. */
285 char *hw_desc; /* Hardware. */
286 char *sw_desc; /* Software version. */
287 char *serial_desc; /* Serial number. */
288 char *dp_desc; /* Datapath description. */
292 struct netdev_monitor *netdev_monitor;
293 struct hmap ports; /* Contains "struct ofport"s. */
294 struct shash port_by_name;
298 struct switch_status *switch_status;
299 struct fail_open *fail_open;
300 struct netflow *netflow;
301 struct ofproto_sflow *sflow;
303 /* In-band control. */
304 struct in_band *in_band;
305 long long int next_in_band_update;
306 struct sockaddr_in *extra_in_band_remotes;
307 size_t n_extra_remotes;
311 struct classifier cls;
312 long long int next_expiration;
316 bool need_revalidate;
317 struct tag_set revalidate_set;
319 /* OpenFlow connections. */
320 struct hmap controllers; /* Controller "struct ofconn"s. */
321 struct list all_conns; /* Contains "struct ofconn"s. */
322 enum ofproto_fail_mode fail_mode;
324 /* OpenFlow listeners. */
325 struct hmap services; /* Contains "struct ofservice"s. */
326 struct pvconn **snoops;
329 /* Hooks for ovs-vswitchd. */
330 const struct ofhooks *ofhooks;
333 /* Used by default ofhooks. */
334 struct mac_learning *ml;
337 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
339 static const struct ofhooks default_ofhooks;
341 static uint64_t pick_datapath_id(const struct ofproto *);
342 static uint64_t pick_fallback_dpid(void);
344 static int ofproto_expire(struct ofproto *);
346 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
348 static void handle_openflow(struct ofconn *, struct ofpbuf *);
350 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
351 static void update_port(struct ofproto *, const char *devname);
352 static int init_ports(struct ofproto *);
353 static void reinit_ports(struct ofproto *);
356 ofproto_create(const char *datapath, const char *datapath_type,
357 const struct ofhooks *ofhooks, void *aux,
358 struct ofproto **ofprotop)
360 struct odp_stats stats;
367 /* Connect to datapath and start listening for messages. */
368 error = dpif_open(datapath, datapath_type, &dpif);
370 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
373 error = dpif_get_dp_stats(dpif, &stats);
375 VLOG_ERR("failed to obtain stats for datapath %s: %s",
376 datapath, strerror(error));
380 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
382 VLOG_ERR("failed to listen on datapath %s: %s",
383 datapath, strerror(error));
387 dpif_flow_flush(dpif);
388 dpif_recv_purge(dpif);
390 /* Initialize settings. */
391 p = xzalloc(sizeof *p);
392 p->fallback_dpid = pick_fallback_dpid();
393 p->datapath_id = p->fallback_dpid;
394 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
395 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
396 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
397 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
398 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
400 /* Initialize datapath. */
402 p->netdev_monitor = netdev_monitor_create();
403 hmap_init(&p->ports);
404 shash_init(&p->port_by_name);
405 p->max_ports = stats.max_ports;
407 /* Initialize submodules. */
408 p->switch_status = switch_status_create(p);
413 /* Initialize in-band control. */
415 p->in_band_queue = -1;
417 /* Initialize flow table. */
418 classifier_init(&p->cls);
419 p->next_expiration = time_msec() + 1000;
421 /* Initialize facet table. */
422 hmap_init(&p->facets);
423 p->need_revalidate = false;
424 tag_set_init(&p->revalidate_set);
426 /* Initialize OpenFlow connections. */
427 list_init(&p->all_conns);
428 hmap_init(&p->controllers);
429 hmap_init(&p->services);
433 /* Initialize hooks. */
435 p->ofhooks = ofhooks;
439 p->ofhooks = &default_ofhooks;
441 p->ml = mac_learning_create();
444 /* Pick final datapath ID. */
445 p->datapath_id = pick_datapath_id(p);
446 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
453 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
455 uint64_t old_dpid = p->datapath_id;
456 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
457 if (p->datapath_id != old_dpid) {
458 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
460 /* Force all active connections to reconnect, since there is no way to
461 * notify a controller that the datapath ID has changed. */
462 ofproto_reconnect_controllers(p);
467 is_discovery_controller(const struct ofproto_controller *c)
469 return !strcmp(c->target, "discover");
473 is_in_band_controller(const struct ofproto_controller *c)
475 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
478 /* Creates a new controller in 'ofproto'. Some of the settings are initially
479 * drawn from 'c', but update_controller() needs to be called later to finish
480 * the new ofconn's configuration. */
482 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
484 struct discovery *discovery;
485 struct ofconn *ofconn;
487 if (is_discovery_controller(c)) {
488 int error = discovery_create(c->accept_re, c->update_resolv_conf,
489 ofproto->dpif, ofproto->switch_status,
498 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
499 ofconn->pktbuf = pktbuf_create();
500 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
502 ofconn->discovery = discovery;
504 char *name = ofconn_make_name(ofproto, c->target);
505 rconn_connect(ofconn->rconn, c->target, name);
508 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
509 hash_string(c->target, 0));
512 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
513 * target or turn discovery on or off (these are done by creating new ofconns
514 * and deleting old ones), but it can update the rest of an ofconn's
517 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
521 ofconn->band = (is_in_band_controller(c)
522 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
524 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
526 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
527 rconn_set_probe_interval(ofconn->rconn, probe_interval);
529 if (ofconn->discovery) {
530 discovery_set_update_resolv_conf(ofconn->discovery,
531 c->update_resolv_conf);
532 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
535 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
539 ofconn_get_target(const struct ofconn *ofconn)
541 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
544 static struct ofconn *
545 find_controller_by_target(struct ofproto *ofproto, const char *target)
547 struct ofconn *ofconn;
549 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
550 hash_string(target, 0), &ofproto->controllers) {
551 if (!strcmp(ofconn_get_target(ofconn), target)) {
559 update_in_band_remotes(struct ofproto *ofproto)
561 const struct ofconn *ofconn;
562 struct sockaddr_in *addrs;
563 size_t max_addrs, n_addrs;
567 /* Allocate enough memory for as many remotes as we could possibly have. */
568 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
569 addrs = xmalloc(max_addrs * sizeof *addrs);
572 /* Add all the remotes. */
574 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
575 struct sockaddr_in *sin = &addrs[n_addrs];
577 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
581 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
582 if (sin->sin_addr.s_addr) {
583 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
586 if (ofconn->discovery) {
590 for (i = 0; i < ofproto->n_extra_remotes; i++) {
591 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
594 /* Create or update or destroy in-band.
596 * Ordinarily we only enable in-band if there's at least one remote
597 * address, but discovery needs the in-band rules for DHCP to be installed
598 * even before we know any remote addresses. */
599 if (n_addrs || discovery) {
600 if (!ofproto->in_band) {
601 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
604 if (ofproto->in_band) {
605 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
607 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
608 ofproto->next_in_band_update = time_msec() + 1000;
610 in_band_destroy(ofproto->in_band);
611 ofproto->in_band = NULL;
619 update_fail_open(struct ofproto *p)
621 struct ofconn *ofconn;
623 if (!hmap_is_empty(&p->controllers)
624 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
625 struct rconn **rconns;
629 p->fail_open = fail_open_create(p, p->switch_status);
633 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
634 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
635 rconns[n++] = ofconn->rconn;
638 fail_open_set_controllers(p->fail_open, rconns, n);
639 /* p->fail_open takes ownership of 'rconns'. */
641 fail_open_destroy(p->fail_open);
647 ofproto_set_controllers(struct ofproto *p,
648 const struct ofproto_controller *controllers,
649 size_t n_controllers)
651 struct shash new_controllers;
652 struct ofconn *ofconn, *next_ofconn;
653 struct ofservice *ofservice, *next_ofservice;
657 /* Create newly configured controllers and services.
658 * Create a name to ofproto_controller mapping in 'new_controllers'. */
659 shash_init(&new_controllers);
660 for (i = 0; i < n_controllers; i++) {
661 const struct ofproto_controller *c = &controllers[i];
663 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
664 if (!find_controller_by_target(p, c->target)) {
665 add_controller(p, c);
667 } else if (!pvconn_verify_name(c->target)) {
668 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
672 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
673 dpif_name(p->dpif), c->target);
677 shash_add_once(&new_controllers, c->target, &controllers[i]);
680 /* Delete controllers that are no longer configured.
681 * Update configuration of all now-existing controllers. */
683 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
684 struct ofproto_controller *c;
686 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
688 ofconn_destroy(ofconn);
690 update_controller(ofconn, c);
697 /* Delete services that are no longer configured.
698 * Update configuration of all now-existing services. */
699 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
700 struct ofproto_controller *c;
702 c = shash_find_data(&new_controllers,
703 pvconn_get_name(ofservice->pvconn));
705 ofservice_destroy(p, ofservice);
707 ofservice_reconfigure(ofservice, c);
711 shash_destroy(&new_controllers);
713 update_in_band_remotes(p);
716 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
717 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
718 struct ofconn, hmap_node);
719 ofconn->ss = switch_status_register(p->switch_status, "remote",
720 rconn_status_cb, ofconn->rconn);
725 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
727 p->fail_mode = fail_mode;
731 /* Drops the connections between 'ofproto' and all of its controllers, forcing
732 * them to reconnect. */
734 ofproto_reconnect_controllers(struct ofproto *ofproto)
736 struct ofconn *ofconn;
738 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
739 rconn_reconnect(ofconn->rconn);
744 any_extras_changed(const struct ofproto *ofproto,
745 const struct sockaddr_in *extras, size_t n)
749 if (n != ofproto->n_extra_remotes) {
753 for (i = 0; i < n; i++) {
754 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
755 const struct sockaddr_in *new = &extras[i];
757 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
758 old->sin_port != new->sin_port) {
766 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
767 * in-band control should guarantee access, in the same way that in-band
768 * control guarantees access to OpenFlow controllers. */
770 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
771 const struct sockaddr_in *extras, size_t n)
773 if (!any_extras_changed(ofproto, extras, n)) {
777 free(ofproto->extra_in_band_remotes);
778 ofproto->n_extra_remotes = n;
779 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
781 update_in_band_remotes(ofproto);
784 /* Sets the OpenFlow queue used by flows set up by in-band control on
785 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
786 * flows will use the default queue. */
788 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
790 if (queue_id != ofproto->in_band_queue) {
791 ofproto->in_band_queue = queue_id;
792 update_in_band_remotes(ofproto);
797 ofproto_set_desc(struct ofproto *p,
798 const char *mfr_desc, const char *hw_desc,
799 const char *sw_desc, const char *serial_desc,
802 struct ofp_desc_stats *ods;
805 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
806 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
807 sizeof ods->mfr_desc);
810 p->mfr_desc = xstrdup(mfr_desc);
813 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
814 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
815 sizeof ods->hw_desc);
818 p->hw_desc = xstrdup(hw_desc);
821 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
822 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
823 sizeof ods->sw_desc);
826 p->sw_desc = xstrdup(sw_desc);
829 if (strlen(serial_desc) >= sizeof ods->serial_num) {
830 VLOG_WARN("truncating serial_desc, must be less than %zu "
832 sizeof ods->serial_num);
834 free(p->serial_desc);
835 p->serial_desc = xstrdup(serial_desc);
838 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
839 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
840 sizeof ods->dp_desc);
843 p->dp_desc = xstrdup(dp_desc);
848 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
849 const struct svec *svec)
851 struct pvconn **pvconns = *pvconnsp;
852 size_t n_pvconns = *n_pvconnsp;
856 for (i = 0; i < n_pvconns; i++) {
857 pvconn_close(pvconns[i]);
861 pvconns = xmalloc(svec->n * sizeof *pvconns);
863 for (i = 0; i < svec->n; i++) {
864 const char *name = svec->names[i];
865 struct pvconn *pvconn;
868 error = pvconn_open(name, &pvconn);
870 pvconns[n_pvconns++] = pvconn;
872 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
880 *n_pvconnsp = n_pvconns;
886 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
888 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
892 ofproto_set_netflow(struct ofproto *ofproto,
893 const struct netflow_options *nf_options)
895 if (nf_options && nf_options->collectors.n) {
896 if (!ofproto->netflow) {
897 ofproto->netflow = netflow_create();
899 return netflow_set_options(ofproto->netflow, nf_options);
901 netflow_destroy(ofproto->netflow);
902 ofproto->netflow = NULL;
908 ofproto_set_sflow(struct ofproto *ofproto,
909 const struct ofproto_sflow_options *oso)
911 struct ofproto_sflow *os = ofproto->sflow;
914 struct ofport *ofport;
916 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
917 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
918 ofproto_sflow_add_port(os, ofport->odp_port,
919 netdev_get_name(ofport->netdev));
922 ofproto_sflow_set_options(os, oso);
924 ofproto_sflow_destroy(os);
925 ofproto->sflow = NULL;
930 ofproto_get_datapath_id(const struct ofproto *ofproto)
932 return ofproto->datapath_id;
936 ofproto_has_primary_controller(const struct ofproto *ofproto)
938 return !hmap_is_empty(&ofproto->controllers);
941 enum ofproto_fail_mode
942 ofproto_get_fail_mode(const struct ofproto *p)
948 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
952 for (i = 0; i < ofproto->n_snoops; i++) {
953 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
958 ofproto_destroy(struct ofproto *p)
960 struct ofservice *ofservice, *next_ofservice;
961 struct ofconn *ofconn, *next_ofconn;
962 struct ofport *ofport, *next_ofport;
969 /* Destroy fail-open and in-band early, since they touch the classifier. */
970 fail_open_destroy(p->fail_open);
973 in_band_destroy(p->in_band);
975 free(p->extra_in_band_remotes);
977 ofproto_flush_flows(p);
978 classifier_destroy(&p->cls);
979 hmap_destroy(&p->facets);
981 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
982 ofconn_destroy(ofconn);
984 hmap_destroy(&p->controllers);
987 netdev_monitor_destroy(p->netdev_monitor);
988 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
989 hmap_remove(&p->ports, &ofport->hmap_node);
992 shash_destroy(&p->port_by_name);
994 switch_status_destroy(p->switch_status);
995 netflow_destroy(p->netflow);
996 ofproto_sflow_destroy(p->sflow);
998 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
999 ofservice_destroy(p, ofservice);
1001 hmap_destroy(&p->services);
1003 for (i = 0; i < p->n_snoops; i++) {
1004 pvconn_close(p->snoops[i]);
1008 mac_learning_destroy(p->ml);
1013 free(p->serial_desc);
1016 hmap_destroy(&p->ports);
1022 ofproto_run(struct ofproto *p)
1024 int error = ofproto_run1(p);
1026 error = ofproto_run2(p, false);
1032 process_port_change(struct ofproto *ofproto, int error, char *devname)
1034 if (error == ENOBUFS) {
1035 reinit_ports(ofproto);
1036 } else if (!error) {
1037 update_port(ofproto, devname);
1042 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1043 * means that 'ofconn' is more interesting for monitoring than a lower return
1046 snoop_preference(const struct ofconn *ofconn)
1048 switch (ofconn->role) {
1049 case NX_ROLE_MASTER:
1056 /* Shouldn't happen. */
1061 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1062 * Connects this vconn to a controller. */
1064 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1066 struct ofconn *ofconn, *best;
1068 /* Pick a controller for monitoring. */
1070 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1071 if (ofconn->type == OFCONN_PRIMARY
1072 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1078 rconn_add_monitor(best->rconn, vconn);
1080 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1086 ofproto_run1(struct ofproto *p)
1088 struct ofconn *ofconn, *next_ofconn;
1089 struct ofservice *ofservice;
1094 if (shash_is_empty(&p->port_by_name)) {
1098 for (i = 0; i < 50; i++) {
1101 error = dpif_recv(p->dpif, &buf);
1103 if (error == ENODEV) {
1104 /* Someone destroyed the datapath behind our back. The caller
1105 * better destroy us and give up, because we're just going to
1106 * spin from here on out. */
1107 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1108 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1109 dpif_name(p->dpif));
1115 handle_odp_msg(p, buf);
1118 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1119 process_port_change(p, error, devname);
1121 while ((error = netdev_monitor_poll(p->netdev_monitor,
1122 &devname)) != EAGAIN) {
1123 process_port_change(p, error, devname);
1127 if (time_msec() >= p->next_in_band_update) {
1128 update_in_band_remotes(p);
1130 in_band_run(p->in_band);
1133 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1137 /* Fail-open maintenance. Do this after processing the ofconns since
1138 * fail-open checks the status of the controller rconn. */
1140 fail_open_run(p->fail_open);
1143 HMAP_FOR_EACH (ofservice, node, &p->services) {
1144 struct vconn *vconn;
1147 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1149 struct rconn *rconn;
1152 rconn = rconn_create(ofservice->probe_interval, 0);
1153 name = ofconn_make_name(p, vconn_get_name(vconn));
1154 rconn_connect_unreliably(rconn, vconn, name);
1157 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1158 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1159 ofservice->burst_limit);
1160 } else if (retval != EAGAIN) {
1161 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1165 for (i = 0; i < p->n_snoops; i++) {
1166 struct vconn *vconn;
1169 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1171 add_snooper(p, vconn);
1172 } else if (retval != EAGAIN) {
1173 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1177 if (time_msec() >= p->next_expiration) {
1178 int delay = ofproto_expire(p);
1179 p->next_expiration = time_msec() + delay;
1180 COVERAGE_INC(ofproto_expiration);
1184 netflow_run(p->netflow);
1187 ofproto_sflow_run(p->sflow);
1194 ofproto_run2(struct ofproto *p, bool revalidate_all)
1196 /* Figure out what we need to revalidate now, if anything. */
1197 struct tag_set revalidate_set = p->revalidate_set;
1198 if (p->need_revalidate) {
1199 revalidate_all = true;
1202 /* Clear the revalidation flags. */
1203 tag_set_init(&p->revalidate_set);
1204 p->need_revalidate = false;
1206 /* Now revalidate if there's anything to do. */
1207 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1208 struct facet *facet, *next;
1210 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1212 || tag_set_intersects(&revalidate_set, facet->tags)) {
1213 facet_revalidate(p, facet);
1222 ofproto_wait(struct ofproto *p)
1224 struct ofservice *ofservice;
1225 struct ofconn *ofconn;
1228 dpif_recv_wait(p->dpif);
1229 dpif_port_poll_wait(p->dpif);
1230 netdev_monitor_poll_wait(p->netdev_monitor);
1231 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1232 ofconn_wait(ofconn);
1235 poll_timer_wait_until(p->next_in_band_update);
1236 in_band_wait(p->in_band);
1239 fail_open_wait(p->fail_open);
1242 ofproto_sflow_wait(p->sflow);
1244 if (!tag_set_is_empty(&p->revalidate_set)) {
1245 poll_immediate_wake();
1247 if (p->need_revalidate) {
1248 /* Shouldn't happen, but if it does just go around again. */
1249 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1250 poll_immediate_wake();
1251 } else if (p->next_expiration != LLONG_MAX) {
1252 poll_timer_wait_until(p->next_expiration);
1254 HMAP_FOR_EACH (ofservice, node, &p->services) {
1255 pvconn_wait(ofservice->pvconn);
1257 for (i = 0; i < p->n_snoops; i++) {
1258 pvconn_wait(p->snoops[i]);
1263 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1265 tag_set_add(&ofproto->revalidate_set, tag);
1269 ofproto_get_revalidate_set(struct ofproto *ofproto)
1271 return &ofproto->revalidate_set;
1275 ofproto_is_alive(const struct ofproto *p)
1277 return !hmap_is_empty(&p->controllers);
1280 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1282 * This is almost the same as calling dpif_port_del() directly on the
1283 * datapath, but it also makes 'ofproto' close its open netdev for the port
1284 * (if any). This makes it possible to create a new netdev of a different
1285 * type under the same name, which otherwise the netdev library would refuse
1286 * to do because of the conflict. (The netdev would eventually get closed on
1287 * the next trip through ofproto_run(), but this interface is more direct.)
1289 * Returns 0 if successful, otherwise a positive errno. */
1291 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1293 struct ofport *ofport = get_port(ofproto, odp_port);
1294 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1297 error = dpif_port_del(ofproto->dpif, odp_port);
1299 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1300 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1301 } else if (ofport) {
1302 /* 'name' is ofport->opp.name and update_port() is going to destroy
1303 * 'ofport'. Just in case update_port() refers to 'name' after it
1304 * destroys 'ofport', make a copy of it around the update_port()
1306 char *devname = xstrdup(name);
1307 update_port(ofproto, devname);
1313 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1314 * true if 'odp_port' exists and should be included, false otherwise. */
1316 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1318 struct ofport *ofport = get_port(ofproto, odp_port);
1319 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1323 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1324 const union ofp_action *actions, size_t n_actions,
1325 const struct ofpbuf *packet)
1327 struct odp_actions odp_actions;
1330 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1336 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1338 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
1342 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1343 * performs the 'n_actions' actions in 'actions'. The new flow will not
1346 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1347 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1348 * controllers; otherwise, it will be hidden.
1350 * The caller retains ownership of 'cls_rule' and 'actions'. */
1352 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1353 const union ofp_action *actions, size_t n_actions)
1356 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1357 rule_insert(p, rule);
1361 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1365 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1368 rule_remove(ofproto, rule);
1373 ofproto_flush_flows(struct ofproto *ofproto)
1375 struct facet *facet, *next_facet;
1376 struct rule *rule, *next_rule;
1377 struct cls_cursor cursor;
1379 COVERAGE_INC(ofproto_flush);
1381 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1382 /* Mark the facet as not installed so that facet_remove() doesn't
1383 * bother trying to uninstall it. There is no point in uninstalling it
1384 * individually since we are about to blow away all the facets with
1385 * dpif_flow_flush(). */
1386 facet->installed = false;
1387 facet_remove(ofproto, facet);
1390 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1391 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1392 rule_remove(ofproto, rule);
1395 dpif_flow_flush(ofproto->dpif);
1396 if (ofproto->in_band) {
1397 in_band_flushed(ofproto->in_band);
1399 if (ofproto->fail_open) {
1400 fail_open_flushed(ofproto->fail_open);
1405 reinit_ports(struct ofproto *p)
1407 struct svec devnames;
1408 struct ofport *ofport;
1409 struct odp_port *odp_ports;
1413 COVERAGE_INC(ofproto_reinit_ports);
1415 svec_init(&devnames);
1416 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1417 svec_add (&devnames, (char *) ofport->opp.name);
1419 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1420 for (i = 0; i < n_odp_ports; i++) {
1421 svec_add (&devnames, odp_ports[i].devname);
1425 svec_sort_unique(&devnames);
1426 for (i = 0; i < devnames.n; i++) {
1427 update_port(p, devnames.names[i]);
1429 svec_destroy(&devnames);
1432 static struct ofport *
1433 make_ofport(const struct odp_port *odp_port)
1435 struct netdev_options netdev_options;
1436 enum netdev_flags flags;
1437 struct ofport *ofport;
1438 struct netdev *netdev;
1441 memset(&netdev_options, 0, sizeof netdev_options);
1442 netdev_options.name = odp_port->devname;
1443 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1445 error = netdev_open(&netdev_options, &netdev);
1447 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1448 "cannot be opened (%s)",
1449 odp_port->devname, odp_port->port,
1450 odp_port->devname, strerror(error));
1454 ofport = xmalloc(sizeof *ofport);
1455 ofport->netdev = netdev;
1456 ofport->odp_port = odp_port->port;
1457 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1458 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1459 memcpy(ofport->opp.name, odp_port->devname,
1460 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1461 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1463 netdev_get_flags(netdev, &flags);
1464 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1466 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1468 netdev_get_features(netdev,
1469 &ofport->opp.curr, &ofport->opp.advertised,
1470 &ofport->opp.supported, &ofport->opp.peer);
1475 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1477 if (get_port(p, odp_port->port)) {
1478 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1481 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1482 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1491 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1493 const struct ofp_phy_port *a = &a_->opp;
1494 const struct ofp_phy_port *b = &b_->opp;
1496 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1497 return (a->port_no == b->port_no
1498 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1499 && !strcmp((char *) a->name, (char *) b->name)
1500 && a->state == b->state
1501 && a->config == b->config
1502 && a->curr == b->curr
1503 && a->advertised == b->advertised
1504 && a->supported == b->supported
1505 && a->peer == b->peer);
1509 send_port_status(struct ofproto *p, const struct ofport *ofport,
1512 /* XXX Should limit the number of queued port status change messages. */
1513 struct ofconn *ofconn;
1514 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1515 struct ofp_port_status *ops;
1518 if (!ofconn_receives_async_msgs(ofconn)) {
1522 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1523 ops->reason = reason;
1524 ops->desc = ofport->opp;
1525 hton_ofp_phy_port(&ops->desc);
1526 queue_tx(b, ofconn, NULL);
1531 ofport_install(struct ofproto *p, struct ofport *ofport)
1533 const char *netdev_name = (const char *) ofport->opp.name;
1535 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1536 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1537 shash_add(&p->port_by_name, netdev_name, ofport);
1539 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1544 ofport_remove(struct ofproto *p, struct ofport *ofport)
1546 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1547 hmap_remove(&p->ports, &ofport->hmap_node);
1548 shash_delete(&p->port_by_name,
1549 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1551 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1556 ofport_free(struct ofport *ofport)
1559 netdev_close(ofport->netdev);
1564 static struct ofport *
1565 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1567 struct ofport *port;
1569 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1570 hash_int(odp_port, 0), &ofproto->ports) {
1571 if (port->odp_port == odp_port) {
1579 update_port(struct ofproto *p, const char *devname)
1581 struct odp_port odp_port;
1582 struct ofport *old_ofport;
1583 struct ofport *new_ofport;
1586 COVERAGE_INC(ofproto_update_port);
1588 /* Query the datapath for port information. */
1589 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1591 /* Find the old ofport. */
1592 old_ofport = shash_find_data(&p->port_by_name, devname);
1595 /* There's no port named 'devname' but there might be a port with
1596 * the same port number. This could happen if a port is deleted
1597 * and then a new one added in its place very quickly, or if a port
1598 * is renamed. In the former case we want to send an OFPPR_DELETE
1599 * and an OFPPR_ADD, and in the latter case we want to send a
1600 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1601 * the old port's ifindex against the new port, or perhaps less
1602 * reliably but more portably by comparing the old port's MAC
1603 * against the new port's MAC. However, this code isn't that smart
1604 * and always sends an OFPPR_MODIFY (XXX). */
1605 old_ofport = get_port(p, odp_port.port);
1607 } else if (error != ENOENT && error != ENODEV) {
1608 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1609 "%s", strerror(error));
1613 /* Create a new ofport. */
1614 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1616 /* Eliminate a few pathological cases. */
1617 if (!old_ofport && !new_ofport) {
1619 } else if (old_ofport && new_ofport) {
1620 /* Most of the 'config' bits are OpenFlow soft state, but
1621 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1622 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1623 * leaves the other bits 0.) */
1624 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1626 if (ofport_equal(old_ofport, new_ofport)) {
1627 /* False alarm--no change. */
1628 ofport_free(new_ofport);
1633 /* Now deal with the normal cases. */
1635 ofport_remove(p, old_ofport);
1638 ofport_install(p, new_ofport);
1640 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1641 (!old_ofport ? OFPPR_ADD
1642 : !new_ofport ? OFPPR_DELETE
1644 ofport_free(old_ofport);
1648 init_ports(struct ofproto *p)
1650 struct odp_port *ports;
1655 error = dpif_port_list(p->dpif, &ports, &n_ports);
1660 for (i = 0; i < n_ports; i++) {
1661 const struct odp_port *odp_port = &ports[i];
1662 if (!ofport_conflicts(p, odp_port)) {
1663 struct ofport *ofport = make_ofport(odp_port);
1665 ofport_install(p, ofport);
1673 static struct ofconn *
1674 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1676 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1677 ofconn->ofproto = p;
1678 list_push_back(&p->all_conns, &ofconn->node);
1679 ofconn->rconn = rconn;
1680 ofconn->type = type;
1681 ofconn->flow_format = NXFF_OPENFLOW10;
1682 ofconn->role = NX_ROLE_OTHER;
1683 ofconn->packet_in_counter = rconn_packet_counter_create ();
1684 ofconn->pktbuf = NULL;
1685 ofconn->miss_send_len = 0;
1686 ofconn->reply_counter = rconn_packet_counter_create ();
1691 ofconn_destroy(struct ofconn *ofconn)
1693 if (ofconn->type == OFCONN_PRIMARY) {
1694 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1696 discovery_destroy(ofconn->discovery);
1698 list_remove(&ofconn->node);
1699 switch_status_unregister(ofconn->ss);
1700 rconn_destroy(ofconn->rconn);
1701 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1702 rconn_packet_counter_destroy(ofconn->reply_counter);
1703 pktbuf_destroy(ofconn->pktbuf);
1708 ofconn_run(struct ofconn *ofconn)
1710 struct ofproto *p = ofconn->ofproto;
1714 if (ofconn->discovery) {
1715 char *controller_name;
1716 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1717 discovery_question_connectivity(ofconn->discovery);
1719 if (discovery_run(ofconn->discovery, &controller_name)) {
1720 if (controller_name) {
1721 char *ofconn_name = ofconn_make_name(p, controller_name);
1722 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1725 rconn_disconnect(ofconn->rconn);
1730 for (i = 0; i < N_SCHEDULERS; i++) {
1731 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1734 rconn_run(ofconn->rconn);
1736 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1737 /* Limit the number of iterations to prevent other tasks from
1739 for (iteration = 0; iteration < 50; iteration++) {
1740 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1745 fail_open_maybe_recover(p->fail_open);
1747 handle_openflow(ofconn, of_msg);
1748 ofpbuf_delete(of_msg);
1752 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1753 ofconn_destroy(ofconn);
1758 ofconn_wait(struct ofconn *ofconn)
1762 if (ofconn->discovery) {
1763 discovery_wait(ofconn->discovery);
1765 for (i = 0; i < N_SCHEDULERS; i++) {
1766 pinsched_wait(ofconn->schedulers[i]);
1768 rconn_run_wait(ofconn->rconn);
1769 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1770 rconn_recv_wait(ofconn->rconn);
1772 COVERAGE_INC(ofproto_ofconn_stuck);
1776 /* Returns true if 'ofconn' should receive asynchronous messages. */
1778 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1780 if (ofconn->type == OFCONN_PRIMARY) {
1781 /* Primary controllers always get asynchronous messages unless they
1782 * have configured themselves as "slaves". */
1783 return ofconn->role != NX_ROLE_SLAVE;
1785 /* Service connections don't get asynchronous messages unless they have
1786 * explicitly asked for them by setting a nonzero miss send length. */
1787 return ofconn->miss_send_len > 0;
1791 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1792 * and 'target', suitable for use in log messages for identifying the
1795 * The name is dynamically allocated. The caller should free it (with free())
1796 * when it is no longer needed. */
1798 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1800 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1804 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1808 for (i = 0; i < N_SCHEDULERS; i++) {
1809 struct pinsched **s = &ofconn->schedulers[i];
1813 *s = pinsched_create(rate, burst,
1814 ofconn->ofproto->switch_status);
1816 pinsched_set_limits(*s, rate, burst);
1819 pinsched_destroy(*s);
1826 ofservice_reconfigure(struct ofservice *ofservice,
1827 const struct ofproto_controller *c)
1829 ofservice->probe_interval = c->probe_interval;
1830 ofservice->rate_limit = c->rate_limit;
1831 ofservice->burst_limit = c->burst_limit;
1834 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1835 * positive errno value. */
1837 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1839 struct ofservice *ofservice;
1840 struct pvconn *pvconn;
1843 error = pvconn_open(c->target, &pvconn);
1848 ofservice = xzalloc(sizeof *ofservice);
1849 hmap_insert(&ofproto->services, &ofservice->node,
1850 hash_string(c->target, 0));
1851 ofservice->pvconn = pvconn;
1853 ofservice_reconfigure(ofservice, c);
1859 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1861 hmap_remove(&ofproto->services, &ofservice->node);
1862 pvconn_close(ofservice->pvconn);
1866 /* Finds and returns the ofservice within 'ofproto' that has the given
1867 * 'target', or a null pointer if none exists. */
1868 static struct ofservice *
1869 ofservice_lookup(struct ofproto *ofproto, const char *target)
1871 struct ofservice *ofservice;
1873 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1874 &ofproto->services) {
1875 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1882 /* Returns true if 'rule' should be hidden from the controller.
1884 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1885 * (e.g. by in-band control) and are intentionally hidden from the
1888 rule_is_hidden(const struct rule *rule)
1890 return rule->cr.priority > UINT16_MAX;
1893 /* Creates and returns a new rule initialized as specified.
1895 * The caller is responsible for inserting the rule into the classifier (with
1896 * rule_insert()). */
1897 static struct rule *
1898 rule_create(const struct cls_rule *cls_rule,
1899 const union ofp_action *actions, size_t n_actions,
1900 uint16_t idle_timeout, uint16_t hard_timeout,
1901 ovs_be64 flow_cookie, bool send_flow_removed)
1903 struct rule *rule = xzalloc(sizeof *rule);
1904 rule->cr = *cls_rule;
1905 rule->idle_timeout = idle_timeout;
1906 rule->hard_timeout = hard_timeout;
1907 rule->flow_cookie = flow_cookie;
1908 rule->used = rule->created = time_msec();
1909 rule->send_flow_removed = send_flow_removed;
1910 list_init(&rule->facets);
1911 if (n_actions > 0) {
1912 rule->n_actions = n_actions;
1913 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1919 static struct rule *
1920 rule_from_cls_rule(const struct cls_rule *cls_rule)
1922 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1926 rule_free(struct rule *rule)
1928 free(rule->actions);
1932 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1933 * destroying any that no longer has a rule (which is probably all of them).
1935 * The caller must have already removed 'rule' from the classifier. */
1937 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1939 struct facet *facet, *next_facet;
1940 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1941 facet_revalidate(ofproto, facet);
1946 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1947 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1950 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1952 const union ofp_action *oa;
1953 struct actions_iterator i;
1955 if (out_port == htons(OFPP_NONE)) {
1958 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1959 oa = actions_next(&i)) {
1960 if (action_outputs_to_port(oa, out_port)) {
1967 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1968 * 'packet', which arrived on 'in_port'.
1970 * Takes ownership of 'packet'. */
1972 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1973 const union odp_action *actions, size_t n_actions,
1974 struct ofpbuf *packet)
1976 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1977 /* As an optimization, avoid a round-trip from userspace to kernel to
1978 * userspace. This also avoids possibly filling up kernel packet
1979 * buffers along the way. */
1980 struct odp_msg *msg;
1982 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1983 msg->type = _ODPL_ACTION_NR;
1984 msg->length = sizeof(struct odp_msg) + packet->size;
1985 msg->port = in_port;
1987 msg->arg = actions[0].controller.arg;
1989 send_packet_in(ofproto, packet);
1995 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
1996 ofpbuf_delete(packet);
2001 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2002 * statistics appropriately. 'packet' must have at least sizeof(struct
2003 * ofp_packet_in) bytes of headroom.
2005 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2006 * applying flow_extract() to 'packet' would yield the same flow as
2009 * 'facet' must have accurately composed ODP actions; that is, it must not be
2010 * in need of revalidation.
2012 * Takes ownership of 'packet'. */
2014 facet_execute(struct ofproto *ofproto, struct facet *facet,
2015 struct ofpbuf *packet)
2017 struct odp_flow_stats stats;
2019 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2021 flow_extract_stats(&facet->flow, packet, &stats);
2022 if (execute_odp_actions(ofproto, facet->flow.in_port,
2023 facet->actions, facet->n_actions, packet)) {
2024 facet_update_stats(ofproto, facet, &stats);
2025 facet->used = time_msec();
2026 netflow_flow_update_time(ofproto->netflow,
2027 &facet->nf_flow, facet->used);
2031 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2032 * statistics (or the statistics for one of its facets) appropriately.
2033 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2035 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2036 * with statistics for 'packet' either way.
2038 * Takes ownership of 'packet'. */
2040 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2041 struct ofpbuf *packet)
2043 struct facet *facet;
2044 struct odp_actions a;
2048 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2050 flow_extract(packet, 0, in_port, &flow);
2052 /* First look for a related facet. If we find one, account it to that. */
2053 facet = facet_lookup_valid(ofproto, &flow);
2054 if (facet && facet->rule == rule) {
2055 facet_execute(ofproto, facet, packet);
2059 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2060 * create a new facet for it and use that. */
2061 if (rule_lookup(ofproto, &flow) == rule) {
2062 facet = facet_create(ofproto, rule, &flow, packet);
2063 facet_execute(ofproto, facet, packet);
2064 facet_install(ofproto, facet, true);
2068 /* We can't account anything to a facet. If we were to try, then that
2069 * facet would have a non-matching rule, busting our invariants. */
2070 if (xlate_actions(rule->actions, rule->n_actions, &flow, ofproto,
2071 packet, &a, NULL, 0, NULL)) {
2072 ofpbuf_delete(packet);
2075 size = packet->size;
2076 if (execute_odp_actions(ofproto, in_port,
2077 a.actions, a.n_actions, packet)) {
2078 rule->used = time_msec();
2079 rule->packet_count++;
2080 rule->byte_count += size;
2084 /* Inserts 'rule' into 'p''s flow table. */
2086 rule_insert(struct ofproto *p, struct rule *rule)
2088 struct rule *displaced_rule;
2090 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2091 if (displaced_rule) {
2092 rule_destroy(p, displaced_rule);
2094 p->need_revalidate = true;
2097 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2098 * 'flow' and an example 'packet' within that flow.
2100 * The caller must already have determined that no facet with an identical
2101 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2102 * 'ofproto''s classifier table. */
2103 static struct facet *
2104 facet_create(struct ofproto *ofproto, struct rule *rule,
2105 const struct flow *flow, const struct ofpbuf *packet)
2107 struct facet *facet;
2109 facet = xzalloc(sizeof *facet);
2110 facet->used = time_msec();
2111 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2112 list_push_back(&rule->facets, &facet->list_node);
2114 facet->flow = *flow;
2115 netflow_flow_init(&facet->nf_flow);
2116 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2118 facet_make_actions(ofproto, facet, packet);
2124 facet_free(struct facet *facet)
2126 free(facet->actions);
2130 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2132 * - Removes 'rule' from the classifier.
2134 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2135 * destroys them), via rule_destroy().
2138 rule_remove(struct ofproto *ofproto, struct rule *rule)
2140 COVERAGE_INC(ofproto_del_rule);
2141 ofproto->need_revalidate = true;
2142 classifier_remove(&ofproto->cls, &rule->cr);
2143 rule_destroy(ofproto, rule);
2146 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2148 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2149 * rule's statistics, via facet_uninstall().
2151 * - Removes 'facet' from its rule and from ofproto->facets.
2154 facet_remove(struct ofproto *ofproto, struct facet *facet)
2156 facet_uninstall(ofproto, facet);
2157 facet_flush_stats(ofproto, facet);
2158 hmap_remove(&ofproto->facets, &facet->hmap_node);
2159 list_remove(&facet->list_node);
2163 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2165 facet_make_actions(struct ofproto *p, struct facet *facet,
2166 const struct ofpbuf *packet)
2168 const struct rule *rule = facet->rule;
2169 struct odp_actions a;
2172 xlate_actions(rule->actions, rule->n_actions, &facet->flow, p,
2173 packet, &a, &facet->tags, &facet->may_install,
2174 &facet->nf_flow.output_iface);
2176 actions_len = a.n_actions * sizeof *a.actions;
2177 if (facet->n_actions != a.n_actions
2178 || memcmp(facet->actions, a.actions, actions_len)) {
2179 free(facet->actions);
2180 facet->n_actions = a.n_actions;
2181 facet->actions = xmemdup(a.actions, actions_len);
2186 facet_put__(struct ofproto *ofproto, struct facet *facet, int flags,
2187 struct odp_flow_put *put)
2189 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2190 odp_flow_key_from_flow(&put->flow.key, &facet->flow);
2191 put->flow.actions = facet->actions;
2192 put->flow.n_actions = facet->n_actions;
2193 put->flow.flags = 0;
2195 return dpif_flow_put(ofproto->dpif, put);
2198 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2199 * 'zero_stats' is true, clears any existing statistics from the datapath for
2202 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2204 if (facet->may_install) {
2205 struct odp_flow_put put;
2208 flags = ODPPF_CREATE | ODPPF_MODIFY;
2210 flags |= ODPPF_ZERO_STATS;
2212 if (!facet_put__(p, facet, flags, &put)) {
2213 facet->installed = true;
2218 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2219 * to the accounting hook function in the ofhooks structure. */
2221 facet_account(struct ofproto *ofproto,
2222 struct facet *facet, uint64_t extra_bytes)
2224 uint64_t total_bytes = facet->byte_count + extra_bytes;
2226 if (ofproto->ofhooks->account_flow_cb
2227 && total_bytes > facet->accounted_bytes)
2229 ofproto->ofhooks->account_flow_cb(
2230 &facet->flow, facet->tags, facet->actions, facet->n_actions,
2231 total_bytes - facet->accounted_bytes, ofproto->aux);
2232 facet->accounted_bytes = total_bytes;
2236 /* If 'rule' is installed in the datapath, uninstalls it. */
2238 facet_uninstall(struct ofproto *p, struct facet *facet)
2240 if (facet->installed) {
2241 struct odp_flow odp_flow;
2243 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
2244 odp_flow.actions = NULL;
2245 odp_flow.n_actions = 0;
2247 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2248 facet_update_stats(p, facet, &odp_flow.stats);
2250 facet->installed = false;
2254 /* Returns true if the only action for 'facet' is to send to the controller.
2255 * (We don't report NetFlow expiration messages for such facets because they
2256 * are just part of the control logic for the network, not real traffic). */
2258 facet_is_controller_flow(struct facet *facet)
2261 && facet->rule->n_actions == 1
2262 && action_outputs_to_port(&facet->rule->actions[0],
2263 htons(OFPP_CONTROLLER)));
2266 /* Folds all of 'facet''s statistics into its rule. Also updates the
2267 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2269 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2271 facet_account(ofproto, facet, 0);
2273 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2274 struct ofexpired expired;
2275 expired.flow = facet->flow;
2276 expired.packet_count = facet->packet_count;
2277 expired.byte_count = facet->byte_count;
2278 expired.used = facet->used;
2279 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2282 facet->rule->packet_count += facet->packet_count;
2283 facet->rule->byte_count += facet->byte_count;
2285 /* Reset counters to prevent double counting if 'facet' ever gets
2287 facet->packet_count = 0;
2288 facet->byte_count = 0;
2289 facet->accounted_bytes = 0;
2291 netflow_flow_clear(&facet->nf_flow);
2294 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2295 * Returns it if found, otherwise a null pointer.
2297 * The returned facet might need revalidation; use facet_lookup_valid()
2298 * instead if that is important. */
2299 static struct facet *
2300 facet_find(struct ofproto *ofproto, const struct flow *flow)
2302 struct facet *facet;
2304 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2306 if (flow_equal(flow, &facet->flow)) {
2314 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2315 * Returns it if found, otherwise a null pointer.
2317 * The returned facet is guaranteed to be valid. */
2318 static struct facet *
2319 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2321 struct facet *facet = facet_find(ofproto, flow);
2323 /* The facet we found might not be valid, since we could be in need of
2324 * revalidation. If it is not valid, don't return it. */
2326 && ofproto->need_revalidate
2327 && !facet_revalidate(ofproto, facet)) {
2328 COVERAGE_INC(ofproto_invalidated);
2335 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2337 * - If the rule found is different from 'facet''s current rule, moves
2338 * 'facet' to the new rule and recompiles its actions.
2340 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2341 * where it is and recompiles its actions anyway.
2343 * - If there is none, destroys 'facet'.
2345 * Returns true if 'facet' still exists, false if it has been destroyed. */
2347 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2349 struct rule *new_rule;
2350 struct odp_actions a;
2352 uint16_t new_nf_output_iface;
2353 bool actions_changed;
2355 COVERAGE_INC(facet_revalidate);
2357 /* Determine the new rule. */
2358 new_rule = rule_lookup(ofproto, &facet->flow);
2360 /* No new rule, so delete the facet. */
2361 facet_remove(ofproto, facet);
2365 /* Calculate new ODP actions.
2367 * We are very cautious about actually modifying 'facet' state at this
2368 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2369 * so, we need to have the old state around to properly compose it. */
2370 xlate_actions(new_rule->actions, new_rule->n_actions, &facet->flow,
2371 ofproto, NULL, &a, &facet->tags, &facet->may_install,
2372 &new_nf_output_iface);
2373 actions_len = a.n_actions * sizeof *a.actions;
2374 actions_changed = (facet->n_actions != a.n_actions
2375 || memcmp(facet->actions, a.actions, actions_len));
2377 /* If the ODP actions changed or the installability changed, then we need
2378 * to talk to the datapath. */
2379 if (actions_changed || facet->may_install != facet->installed) {
2380 if (facet->may_install) {
2381 struct odp_flow_put put;
2383 memset(&put.flow.stats, 0, sizeof put.flow.stats);
2384 odp_flow_key_from_flow(&put.flow.key, &facet->flow);
2385 put.flow.actions = a.actions;
2386 put.flow.n_actions = a.n_actions;
2388 put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS;
2389 dpif_flow_put(ofproto->dpif, &put);
2391 facet_update_stats(ofproto, facet, &put.flow.stats);
2393 facet_uninstall(ofproto, facet);
2396 /* The datapath flow is gone or has zeroed stats, so push stats out of
2397 * 'facet' into 'rule'. */
2398 facet_flush_stats(ofproto, facet);
2401 /* Update 'facet' now that we've taken care of all the old state. */
2402 facet->nf_flow.output_iface = new_nf_output_iface;
2403 if (actions_changed) {
2404 free(facet->actions);
2405 facet->n_actions = a.n_actions;
2406 facet->actions = xmemdup(a.actions, actions_len);
2408 if (facet->rule != new_rule) {
2409 COVERAGE_INC(facet_changed_rule);
2410 list_remove(&facet->list_node);
2411 list_push_back(&new_rule->facets, &facet->list_node);
2412 facet->rule = new_rule;
2413 facet->used = new_rule->created;
2420 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2421 struct rconn_packet_counter *counter)
2423 update_openflow_length(msg);
2424 if (rconn_send(ofconn->rconn, msg, counter)) {
2430 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2433 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2435 COVERAGE_INC(ofproto_error);
2436 queue_tx(buf, ofconn, ofconn->reply_counter);
2441 hton_ofp_phy_port(struct ofp_phy_port *opp)
2443 opp->port_no = htons(opp->port_no);
2444 opp->config = htonl(opp->config);
2445 opp->state = htonl(opp->state);
2446 opp->curr = htonl(opp->curr);
2447 opp->advertised = htonl(opp->advertised);
2448 opp->supported = htonl(opp->supported);
2449 opp->peer = htonl(opp->peer);
2453 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2455 struct ofp_header *rq = oh;
2456 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2461 handle_features_request(struct ofconn *ofconn, struct ofp_header *oh)
2463 struct ofp_switch_features *osf;
2465 struct ofport *port;
2467 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2468 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2469 osf->n_buffers = htonl(pktbuf_capacity());
2471 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2472 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2473 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2474 (1u << OFPAT_SET_VLAN_VID) |
2475 (1u << OFPAT_SET_VLAN_PCP) |
2476 (1u << OFPAT_STRIP_VLAN) |
2477 (1u << OFPAT_SET_DL_SRC) |
2478 (1u << OFPAT_SET_DL_DST) |
2479 (1u << OFPAT_SET_NW_SRC) |
2480 (1u << OFPAT_SET_NW_DST) |
2481 (1u << OFPAT_SET_NW_TOS) |
2482 (1u << OFPAT_SET_TP_SRC) |
2483 (1u << OFPAT_SET_TP_DST) |
2484 (1u << OFPAT_ENQUEUE));
2486 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2487 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2490 queue_tx(buf, ofconn, ofconn->reply_counter);
2495 handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh)
2498 struct ofp_switch_config *osc;
2502 /* Figure out flags. */
2503 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2504 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2507 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2508 osc->flags = htons(flags);
2509 osc->miss_send_len = htons(ofconn->miss_send_len);
2510 queue_tx(buf, ofconn, ofconn->reply_counter);
2516 handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc)
2521 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2525 flags = ntohs(osc->flags);
2527 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2528 switch (flags & OFPC_FRAG_MASK) {
2529 case OFPC_FRAG_NORMAL:
2530 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2532 case OFPC_FRAG_DROP:
2533 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2536 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2542 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2548 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2550 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2551 a->controller.arg = max_len;
2554 struct action_xlate_ctx {
2556 struct flow flow; /* Flow to which these actions correspond. */
2557 int recurse; /* Recursion level, via xlate_table_action. */
2558 struct ofproto *ofproto;
2559 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2560 * null pointer if we are revalidating
2561 * without a packet to refer to. */
2564 struct odp_actions *out; /* Datapath actions. */
2565 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
2566 bool may_set_up_flow; /* True ordinarily; false if the actions must
2567 * be reassessed for every packet. */
2568 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2571 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2572 * flow translation. */
2573 #define MAX_RESUBMIT_RECURSION 8
2575 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2576 struct action_xlate_ctx *ctx);
2579 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2581 const struct ofport *ofport = get_port(ctx->ofproto, port);
2584 if (ofport->opp.config & OFPPC_NO_FWD) {
2585 /* Forwarding disabled on port. */
2590 * We don't have an ofport record for this port, but it doesn't hurt to
2591 * allow forwarding to it anyhow. Maybe such a port will appear later
2592 * and we're pre-populating the flow table.
2596 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2597 ctx->nf_output_iface = port;
2600 static struct rule *
2601 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2603 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2607 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2609 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2610 uint16_t old_in_port;
2613 /* Look up a flow with 'in_port' as the input port. Then restore the
2614 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2615 * have surprising behavior). */
2616 old_in_port = ctx->flow.in_port;
2617 ctx->flow.in_port = in_port;
2618 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2619 ctx->flow.in_port = old_in_port;
2623 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2627 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2629 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2630 MAX_RESUBMIT_RECURSION);
2635 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2636 uint16_t *nf_output_iface, struct odp_actions *actions)
2638 struct ofport *ofport;
2640 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2641 uint16_t odp_port = ofport->odp_port;
2642 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2643 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2646 *nf_output_iface = NF_OUT_FLOOD;
2650 xlate_output_action__(struct action_xlate_ctx *ctx,
2651 uint16_t port, uint16_t max_len)
2654 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2656 ctx->nf_output_iface = NF_OUT_DROP;
2660 add_output_action(ctx, ctx->flow.in_port);
2663 xlate_table_action(ctx, ctx->flow.in_port);
2666 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2667 ctx->out, &ctx->tags,
2668 &ctx->nf_output_iface,
2669 ctx->ofproto->aux)) {
2670 COVERAGE_INC(ofproto_uninstallable);
2671 ctx->may_set_up_flow = false;
2675 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2676 &ctx->nf_output_iface, ctx->out);
2679 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2680 &ctx->nf_output_iface, ctx->out);
2682 case OFPP_CONTROLLER:
2683 add_controller_action(ctx->out, max_len);
2686 add_output_action(ctx, ODPP_LOCAL);
2689 odp_port = ofp_port_to_odp_port(port);
2690 if (odp_port != ctx->flow.in_port) {
2691 add_output_action(ctx, odp_port);
2696 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2697 ctx->nf_output_iface = NF_OUT_FLOOD;
2698 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2699 ctx->nf_output_iface = prev_nf_output_iface;
2700 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2701 ctx->nf_output_iface != NF_OUT_FLOOD) {
2702 ctx->nf_output_iface = NF_OUT_MULTI;
2707 xlate_output_action(struct action_xlate_ctx *ctx,
2708 const struct ofp_action_output *oao)
2710 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2713 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2714 * optimization, because we're going to add another action that sets the
2715 * priority immediately after, or because there are no actions following the
2718 remove_pop_action(struct action_xlate_ctx *ctx)
2720 size_t n = ctx->out->n_actions;
2721 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2722 ctx->out->n_actions--;
2727 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2728 const struct ofp_action_enqueue *oae)
2730 uint16_t ofp_port, odp_port;
2734 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2737 /* Fall back to ordinary output action. */
2738 xlate_output_action__(ctx, ntohs(oae->port), 0);
2742 /* Figure out ODP output port. */
2743 ofp_port = ntohs(oae->port);
2744 if (ofp_port != OFPP_IN_PORT) {
2745 odp_port = ofp_port_to_odp_port(ofp_port);
2747 odp_port = ctx->flow.in_port;
2750 /* Add ODP actions. */
2751 remove_pop_action(ctx);
2752 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2754 add_output_action(ctx, odp_port);
2755 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2757 /* Update NetFlow output port. */
2758 if (ctx->nf_output_iface == NF_OUT_DROP) {
2759 ctx->nf_output_iface = odp_port;
2760 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2761 ctx->nf_output_iface = NF_OUT_MULTI;
2766 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2767 const struct nx_action_set_queue *nasq)
2772 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2775 /* Couldn't translate queue to a priority, so ignore. A warning
2776 * has already been logged. */
2780 remove_pop_action(ctx);
2781 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2786 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2788 ovs_be16 tci = ctx->flow.vlan_tci;
2789 if (!(tci & htons(VLAN_CFI))) {
2790 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2792 union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2793 oa->dl_tci.tci = tci & ~htons(VLAN_CFI);
2798 xlate_reg_move_action(struct action_xlate_ctx *ctx,
2799 const struct nx_action_reg_move *narm)
2801 ovs_be16 old_tci = ctx->flow.vlan_tci;
2803 nxm_execute_reg_move(narm, &ctx->flow);
2805 if (ctx->flow.vlan_tci != old_tci) {
2806 xlate_set_dl_tci(ctx);
2811 xlate_nicira_action(struct action_xlate_ctx *ctx,
2812 const struct nx_action_header *nah)
2814 const struct nx_action_resubmit *nar;
2815 const struct nx_action_set_tunnel *nast;
2816 const struct nx_action_set_queue *nasq;
2817 union odp_action *oa;
2818 int subtype = ntohs(nah->subtype);
2820 assert(nah->vendor == htonl(NX_VENDOR_ID));
2822 case NXAST_RESUBMIT:
2823 nar = (const struct nx_action_resubmit *) nah;
2824 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2827 case NXAST_SET_TUNNEL:
2828 nast = (const struct nx_action_set_tunnel *) nah;
2829 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2830 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2833 case NXAST_DROP_SPOOFED_ARP:
2834 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2835 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2839 case NXAST_SET_QUEUE:
2840 nasq = (const struct nx_action_set_queue *) nah;
2841 xlate_set_queue_action(ctx, nasq);
2844 case NXAST_POP_QUEUE:
2845 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2848 case NXAST_REG_MOVE:
2849 xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah);
2852 case NXAST_REG_LOAD:
2853 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2857 /* Nothing to do. */
2860 /* If you add a new action here that modifies flow data, don't forget to
2861 * update the flow key in ctx->flow at the same time. */
2864 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2870 do_xlate_actions(const union ofp_action *in, size_t n_in,
2871 struct action_xlate_ctx *ctx)
2873 struct actions_iterator iter;
2874 const union ofp_action *ia;
2875 const struct ofport *port;
2877 port = get_port(ctx->ofproto, ctx->flow.in_port);
2878 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2879 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2880 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2881 /* Drop this flow. */
2885 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2886 uint16_t type = ntohs(ia->type);
2887 union odp_action *oa;
2891 xlate_output_action(ctx, &ia->output);
2894 case OFPAT_SET_VLAN_VID:
2895 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2896 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2897 xlate_set_dl_tci(ctx);
2900 case OFPAT_SET_VLAN_PCP:
2901 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2902 ctx->flow.vlan_tci |= htons(
2903 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2904 xlate_set_dl_tci(ctx);
2907 case OFPAT_STRIP_VLAN:
2908 ctx->flow.vlan_tci = htons(0);
2909 xlate_set_dl_tci(ctx);
2912 case OFPAT_SET_DL_SRC:
2913 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2914 memcpy(oa->dl_addr.dl_addr,
2915 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2916 memcpy(ctx->flow.dl_src,
2917 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2920 case OFPAT_SET_DL_DST:
2921 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2922 memcpy(oa->dl_addr.dl_addr,
2923 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2924 memcpy(ctx->flow.dl_dst,
2925 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2928 case OFPAT_SET_NW_SRC:
2929 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2930 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2933 case OFPAT_SET_NW_DST:
2934 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2935 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2938 case OFPAT_SET_NW_TOS:
2939 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2940 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2943 case OFPAT_SET_TP_SRC:
2944 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2945 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2948 case OFPAT_SET_TP_DST:
2949 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2950 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2954 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2958 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2962 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2969 xlate_actions(const union ofp_action *in, size_t n_in,
2970 const struct flow *flow, struct ofproto *ofproto,
2971 const struct ofpbuf *packet,
2972 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2973 uint16_t *nf_output_iface)
2975 struct action_xlate_ctx ctx;
2977 COVERAGE_INC(ofproto_ofp2odp);
2978 odp_actions_init(out);
2981 ctx.ofproto = ofproto;
2982 ctx.packet = packet;
2985 ctx.may_set_up_flow = true;
2986 ctx.nf_output_iface = NF_OUT_DROP;
2987 do_xlate_actions(in, n_in, &ctx);
2988 remove_pop_action(&ctx);
2990 /* Check with in-band control to see if we're allowed to set up this
2992 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2993 ctx.may_set_up_flow = false;
2999 if (may_set_up_flow) {
3000 *may_set_up_flow = ctx.may_set_up_flow;
3002 if (nf_output_iface) {
3003 *nf_output_iface = ctx.nf_output_iface;
3005 if (odp_actions_overflow(out)) {
3006 COVERAGE_INC(odp_overflow);
3007 odp_actions_init(out);
3008 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
3013 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3014 * error message code (composed with ofp_mkerr()) for the caller to propagate
3015 * upward. Otherwise, returns 0.
3017 * The log message mentions 'msg_type'. */
3019 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3021 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3022 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3023 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3026 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3033 handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh)
3035 struct ofproto *p = ofconn->ofproto;
3036 struct ofp_packet_out *opo;
3037 struct ofpbuf payload, *buffer;
3038 union ofp_action *ofp_actions;
3039 struct odp_actions odp_actions;
3040 struct ofpbuf request;
3042 size_t n_ofp_actions;
3046 COVERAGE_INC(ofproto_packet_out);
3048 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3053 /* Get ofp_packet_out. */
3055 request.size = ntohs(oh->length);
3056 opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions));
3058 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3062 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3063 &ofp_actions, &n_ofp_actions);
3069 if (opo->buffer_id != htonl(UINT32_MAX)) {
3070 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3072 if (error || !buffer) {
3081 /* Extract flow, check actions. */
3082 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3084 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3090 error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload,
3091 &odp_actions, NULL, NULL, NULL);
3093 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions,
3098 ofpbuf_delete(buffer);
3103 update_port_config(struct ofproto *p, struct ofport *port,
3104 uint32_t config, uint32_t mask)
3106 mask &= config ^ port->opp.config;
3107 if (mask & OFPPC_PORT_DOWN) {
3108 if (config & OFPPC_PORT_DOWN) {
3109 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3111 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3114 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3115 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3116 if (mask & REVALIDATE_BITS) {
3117 COVERAGE_INC(ofproto_costly_flags);
3118 port->opp.config ^= mask & REVALIDATE_BITS;
3119 p->need_revalidate = true;
3121 #undef REVALIDATE_BITS
3122 if (mask & OFPPC_NO_PACKET_IN) {
3123 port->opp.config ^= OFPPC_NO_PACKET_IN;
3128 handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh)
3130 struct ofproto *p = ofconn->ofproto;
3131 const struct ofp_port_mod *opm;
3132 struct ofport *port;
3135 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3139 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
3143 opm = (struct ofp_port_mod *) oh;
3145 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3147 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3148 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3149 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3151 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3152 if (opm->advertise) {
3153 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3159 static struct ofpbuf *
3160 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3162 struct ofp_stats_reply *osr;
3165 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3166 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3168 osr->flags = htons(0);
3172 static struct ofpbuf *
3173 start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3175 return make_ofp_stats_reply(request->header.xid, request->type, body_len);
3179 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3180 struct ofpbuf **msgp)
3182 struct ofpbuf *msg = *msgp;
3183 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3184 if (nbytes + msg->size > UINT16_MAX) {
3185 struct ofp_stats_reply *reply = msg->data;
3186 reply->flags = htons(OFPSF_REPLY_MORE);
3187 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3188 queue_tx(msg, ofconn, ofconn->reply_counter);
3190 return ofpbuf_put_uninit(*msgp, nbytes);
3193 static struct ofpbuf *
3194 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3196 struct nicira_stats_msg *nsm;
3199 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3200 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3201 nsm->type = htons(OFPST_VENDOR);
3202 nsm->flags = htons(0);
3203 nsm->vendor = htonl(NX_VENDOR_ID);
3204 nsm->subtype = htonl(subtype);
3208 static struct ofpbuf *
3209 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3211 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3215 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3216 struct ofpbuf **msgp)
3218 struct ofpbuf *msg = *msgp;
3219 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3220 if (nbytes + msg->size > UINT16_MAX) {
3221 struct nicira_stats_msg *reply = msg->data;
3222 reply->flags = htons(OFPSF_REPLY_MORE);
3223 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3224 queue_tx(msg, ofconn, ofconn->reply_counter);
3226 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3230 handle_desc_stats_request(struct ofconn *ofconn,
3231 struct ofp_stats_request *request)
3233 struct ofproto *p = ofconn->ofproto;
3234 struct ofp_desc_stats *ods;
3237 msg = start_ofp_stats_reply(request, sizeof *ods);
3238 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3239 memset(ods, 0, sizeof *ods);
3240 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3241 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3242 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3243 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3244 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3245 queue_tx(msg, ofconn, ofconn->reply_counter);
3251 handle_table_stats_request(struct ofconn *ofconn,
3252 struct ofp_stats_request *request)
3254 struct ofproto *p = ofconn->ofproto;
3255 struct ofp_table_stats *ots;
3258 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3260 /* Classifier table. */
3261 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3262 memset(ots, 0, sizeof *ots);
3263 strcpy(ots->name, "classifier");
3264 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3265 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3266 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3267 ots->active_count = htonl(classifier_count(&p->cls));
3268 ots->lookup_count = htonll(0); /* XXX */
3269 ots->matched_count = htonll(0); /* XXX */
3271 queue_tx(msg, ofconn, ofconn->reply_counter);
3276 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3277 struct ofpbuf **msgp)
3279 struct netdev_stats stats;
3280 struct ofp_port_stats *ops;
3282 /* Intentionally ignore return value, since errors will set
3283 * 'stats' to all-1s, which is correct for OpenFlow, and
3284 * netdev_get_stats() will log errors. */
3285 netdev_get_stats(port->netdev, &stats);
3287 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3288 ops->port_no = htons(port->opp.port_no);
3289 memset(ops->pad, 0, sizeof ops->pad);
3290 ops->rx_packets = htonll(stats.rx_packets);
3291 ops->tx_packets = htonll(stats.tx_packets);
3292 ops->rx_bytes = htonll(stats.rx_bytes);
3293 ops->tx_bytes = htonll(stats.tx_bytes);
3294 ops->rx_dropped = htonll(stats.rx_dropped);
3295 ops->tx_dropped = htonll(stats.tx_dropped);
3296 ops->rx_errors = htonll(stats.rx_errors);
3297 ops->tx_errors = htonll(stats.tx_errors);
3298 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3299 ops->rx_over_err = htonll(stats.rx_over_errors);
3300 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3301 ops->collisions = htonll(stats.collisions);
3305 handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr,
3308 struct ofproto *p = ofconn->ofproto;
3309 struct ofp_port_stats_request *psr;
3310 struct ofp_port_stats *ops;
3312 struct ofport *port;
3314 if (arg_size != sizeof *psr) {
3315 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3317 psr = (struct ofp_port_stats_request *) osr->body;
3319 msg = start_ofp_stats_reply(osr, sizeof *ops * 16);
3320 if (psr->port_no != htons(OFPP_NONE)) {
3321 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3323 append_port_stat(port, ofconn, &msg);
3326 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3327 append_port_stat(port, ofconn, &msg);
3331 queue_tx(msg, ofconn, ofconn->reply_counter);
3335 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3336 * '*packet_countp' and '*byte_countp'. The returned statistics include
3337 * statistics for all of 'rule''s facets. */
3339 query_stats(struct ofproto *p, struct rule *rule,
3340 uint64_t *packet_countp, uint64_t *byte_countp)
3342 uint64_t packet_count, byte_count;
3343 struct facet *facet;
3344 struct odp_flow *odp_flows;
3347 /* Start from historical data for 'rule' itself that are no longer tracked
3348 * by the datapath. This counts, for example, facets that have expired. */
3349 packet_count = rule->packet_count;
3350 byte_count = rule->byte_count;
3352 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3354 * Also, add any statistics that are not tracked by the datapath for each
3355 * facet. This includes, for example, statistics for packets that were
3356 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3358 odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows);
3360 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3361 struct odp_flow *odp_flow = &odp_flows[n_odp_flows++];
3362 odp_flow_key_from_flow(&odp_flow->key, &facet->flow);
3363 packet_count += facet->packet_count;
3364 byte_count += facet->byte_count;
3367 /* Fetch up-to-date statistics from the datapath and add them in. */
3368 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3371 for (i = 0; i < n_odp_flows; i++) {
3372 struct odp_flow *odp_flow = &odp_flows[i];
3373 packet_count += odp_flow->stats.n_packets;
3374 byte_count += odp_flow->stats.n_bytes;
3379 /* Return the stats to the caller. */
3380 *packet_countp = packet_count;
3381 *byte_countp = byte_count;
3385 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3387 long long int msecs = time_msec() - start;
3388 *sec = htonl(msecs / 1000);
3389 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3393 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3394 ovs_be16 out_port, struct ofpbuf **replyp)
3396 struct ofp_flow_stats *ofs;
3397 uint64_t packet_count, byte_count;
3398 size_t act_len, len;
3400 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3404 act_len = sizeof *rule->actions * rule->n_actions;
3405 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3407 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3409 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3410 ofs->length = htons(len);
3413 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match);
3414 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3415 ofs->cookie = rule->flow_cookie;
3416 ofs->priority = htons(rule->cr.priority);
3417 ofs->idle_timeout = htons(rule->idle_timeout);
3418 ofs->hard_timeout = htons(rule->hard_timeout);
3419 memset(ofs->pad2, 0, sizeof ofs->pad2);
3420 ofs->packet_count = htonll(packet_count);
3421 ofs->byte_count = htonll(byte_count);
3422 if (rule->n_actions > 0) {
3423 memcpy(ofs->actions, rule->actions, act_len);
3428 is_valid_table(uint8_t table_id)
3430 return table_id == 0 || table_id == 0xff;
3434 handle_flow_stats_request(struct ofconn *ofconn,
3435 const struct ofp_stats_request *osr, size_t arg_size)
3437 struct ofp_flow_stats_request *fsr;
3438 struct ofpbuf *reply;
3440 if (arg_size != sizeof *fsr) {
3441 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3443 fsr = (struct ofp_flow_stats_request *) osr->body;
3445 COVERAGE_INC(ofproto_flows_req);
3446 reply = start_ofp_stats_reply(osr, 1024);
3447 if (is_valid_table(fsr->table_id)) {
3448 struct cls_cursor cursor;
3449 struct cls_rule target;
3452 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3454 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3455 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3456 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3459 queue_tx(reply, ofconn, ofconn->reply_counter);
3465 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3466 ovs_be16 out_port, struct ofpbuf **replyp)
3468 struct nx_flow_stats *nfs;
3469 uint64_t packet_count, byte_count;
3470 size_t act_len, start_len;
3471 struct ofpbuf *reply;
3473 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3477 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3479 act_len = sizeof *rule->actions * rule->n_actions;
3481 start_len = (*replyp)->size;
3482 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3485 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3488 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3489 nfs->cookie = rule->flow_cookie;
3490 nfs->priority = htons(rule->cr.priority);
3491 nfs->idle_timeout = htons(rule->idle_timeout);
3492 nfs->hard_timeout = htons(rule->hard_timeout);
3493 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3494 memset(nfs->pad2, 0, sizeof nfs->pad2);
3495 nfs->packet_count = htonll(packet_count);
3496 nfs->byte_count = htonll(byte_count);
3497 if (rule->n_actions > 0) {
3498 ofpbuf_put(reply, rule->actions, act_len);
3500 nfs->length = htons(reply->size - start_len);
3504 handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b)
3506 struct nx_flow_stats_request *nfsr;
3507 struct cls_rule target;
3508 struct ofpbuf *reply;
3511 /* Dissect the message. */
3512 nfsr = ofpbuf_try_pull(b, sizeof *nfsr);
3514 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3516 error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target);
3521 COVERAGE_INC(ofproto_flows_req);
3522 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3523 if (is_valid_table(nfsr->table_id)) {
3524 struct cls_cursor cursor;
3527 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3528 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3529 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3532 queue_tx(reply, ofconn, ofconn->reply_counter);
3538 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3540 struct ofp_match match;
3541 uint64_t packet_count, byte_count;
3542 size_t act_len = sizeof *rule->actions * rule->n_actions;
3544 query_stats(ofproto, rule, &packet_count, &byte_count);
3545 ofputil_cls_rule_to_match(&rule->cr, NXFF_OPENFLOW10, &match);
3547 ds_put_format(results, "duration=%llds, ",
3548 (time_msec() - rule->created) / 1000);
3549 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3550 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3551 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3552 ofp_print_match(results, &match, true);
3554 ofp_print_actions(results, &rule->actions->header, act_len);
3556 ds_put_cstr(results, "drop");
3558 ds_put_cstr(results, "\n");
3561 /* Adds a pretty-printed description of all flows to 'results', including
3562 * those marked hidden by secchan (e.g., by in-band control). */
3564 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3566 struct cls_cursor cursor;
3569 cls_cursor_init(&cursor, &p->cls, NULL);
3570 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3571 flow_stats_ds(p, rule, results);
3576 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3577 ovs_be16 out_port, uint8_t table_id,
3578 struct ofp_aggregate_stats_reply *oasr)
3580 uint64_t total_packets = 0;
3581 uint64_t total_bytes = 0;
3584 COVERAGE_INC(ofproto_agg_request);
3586 if (is_valid_table(table_id)) {
3587 struct cls_cursor cursor;
3590 cls_cursor_init(&cursor, &ofproto->cls, target);
3591 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3592 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3593 uint64_t packet_count;
3594 uint64_t byte_count;
3596 query_stats(ofproto, rule, &packet_count, &byte_count);
3598 total_packets += packet_count;
3599 total_bytes += byte_count;
3605 oasr->flow_count = htonl(n_flows);
3606 oasr->packet_count = htonll(total_packets);
3607 oasr->byte_count = htonll(total_bytes);
3608 memset(oasr->pad, 0, sizeof oasr->pad);
3612 handle_aggregate_stats_request(struct ofconn *ofconn,
3613 const struct ofp_stats_request *osr,
3616 struct ofp_aggregate_stats_request *request;
3617 struct ofp_aggregate_stats_reply *reply;
3618 struct cls_rule target;
3621 if (arg_size != sizeof *request) {
3622 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3624 request = (struct ofp_aggregate_stats_request *) osr->body;
3626 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3629 msg = start_ofp_stats_reply(osr, sizeof *reply);
3630 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3631 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3632 request->table_id, reply);
3633 queue_tx(msg, ofconn, ofconn->reply_counter);
3638 handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b)
3640 struct nx_aggregate_stats_request *request;
3641 struct ofp_aggregate_stats_reply *reply;
3642 struct cls_rule target;
3646 /* Dissect the message. */
3647 request = ofpbuf_try_pull(b, sizeof *request);
3649 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3651 error = nx_pull_match(b, ntohs(request->match_len), 0, &target);
3657 COVERAGE_INC(ofproto_flows_req);
3658 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3659 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3660 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3661 request->table_id, reply);
3662 queue_tx(buf, ofconn, ofconn->reply_counter);
3667 struct queue_stats_cbdata {
3668 struct ofconn *ofconn;
3669 struct ofport *ofport;
3674 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3675 const struct netdev_queue_stats *stats)
3677 struct ofp_queue_stats *reply;
3679 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3680 reply->port_no = htons(cbdata->ofport->opp.port_no);
3681 memset(reply->pad, 0, sizeof reply->pad);
3682 reply->queue_id = htonl(queue_id);
3683 reply->tx_bytes = htonll(stats->tx_bytes);
3684 reply->tx_packets = htonll(stats->tx_packets);
3685 reply->tx_errors = htonll(stats->tx_errors);
3689 handle_queue_stats_dump_cb(uint32_t queue_id,
3690 struct netdev_queue_stats *stats,
3693 struct queue_stats_cbdata *cbdata = cbdata_;
3695 put_queue_stats(cbdata, queue_id, stats);
3699 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3700 struct queue_stats_cbdata *cbdata)
3702 cbdata->ofport = port;
3703 if (queue_id == OFPQ_ALL) {
3704 netdev_dump_queue_stats(port->netdev,
3705 handle_queue_stats_dump_cb, cbdata);
3707 struct netdev_queue_stats stats;
3709 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3710 put_queue_stats(cbdata, queue_id, &stats);
3716 handle_queue_stats_request(struct ofconn *ofconn,
3717 const struct ofp_stats_request *osr,
3720 struct ofproto *ofproto = ofconn->ofproto;
3721 struct ofp_queue_stats_request *qsr;
3722 struct queue_stats_cbdata cbdata;
3723 struct ofport *port;
3724 unsigned int port_no;
3727 if (arg_size != sizeof *qsr) {
3728 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3730 qsr = (struct ofp_queue_stats_request *) osr->body;
3732 COVERAGE_INC(ofproto_queue_req);
3734 cbdata.ofconn = ofconn;
3735 cbdata.msg = start_ofp_stats_reply(osr, 128);
3737 port_no = ntohs(qsr->port_no);
3738 queue_id = ntohl(qsr->queue_id);
3739 if (port_no == OFPP_ALL) {
3740 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3741 handle_queue_stats_for_port(port, queue_id, &cbdata);
3743 } else if (port_no < ofproto->max_ports) {
3744 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3746 handle_queue_stats_for_port(port, queue_id, &cbdata);
3749 ofpbuf_delete(cbdata.msg);
3750 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3752 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3758 handle_vendor_stats_request(struct ofconn *ofconn,
3759 struct ofp_stats_request *osr, size_t arg_size)
3761 struct nicira_stats_msg *nsm;
3766 VLOG_WARN_RL(&rl, "truncated vendor stats request body");
3767 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3770 memcpy(&vendor, osr->body, sizeof vendor);
3771 if (vendor != htonl(NX_VENDOR_ID)) {
3772 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3775 if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) {
3776 VLOG_WARN_RL(&rl, "truncated Nicira stats request");
3777 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3780 nsm = (struct nicira_stats_msg *) osr;
3782 b.size = ntohs(nsm->header.length);
3783 switch (ntohl(nsm->subtype)) {
3785 return handle_nxst_flow(ofconn, &b);
3787 case NXST_AGGREGATE:
3788 return handle_nxst_aggregate(ofconn, &b);
3791 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3796 handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh)
3798 struct ofp_stats_request *osr;
3802 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3807 osr = (struct ofp_stats_request *) oh;
3809 switch (ntohs(osr->type)) {
3811 return handle_desc_stats_request(ofconn, osr);
3814 return handle_flow_stats_request(ofconn, osr, arg_size);
3816 case OFPST_AGGREGATE:
3817 return handle_aggregate_stats_request(ofconn, osr, arg_size);
3820 return handle_table_stats_request(ofconn, osr);
3823 return handle_port_stats_request(ofconn, osr, arg_size);
3826 return handle_queue_stats_request(ofconn, osr, arg_size);
3829 return handle_vendor_stats_request(ofconn, osr, arg_size);
3832 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3836 static long long int
3837 msec_from_nsec(uint64_t sec, uint32_t nsec)
3839 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3843 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3844 const struct odp_flow_stats *stats)
3846 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3847 if (used > facet->used) {
3849 if (used > facet->rule->used) {
3850 facet->rule->used = used;
3852 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3856 /* Folds the statistics from 'stats' into the counters in 'facet'.
3858 * Because of the meaning of a facet's counters, it only makes sense to do this
3859 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3860 * packet that was sent by hand or if it represents statistics that have been
3861 * cleared out of the datapath. */
3863 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3864 const struct odp_flow_stats *stats)
3866 if (stats->n_packets) {
3867 facet_update_time(ofproto, facet, stats);
3868 facet->packet_count += stats->n_packets;
3869 facet->byte_count += stats->n_bytes;
3870 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3878 uint16_t idle_timeout;
3879 uint16_t hard_timeout;
3883 union ofp_action *actions;
3887 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3888 * in which no matching flow already exists in the flow table.
3890 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3891 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3892 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3894 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3897 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3899 struct ofproto *p = ofconn->ofproto;
3900 struct ofpbuf *packet;
3905 if (fm->flags & OFPFF_CHECK_OVERLAP
3906 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3907 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3911 if (fm->buffer_id != UINT32_MAX) {
3912 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3916 in_port = UINT16_MAX;
3919 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3920 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3921 fm->flags & OFPFF_SEND_FLOW_REM);
3922 rule_insert(p, rule);
3924 rule_execute(p, rule, in_port, packet);
3929 static struct rule *
3930 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3932 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3936 send_buffered_packet(struct ofconn *ofconn,
3937 struct rule *rule, uint32_t buffer_id)
3939 struct ofpbuf *packet;
3943 if (buffer_id == UINT32_MAX) {
3947 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3952 rule_execute(ofconn->ofproto, rule, in_port, packet);
3957 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3959 struct modify_flows_cbdata {
3960 struct ofproto *ofproto;
3961 const struct flow_mod *fm;
3965 static int modify_flow(struct ofproto *, const struct flow_mod *,
3968 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3969 * encoded by ofp_mkerr() on failure.
3971 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3974 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3976 struct ofproto *p = ofconn->ofproto;
3977 struct rule *match = NULL;
3978 struct cls_cursor cursor;
3981 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3982 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3983 if (!rule_is_hidden(rule)) {
3985 modify_flow(p, fm, rule);
3990 /* This credits the packet to whichever flow happened to match last.
3991 * That's weird. Maybe we should do a lookup for the flow that
3992 * actually matches the packet? Who knows. */
3993 send_buffered_packet(ofconn, match, fm->buffer_id);
3996 return add_flow(ofconn, fm);
4000 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4001 * code as encoded by ofp_mkerr() on failure.
4003 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4006 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4008 struct ofproto *p = ofconn->ofproto;
4009 struct rule *rule = find_flow_strict(p, fm);
4010 if (rule && !rule_is_hidden(rule)) {
4011 modify_flow(p, fm, rule);
4012 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4014 return add_flow(ofconn, fm);
4018 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4019 * been identified as a flow in 'p''s flow table to be modified, by changing
4020 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4021 * ofp_action[] structures). */
4023 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4025 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4027 rule->flow_cookie = fm->cookie;
4029 /* If the actions are the same, do nothing. */
4030 if (fm->n_actions == rule->n_actions
4032 || !memcmp(fm->actions, rule->actions, actions_len))) {
4036 /* Replace actions. */
4037 free(rule->actions);
4038 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4039 rule->n_actions = fm->n_actions;
4041 p->need_revalidate = true;
4046 /* OFPFC_DELETE implementation. */
4048 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4050 /* Implements OFPFC_DELETE. */
4052 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4054 struct rule *rule, *next_rule;
4055 struct cls_cursor cursor;
4057 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4058 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4059 delete_flow(p, rule, htons(fm->out_port));
4063 /* Implements OFPFC_DELETE_STRICT. */
4065 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4067 struct rule *rule = find_flow_strict(p, fm);
4069 delete_flow(p, rule, htons(fm->out_port));
4073 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4074 * been identified as a flow to delete from 'p''s flow table, by deleting the
4075 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4078 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4079 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4080 * specified 'out_port'. */
4082 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4084 if (rule_is_hidden(rule)) {
4088 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4092 rule_send_removed(p, rule, OFPRR_DELETE);
4093 rule_remove(p, rule);
4097 flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm)
4099 struct ofproto *p = ofconn->ofproto;
4102 error = reject_slave_controller(ofconn, "flow_mod");
4107 error = validate_actions(fm->actions, fm->n_actions,
4108 &fm->cr.flow, p->max_ports);
4113 /* We do not support the emergency flow cache. It will hopefully
4114 * get dropped from OpenFlow in the near future. */
4115 if (fm->flags & OFPFF_EMERG) {
4116 /* There isn't a good fit for an error code, so just state that the
4117 * flow table is full. */
4118 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4121 switch (fm->command) {
4123 return add_flow(ofconn, fm);
4126 return modify_flows_loose(ofconn, fm);
4128 case OFPFC_MODIFY_STRICT:
4129 return modify_flow_strict(ofconn, fm);
4132 delete_flows_loose(p, fm);
4135 case OFPFC_DELETE_STRICT:
4136 delete_flow_strict(p, fm);
4140 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4145 handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4147 struct ofp_match orig_match;
4148 struct ofp_flow_mod *ofm;
4154 b.size = ntohs(oh->length);
4156 /* Dissect the message. */
4157 ofm = ofpbuf_try_pull(&b, sizeof *ofm);
4159 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4161 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4166 /* Normalize ofm->match. If normalization actually changes anything, then
4167 * log the differences. */
4168 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
4169 orig_match = ofm->match;
4170 normalize_match(&ofm->match);
4171 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
4172 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4173 if (!VLOG_DROP_INFO(&normal_rl)) {
4174 char *old = ofp_match_to_literal_string(&orig_match);
4175 char *new = ofp_match_to_literal_string(&ofm->match);
4176 VLOG_INFO("%s: normalization changed ofp_match, details:",
4177 rconn_get_name(ofconn->rconn));
4178 VLOG_INFO(" pre: %s", old);
4179 VLOG_INFO("post: %s", new);
4185 /* Translate the message. */
4186 ofputil_cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
4187 ofconn->flow_format, ofm->cookie, &fm.cr);
4188 fm.cookie = ofm->cookie;
4189 fm.command = ntohs(ofm->command);
4190 fm.idle_timeout = ntohs(ofm->idle_timeout);
4191 fm.hard_timeout = ntohs(ofm->hard_timeout);
4192 fm.buffer_id = ntohl(ofm->buffer_id);
4193 fm.out_port = ntohs(ofm->out_port);
4194 fm.flags = ntohs(ofm->flags);
4196 /* Execute the command. */
4197 return flow_mod_core(ofconn, &fm);
4201 handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4203 struct nx_flow_mod *nfm;
4209 b.size = ntohs(oh->length);
4211 /* Dissect the message. */
4212 nfm = ofpbuf_try_pull(&b, sizeof *nfm);
4214 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4216 error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority),
4221 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4226 /* Translate the message. */
4227 fm.cookie = nfm->cookie;
4228 fm.command = ntohs(nfm->command);
4229 fm.idle_timeout = ntohs(nfm->idle_timeout);
4230 fm.hard_timeout = ntohs(nfm->hard_timeout);
4231 fm.buffer_id = ntohl(nfm->buffer_id);
4232 fm.out_port = ntohs(nfm->out_port);
4233 fm.flags = ntohs(nfm->flags);
4235 /* Execute the command. */
4236 return flow_mod_core(ofconn, &fm);
4240 handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg)
4244 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4249 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4254 handle_role_request(struct ofconn *ofconn, struct nicira_header *msg)
4256 struct nx_role_request *nrr;
4257 struct nx_role_request *reply;
4261 if (ntohs(msg->header.length) != sizeof *nrr) {
4262 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
4263 ntohs(msg->header.length), sizeof *nrr);
4264 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4266 nrr = (struct nx_role_request *) msg;
4268 if (ofconn->type != OFCONN_PRIMARY) {
4269 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4271 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4274 role = ntohl(nrr->role);
4275 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4276 && role != NX_ROLE_SLAVE) {
4277 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4279 /* There's no good error code for this. */
4280 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4283 if (role == NX_ROLE_MASTER) {
4284 struct ofconn *other;
4286 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4287 if (other->role == NX_ROLE_MASTER) {
4288 other->role = NX_ROLE_SLAVE;
4292 ofconn->role = role;
4294 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid,
4296 reply->role = htonl(role);
4297 queue_tx(buf, ofconn, ofconn->reply_counter);
4303 handle_nxt_set_flow_format(struct ofconn *ofconn,
4304 struct nxt_set_flow_format *msg)
4309 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4314 format = ntohl(msg->format);
4315 if (format == NXFF_OPENFLOW10
4316 || format == NXFF_TUN_ID_FROM_COOKIE
4317 || format == NXFF_NXM) {
4318 ofconn->flow_format = format;
4321 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4326 handle_vendor(struct ofconn *ofconn, void *msg)
4328 struct ofproto *p = ofconn->ofproto;
4329 struct ofp_vendor_header *ovh = msg;
4330 struct nicira_header *nh;
4332 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4333 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4334 "(expected at least %zu)",
4335 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4336 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4338 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4339 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4341 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4342 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4343 "(expected at least %zu)",
4344 ntohs(ovh->header.length), sizeof(struct nicira_header));
4345 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4349 switch (ntohl(nh->subtype)) {
4350 case NXT_STATUS_REQUEST:
4351 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4354 case NXT_TUN_ID_FROM_COOKIE:
4355 return handle_tun_id_from_cookie(ofconn, msg);
4357 case NXT_ROLE_REQUEST:
4358 return handle_role_request(ofconn, msg);
4360 case NXT_SET_FLOW_FORMAT:
4361 return handle_nxt_set_flow_format(ofconn, msg);
4364 return handle_nxt_flow_mod(ofconn, &ovh->header);
4367 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4371 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4373 struct ofp_header *ob;
4376 /* Currently, everything executes synchronously, so we can just
4377 * immediately send the barrier reply. */
4378 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4379 queue_tx(buf, ofconn, ofconn->reply_counter);
4384 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4386 struct ofp_header *oh = ofp_msg->data;
4389 COVERAGE_INC(ofproto_recv_openflow);
4391 case OFPT_ECHO_REQUEST:
4392 error = handle_echo_request(ofconn, oh);
4395 case OFPT_ECHO_REPLY:
4399 case OFPT_FEATURES_REQUEST:
4400 error = handle_features_request(ofconn, oh);
4403 case OFPT_GET_CONFIG_REQUEST:
4404 error = handle_get_config_request(ofconn, oh);
4407 case OFPT_SET_CONFIG:
4408 error = handle_set_config(ofconn, ofp_msg->data);
4411 case OFPT_PACKET_OUT:
4412 error = handle_packet_out(ofconn, ofp_msg->data);
4416 error = handle_port_mod(ofconn, oh);
4420 error = handle_ofpt_flow_mod(ofconn, ofp_msg->data);
4423 case OFPT_STATS_REQUEST:
4424 error = handle_stats_request(ofconn, oh);
4428 error = handle_vendor(ofconn, ofp_msg->data);
4431 case OFPT_BARRIER_REQUEST:
4432 error = handle_barrier_request(ofconn, oh);
4436 if (VLOG_IS_WARN_ENABLED()) {
4437 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4438 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4441 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4446 send_error_oh(ofconn, ofp_msg->data, error);
4451 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4453 struct odp_msg *msg = packet->data;
4454 struct ofpbuf payload;
4455 struct facet *facet;
4458 payload.data = msg + 1;
4459 payload.size = msg->length - sizeof *msg;
4460 flow_extract(&payload, msg->arg, msg->port, &flow);
4462 /* Check with in-band control to see if this packet should be sent
4463 * to the local port regardless of the flow table. */
4464 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4465 union odp_action action;
4467 memset(&action, 0, sizeof(action));
4468 action.output.type = ODPAT_OUTPUT;
4469 action.output.port = ODPP_LOCAL;
4470 dpif_execute(p->dpif, &action, 1, &payload);
4473 facet = facet_lookup_valid(p, &flow);
4475 struct rule *rule = rule_lookup(p, &flow);
4477 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4478 struct ofport *port = get_port(p, msg->port);
4480 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4481 COVERAGE_INC(ofproto_no_packet_in);
4482 /* XXX install 'drop' flow entry */
4483 ofpbuf_delete(packet);
4487 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4491 COVERAGE_INC(ofproto_packet_in);
4492 send_packet_in(p, packet);
4496 facet = facet_create(p, rule, &flow, packet);
4497 } else if (!facet->may_install) {
4498 /* The facet is not installable, that is, we need to process every
4499 * packet, so process the current packet's actions into 'facet'. */
4500 facet_make_actions(p, facet, packet);
4503 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4505 * Extra-special case for fail-open mode.
4507 * We are in fail-open mode and the packet matched the fail-open rule,
4508 * but we are connected to a controller too. We should send the packet
4509 * up to the controller in the hope that it will try to set up a flow
4510 * and thereby allow us to exit fail-open.
4512 * See the top-level comment in fail-open.c for more information.
4514 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4515 DPIF_RECV_MSG_PADDING));
4518 ofpbuf_pull(packet, sizeof *msg);
4519 facet_execute(p, facet, packet);
4520 facet_install(p, facet, false);
4524 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4526 struct odp_msg *msg = packet->data;
4528 switch (msg->type) {
4529 case _ODPL_ACTION_NR:
4530 COVERAGE_INC(ofproto_ctlr_action);
4531 send_packet_in(p, packet);
4534 case _ODPL_SFLOW_NR:
4536 ofproto_sflow_received(p->sflow, msg);
4538 ofpbuf_delete(packet);
4542 handle_odp_miss_msg(p, packet);
4546 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4552 /* Flow expiration. */
4554 static int ofproto_dp_max_idle(const struct ofproto *);
4555 static void ofproto_update_used(struct ofproto *);
4556 static void rule_expire(struct ofproto *, struct rule *);
4557 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4559 /* This function is called periodically by ofproto_run(). Its job is to
4560 * collect updates for the flows that have been installed into the datapath,
4561 * most importantly when they last were used, and then use that information to
4562 * expire flows that have not been used recently.
4564 * Returns the number of milliseconds after which it should be called again. */
4566 ofproto_expire(struct ofproto *ofproto)
4568 struct rule *rule, *next_rule;
4569 struct cls_cursor cursor;
4572 /* Update 'used' for each flow in the datapath. */
4573 ofproto_update_used(ofproto);
4575 /* Expire facets that have been idle too long. */
4576 dp_max_idle = ofproto_dp_max_idle(ofproto);
4577 ofproto_expire_facets(ofproto, dp_max_idle);
4579 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4580 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4581 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4582 rule_expire(ofproto, rule);
4585 /* Let the hook know that we're at a stable point: all outstanding data
4586 * in existing flows has been accounted to the account_cb. Thus, the
4587 * hook can now reasonably do operations that depend on having accurate
4588 * flow volume accounting (currently, that's just bond rebalancing). */
4589 if (ofproto->ofhooks->account_checkpoint_cb) {
4590 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4593 return MIN(dp_max_idle, 1000);
4596 /* Update 'used' member of installed facets. */
4598 ofproto_update_used(struct ofproto *p)
4600 struct odp_flow *flows;
4605 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4610 for (i = 0; i < n_flows; i++) {
4611 struct odp_flow *f = &flows[i];
4612 struct facet *facet;
4615 odp_flow_key_to_flow(&f->key, &flow);
4616 facet = facet_find(p, &flow);
4618 if (facet && facet->installed) {
4619 facet_update_time(p, facet, &f->stats);
4620 facet_account(p, facet, f->stats.n_bytes);
4622 /* There's a flow in the datapath that we know nothing about.
4624 COVERAGE_INC(ofproto_unexpected_rule);
4625 dpif_flow_del(p->dpif, f);
4632 /* Calculates and returns the number of milliseconds of idle time after which
4633 * facets should expire from the datapath and we should fold their statistics
4634 * into their parent rules in userspace. */
4636 ofproto_dp_max_idle(const struct ofproto *ofproto)
4639 * Idle time histogram.
4641 * Most of the time a switch has a relatively small number of facets. When
4642 * this is the case we might as well keep statistics for all of them in
4643 * userspace and to cache them in the kernel datapath for performance as
4646 * As the number of facets increases, the memory required to maintain
4647 * statistics about them in userspace and in the kernel becomes
4648 * significant. However, with a large number of facets it is likely that
4649 * only a few of them are "heavy hitters" that consume a large amount of
4650 * bandwidth. At this point, only heavy hitters are worth caching in the
4651 * kernel and maintaining in userspaces; other facets we can discard.
4653 * The technique used to compute the idle time is to build a histogram with
4654 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4655 * that is installed in the kernel gets dropped in the appropriate bucket.
4656 * After the histogram has been built, we compute the cutoff so that only
4657 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4658 * cached. At least the most-recently-used bucket of facets is kept, so
4659 * actually an arbitrary number of facets can be kept in any given
4660 * expiration run (though the next run will delete most of those unless
4661 * they receive additional data).
4663 * This requires a second pass through the facets, in addition to the pass
4664 * made by ofproto_update_used(), because the former function never looks
4665 * at uninstallable facets.
4667 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4668 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4669 int buckets[N_BUCKETS] = { 0 };
4670 struct facet *facet;
4675 total = hmap_count(&ofproto->facets);
4676 if (total <= 1000) {
4677 return N_BUCKETS * BUCKET_WIDTH;
4680 /* Build histogram. */
4682 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4683 long long int idle = now - facet->used;
4684 int bucket = (idle <= 0 ? 0
4685 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4686 : (unsigned int) idle / BUCKET_WIDTH);
4690 /* Find the first bucket whose flows should be expired. */
4691 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4692 if (buckets[bucket]) {
4695 subtotal += buckets[bucket++];
4696 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4701 if (VLOG_IS_DBG_ENABLED()) {
4705 ds_put_cstr(&s, "keep");
4706 for (i = 0; i < N_BUCKETS; i++) {
4708 ds_put_cstr(&s, ", drop");
4711 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4714 VLOG_INFO("%s: %s (msec:count)",
4715 dpif_name(ofproto->dpif), ds_cstr(&s));
4719 return bucket * BUCKET_WIDTH;
4723 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4725 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4726 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4727 struct ofexpired expired;
4728 struct odp_flow odp_flow;
4730 /* Get updated flow stats.
4732 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4733 * updated TCP flags and (2) the dpif_flow_list_all() in
4734 * ofproto_update_used() zeroed TCP flags. */
4735 memset(&odp_flow, 0, sizeof odp_flow);
4736 if (facet->installed) {
4737 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
4738 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4739 dpif_flow_get(ofproto->dpif, &odp_flow);
4741 if (odp_flow.stats.n_packets) {
4742 facet_update_time(ofproto, facet, &odp_flow.stats);
4743 netflow_flow_update_flags(&facet->nf_flow,
4744 odp_flow.stats.tcp_flags);
4748 expired.flow = facet->flow;
4749 expired.packet_count = facet->packet_count +
4750 odp_flow.stats.n_packets;
4751 expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes;
4752 expired.used = facet->used;
4754 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4759 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4761 long long int cutoff = time_msec() - dp_max_idle;
4762 struct facet *facet, *next_facet;
4764 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4765 facet_active_timeout(ofproto, facet);
4766 if (facet->used < cutoff) {
4767 facet_remove(ofproto, facet);
4772 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4773 * then delete it entirely. */
4775 rule_expire(struct ofproto *ofproto, struct rule *rule)
4777 struct facet *facet, *next_facet;
4781 /* Has 'rule' expired? */
4783 if (rule->hard_timeout
4784 && now > rule->created + rule->hard_timeout * 1000) {
4785 reason = OFPRR_HARD_TIMEOUT;
4786 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4787 && now >rule->used + rule->idle_timeout * 1000) {
4788 reason = OFPRR_IDLE_TIMEOUT;
4793 COVERAGE_INC(ofproto_expired);
4795 /* Update stats. (This is a no-op if the rule expired due to an idle
4796 * timeout, because that only happens when the rule has no facets left.) */
4797 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4798 facet_remove(ofproto, facet);
4801 /* Get rid of the rule. */
4802 if (!rule_is_hidden(rule)) {
4803 rule_send_removed(ofproto, rule, reason);
4805 rule_remove(ofproto, rule);
4808 static struct ofpbuf *
4809 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4812 struct ofp_flow_removed *ofr;
4815 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4816 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match);
4817 ofr->cookie = rule->flow_cookie;
4818 ofr->priority = htons(rule->cr.priority);
4819 ofr->reason = reason;
4820 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4821 ofr->idle_timeout = htons(rule->idle_timeout);
4822 ofr->packet_count = htonll(rule->packet_count);
4823 ofr->byte_count = htonll(rule->byte_count);
4828 static struct ofpbuf *
4829 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4831 struct nx_flow_removed *nfr;
4835 nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf);
4837 match_len = nx_put_match(buf, &rule->cr);
4839 nfr->cookie = rule->flow_cookie;
4840 nfr->priority = htons(rule->cr.priority);
4841 nfr->reason = reason;
4842 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4843 nfr->idle_timeout = htons(rule->idle_timeout);
4844 nfr->match_len = htons(match_len);
4845 nfr->packet_count = htonll(rule->packet_count);
4846 nfr->byte_count = htonll(rule->byte_count);
4852 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4854 struct ofconn *ofconn;
4856 if (!rule->send_flow_removed) {
4860 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4863 if (!rconn_is_connected(ofconn->rconn)
4864 || !ofconn_receives_async_msgs(ofconn)) {
4868 msg = (ofconn->flow_format == NXFF_NXM
4869 ? compose_nx_flow_removed(rule, reason)
4870 : compose_ofp_flow_removed(ofconn, rule, reason));
4872 /* Account flow expirations under ofconn->reply_counter, the counter
4873 * for replies to OpenFlow requests. That works because preventing
4874 * OpenFlow requests from being processed also prevents new flows from
4875 * being added (and expiring). (It also prevents processing OpenFlow
4876 * requests that would not add new flows, so it is imperfect.) */
4877 queue_tx(msg, ofconn, ofconn->reply_counter);
4881 /* pinsched callback for sending 'packet' on 'ofconn'. */
4883 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4885 struct ofconn *ofconn = ofconn_;
4887 rconn_send_with_limit(ofconn->rconn, packet,
4888 ofconn->packet_in_counter, 100);
4891 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4892 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4893 * packet scheduler for sending.
4895 * 'max_len' specifies the maximum number of bytes of the packet to send on
4896 * 'ofconn' (INT_MAX specifies no limit).
4898 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4899 * ownership is transferred to this function. */
4901 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4904 struct ofproto *ofproto = ofconn->ofproto;
4905 struct ofp_packet_in *opi = packet->data;
4906 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4907 int send_len, trim_size;
4911 if (opi->reason == OFPR_ACTION) {
4912 buffer_id = UINT32_MAX;
4913 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4914 buffer_id = pktbuf_get_null();
4915 } else if (!ofconn->pktbuf) {
4916 buffer_id = UINT32_MAX;
4918 struct ofpbuf payload;
4919 payload.data = opi->data;
4920 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4921 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4924 /* Figure out how much of the packet to send. */
4925 send_len = ntohs(opi->total_len);
4926 if (buffer_id != UINT32_MAX) {
4927 send_len = MIN(send_len, ofconn->miss_send_len);
4929 send_len = MIN(send_len, max_len);
4931 /* Adjust packet length and clone if necessary. */
4932 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4934 packet = ofpbuf_clone_data(packet->data, trim_size);
4937 packet->size = trim_size;
4940 /* Update packet headers. */
4941 opi->buffer_id = htonl(buffer_id);
4942 update_openflow_length(packet);
4944 /* Hand over to packet scheduler. It might immediately call into
4945 * do_send_packet_in() or it might buffer it for a while (until a later
4946 * call to pinsched_run()). */
4947 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4948 packet, do_send_packet_in, ofconn);
4951 /* Replace struct odp_msg header in 'packet' by equivalent struct
4952 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4953 * returned by dpif_recv()).
4955 * The conversion is not complete: the caller still needs to trim any unneeded
4956 * payload off the end of the buffer, set the length in the OpenFlow header,
4957 * and set buffer_id. Those require us to know the controller settings and so
4958 * must be done on a per-controller basis.
4960 * Returns the maximum number of bytes of the packet that should be sent to
4961 * the controller (INT_MAX if no limit). */
4963 do_convert_to_packet_in(struct ofpbuf *packet)
4965 struct odp_msg *msg = packet->data;
4966 struct ofp_packet_in *opi;
4972 /* Extract relevant header fields */
4973 if (msg->type == _ODPL_ACTION_NR) {
4974 reason = OFPR_ACTION;
4977 reason = OFPR_NO_MATCH;
4980 total_len = msg->length - sizeof *msg;
4981 in_port = odp_port_to_ofp_port(msg->port);
4983 /* Repurpose packet buffer by overwriting header. */
4984 ofpbuf_pull(packet, sizeof(struct odp_msg));
4985 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4986 opi->header.version = OFP_VERSION;
4987 opi->header.type = OFPT_PACKET_IN;
4988 opi->total_len = htons(total_len);
4989 opi->in_port = htons(in_port);
4990 opi->reason = reason;
4995 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4996 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4997 * as necessary according to their individual configurations.
4999 * 'packet' must have sufficient headroom to convert it into a struct
5000 * ofp_packet_in (e.g. as returned by dpif_recv()).
5002 * Takes ownership of 'packet'. */
5004 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
5006 struct ofconn *ofconn, *prev;
5009 max_len = do_convert_to_packet_in(packet);
5012 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5013 if (ofconn_receives_async_msgs(ofconn)) {
5015 schedule_packet_in(prev, packet, max_len, true);
5021 schedule_packet_in(prev, packet, max_len, false);
5023 ofpbuf_delete(packet);
5028 pick_datapath_id(const struct ofproto *ofproto)
5030 const struct ofport *port;
5032 port = get_port(ofproto, ODPP_LOCAL);
5034 uint8_t ea[ETH_ADDR_LEN];
5037 error = netdev_get_etheraddr(port->netdev, ea);
5039 return eth_addr_to_uint64(ea);
5041 VLOG_WARN("could not get MAC address for %s (%s)",
5042 netdev_get_name(port->netdev), strerror(error));
5044 return ofproto->fallback_dpid;
5048 pick_fallback_dpid(void)
5050 uint8_t ea[ETH_ADDR_LEN];
5051 eth_addr_nicira_random(ea);
5052 return eth_addr_to_uint64(ea);
5056 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5057 struct odp_actions *actions, tag_type *tags,
5058 uint16_t *nf_output_iface, void *ofproto_)
5060 struct ofproto *ofproto = ofproto_;
5063 /* Drop frames for reserved multicast addresses. */
5064 if (eth_addr_is_reserved(flow->dl_dst)) {
5068 /* Learn source MAC (but don't try to learn from revalidation). */
5069 if (packet != NULL) {
5070 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5072 GRAT_ARP_LOCK_NONE);
5074 /* The log messages here could actually be useful in debugging,
5075 * so keep the rate limit relatively high. */
5076 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5077 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5078 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5079 ofproto_revalidate(ofproto, rev_tag);
5083 /* Determine output port. */
5084 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5087 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5088 nf_output_iface, actions);
5089 } else if (out_port != flow->in_port) {
5090 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
5091 *nf_output_iface = out_port;
5099 static const struct ofhooks default_ofhooks = {
5100 default_normal_ofhook_cb,