2 * Copyright (c) 2009, 2010, 2011 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"
38 #include "multipath.h"
44 #include "ofp-print.h"
46 #include "ofproto-sflow.h"
48 #include "openflow/nicira-ext.h"
49 #include "openflow/openflow.h"
50 #include "openvswitch/datapath-protocol.h"
54 #include "poll-loop.h"
58 #include "stream-ssl.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(odp_overflow);
71 COVERAGE_DEFINE(ofproto_agg_request);
72 COVERAGE_DEFINE(ofproto_costly_flags);
73 COVERAGE_DEFINE(ofproto_ctlr_action);
74 COVERAGE_DEFINE(ofproto_del_rule);
75 COVERAGE_DEFINE(ofproto_error);
76 COVERAGE_DEFINE(ofproto_expiration);
77 COVERAGE_DEFINE(ofproto_expired);
78 COVERAGE_DEFINE(ofproto_flows_req);
79 COVERAGE_DEFINE(ofproto_flush);
80 COVERAGE_DEFINE(ofproto_invalidated);
81 COVERAGE_DEFINE(ofproto_no_packet_in);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 #include "sflow_api.h"
98 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
99 struct netdev *netdev;
100 struct ofp_phy_port opp; /* In host byte order. */
104 static void ofport_free(struct ofport *);
105 static void hton_ofp_phy_port(struct ofp_phy_port *);
107 struct action_xlate_ctx {
108 /* action_xlate_ctx_init() initializes these members. */
111 struct ofproto *ofproto;
113 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
114 * this flow when actions change header fields. */
117 /* The packet corresponding to 'flow', or a null pointer if we are
118 * revalidating without a packet to refer to. */
119 const struct ofpbuf *packet;
121 /* If nonnull, called just before executing a resubmit action.
123 * This is normally null so the client has to set it manually after
124 * calling action_xlate_ctx_init(). */
125 void (*resubmit_hook)(struct action_xlate_ctx *, const struct rule *);
127 /* xlate_actions() initializes and uses these members. The client might want
128 * to look at them after it returns. */
130 struct ofpbuf *odp_actions; /* Datapath actions. */
131 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
132 bool may_set_up_flow; /* True ordinarily; false if the actions must
133 * be reassessed for every packet. */
134 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
136 /* xlate_actions() initializes and uses these members, but the client has no
137 * reason to look at them. */
139 int recurse; /* Recursion level, via xlate_table_action. */
140 int last_pop_priority; /* Offset in 'odp_actions' just past most
141 * recently added ODPAT_SET_PRIORITY. */
144 static void action_xlate_ctx_init(struct action_xlate_ctx *,
145 struct ofproto *, const struct flow *,
146 const struct ofpbuf *);
147 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
148 const union ofp_action *in, size_t n_in);
150 /* An OpenFlow flow. */
152 long long int used; /* Time last used; time created if not used. */
153 long long int created; /* Creation time. */
157 * - Do include packets and bytes from facets that have been deleted or
158 * whose own statistics have been folded into the rule.
160 * - Do include packets and bytes sent "by hand" that were accounted to
161 * the rule without any facet being involved (this is a rare corner
162 * case in rule_execute()).
164 * - Do not include packet or bytes that can be obtained from any facet's
165 * packet_count or byte_count member or that can be obtained from the
166 * datapath by, e.g., dpif_flow_get() for any facet.
168 uint64_t packet_count; /* Number of packets received. */
169 uint64_t byte_count; /* Number of bytes received. */
171 ovs_be64 flow_cookie; /* Controller-issued identifier. */
173 struct cls_rule cr; /* In owning ofproto's classifier. */
174 uint16_t idle_timeout; /* In seconds from time of last use. */
175 uint16_t hard_timeout; /* In seconds from time of creation. */
176 bool send_flow_removed; /* Send a flow removed message? */
177 int n_actions; /* Number of elements in actions[]. */
178 union ofp_action *actions; /* OpenFlow actions. */
179 struct list facets; /* List of "struct facet"s. */
182 static struct rule *rule_from_cls_rule(const struct cls_rule *);
183 static bool rule_is_hidden(const struct rule *);
185 static struct rule *rule_create(const struct cls_rule *,
186 const union ofp_action *, size_t n_actions,
187 uint16_t idle_timeout, uint16_t hard_timeout,
188 ovs_be64 flow_cookie, bool send_flow_removed);
189 static void rule_destroy(struct ofproto *, struct rule *);
190 static void rule_free(struct rule *);
192 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
193 static void rule_insert(struct ofproto *, struct rule *);
194 static void rule_remove(struct ofproto *, struct rule *);
196 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
198 /* An exact-match instantiation of an OpenFlow flow. */
200 long long int used; /* Time last used; time created if not used. */
204 * - Do include packets and bytes sent "by hand", e.g. with
207 * - Do include packets and bytes that were obtained from the datapath
208 * when a flow was deleted (e.g. dpif_flow_del()) or when its
209 * statistics were reset (e.g. dpif_flow_put() with
210 * DPIF_FP_ZERO_STATS).
212 * - Do not include any packets or bytes that can currently be obtained
213 * from the datapath by, e.g., dpif_flow_get().
215 uint64_t packet_count; /* Number of packets received. */
216 uint64_t byte_count; /* Number of bytes received. */
218 /* Number of bytes passed to account_cb. This may include bytes that can
219 * currently obtained from the datapath (thus, it can be greater than
221 uint64_t accounted_bytes;
223 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
224 struct list list_node; /* In owning rule's 'facets' list. */
225 struct rule *rule; /* Owning rule. */
226 struct flow flow; /* Exact-match flow. */
227 bool installed; /* Installed in datapath? */
228 bool may_install; /* True ordinarily; false if actions must
229 * be reassessed for every packet. */
230 size_t actions_len; /* Number of bytes in actions[]. */
231 struct nlattr *actions; /* Datapath actions. */
232 tag_type tags; /* Tags (set only by hooks). */
233 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
236 static struct facet *facet_create(struct ofproto *, struct rule *,
238 const struct ofpbuf *packet);
239 static void facet_remove(struct ofproto *, struct facet *);
240 static void facet_free(struct facet *);
242 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
243 static bool facet_revalidate(struct ofproto *, struct facet *);
245 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
246 static void facet_uninstall(struct ofproto *, struct facet *);
247 static void facet_flush_stats(struct ofproto *, struct facet *);
249 static void facet_make_actions(struct ofproto *, struct facet *,
250 const struct ofpbuf *packet);
251 static void facet_update_stats(struct ofproto *, struct facet *,
252 const struct dpif_flow_stats *);
254 /* ofproto supports two kinds of OpenFlow connections:
256 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
257 * maintains persistent connections to these controllers and by default
258 * sends them asynchronous messages such as packet-ins.
260 * - "Service" connections, e.g. from ovs-ofctl. When these connections
261 * drop, it is the other side's responsibility to reconnect them if
262 * necessary. ofproto does not send them asynchronous messages by default.
264 * Currently, active (tcp, ssl, unix) connections are always "primary"
265 * connections and passive (ptcp, pssl, punix) connections are always "service"
266 * connections. There is no inherent reason for this, but it reflects the
270 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
271 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
274 /* A listener for incoming OpenFlow "service" connections. */
276 struct hmap_node node; /* In struct ofproto's "services" hmap. */
277 struct pvconn *pvconn; /* OpenFlow connection listener. */
279 /* These are not used by ofservice directly. They are settings for
280 * accepted "struct ofconn"s from the pvconn. */
281 int probe_interval; /* Max idle time before probing, in seconds. */
282 int rate_limit; /* Max packet-in rate in packets per second. */
283 int burst_limit; /* Limit on accumulating packet credits. */
286 static struct ofservice *ofservice_lookup(struct ofproto *,
288 static int ofservice_create(struct ofproto *,
289 const struct ofproto_controller *);
290 static void ofservice_reconfigure(struct ofservice *,
291 const struct ofproto_controller *);
292 static void ofservice_destroy(struct ofproto *, struct ofservice *);
294 /* An OpenFlow connection. */
296 struct ofproto *ofproto; /* The ofproto that owns this connection. */
297 struct list node; /* In struct ofproto's "all_conns" list. */
298 struct rconn *rconn; /* OpenFlow connection. */
299 enum ofconn_type type; /* Type. */
300 enum nx_flow_format flow_format; /* Currently selected flow format. */
302 /* OFPT_PACKET_IN related data. */
303 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
304 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
305 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
306 int miss_send_len; /* Bytes to send of buffered packets. */
308 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
309 * requests, and the maximum number before we stop reading OpenFlow
311 #define OFCONN_REPLY_MAX 100
312 struct rconn_packet_counter *reply_counter;
314 /* type == OFCONN_PRIMARY only. */
315 enum nx_role role; /* Role. */
316 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
317 struct discovery *discovery; /* Controller discovery object, if enabled. */
318 struct status_category *ss; /* Switch status category. */
319 enum ofproto_band band; /* In-band or out-of-band? */
322 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
323 * "schedulers" array. Their values are 0 and 1, and their meanings and values
324 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
325 * case anything ever changes, check their values here. */
326 #define N_SCHEDULERS 2
327 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
328 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
329 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
330 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
332 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
334 static void ofconn_destroy(struct ofconn *);
335 static void ofconn_run(struct ofconn *);
336 static void ofconn_wait(struct ofconn *);
337 static bool ofconn_receives_async_msgs(const struct ofconn *);
338 static char *ofconn_make_name(const struct ofproto *, const char *target);
339 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
341 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
342 struct rconn_packet_counter *counter);
344 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
345 const struct flow *, bool clone);
346 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
350 uint64_t datapath_id; /* Datapath ID. */
351 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
352 char *mfr_desc; /* Manufacturer. */
353 char *hw_desc; /* Hardware. */
354 char *sw_desc; /* Software version. */
355 char *serial_desc; /* Serial number. */
356 char *dp_desc; /* Datapath description. */
360 struct netdev_monitor *netdev_monitor;
361 struct hmap ports; /* Contains "struct ofport"s. */
362 struct shash port_by_name;
366 struct switch_status *switch_status;
367 struct fail_open *fail_open;
368 struct netflow *netflow;
369 struct ofproto_sflow *sflow;
371 /* In-band control. */
372 struct in_band *in_band;
373 long long int next_in_band_update;
374 struct sockaddr_in *extra_in_band_remotes;
375 size_t n_extra_remotes;
379 struct classifier cls;
380 long long int next_expiration;
384 bool need_revalidate;
385 struct tag_set revalidate_set;
387 /* OpenFlow connections. */
388 struct hmap controllers; /* Controller "struct ofconn"s. */
389 struct list all_conns; /* Contains "struct ofconn"s. */
390 enum ofproto_fail_mode fail_mode;
392 /* OpenFlow listeners. */
393 struct hmap services; /* Contains "struct ofservice"s. */
394 struct pvconn **snoops;
397 /* Hooks for ovs-vswitchd. */
398 const struct ofhooks *ofhooks;
401 /* Used by default ofhooks. */
402 struct mac_learning *ml;
405 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
406 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
408 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
410 static const struct ofhooks default_ofhooks;
412 static uint64_t pick_datapath_id(const struct ofproto *);
413 static uint64_t pick_fallback_dpid(void);
415 static int ofproto_expire(struct ofproto *);
417 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
419 static void handle_openflow(struct ofconn *, struct ofpbuf *);
421 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
422 static void update_port(struct ofproto *, const char *devname);
423 static int init_ports(struct ofproto *);
424 static void reinit_ports(struct ofproto *);
426 static void ofproto_unixctl_init(void);
429 ofproto_create(const char *datapath, const char *datapath_type,
430 const struct ofhooks *ofhooks, void *aux,
431 struct ofproto **ofprotop)
439 ofproto_unixctl_init();
441 /* Connect to datapath and start listening for messages. */
442 error = dpif_open(datapath, datapath_type, &dpif);
444 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
447 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
449 VLOG_ERR("failed to listen on datapath %s: %s",
450 datapath, strerror(error));
454 dpif_flow_flush(dpif);
455 dpif_recv_purge(dpif);
457 /* Initialize settings. */
458 p = xzalloc(sizeof *p);
459 p->fallback_dpid = pick_fallback_dpid();
460 p->datapath_id = p->fallback_dpid;
461 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
462 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
463 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
464 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
465 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
467 /* Initialize datapath. */
469 p->netdev_monitor = netdev_monitor_create();
470 hmap_init(&p->ports);
471 shash_init(&p->port_by_name);
472 p->max_ports = dpif_get_max_ports(dpif);
474 /* Initialize submodules. */
475 p->switch_status = switch_status_create(p);
480 /* Initialize in-band control. */
482 p->in_band_queue = -1;
484 /* Initialize flow table. */
485 classifier_init(&p->cls);
486 p->next_expiration = time_msec() + 1000;
488 /* Initialize facet table. */
489 hmap_init(&p->facets);
490 p->need_revalidate = false;
491 tag_set_init(&p->revalidate_set);
493 /* Initialize OpenFlow connections. */
494 list_init(&p->all_conns);
495 hmap_init(&p->controllers);
496 hmap_init(&p->services);
500 /* Initialize hooks. */
502 p->ofhooks = ofhooks;
506 p->ofhooks = &default_ofhooks;
508 p->ml = mac_learning_create();
511 /* Pick final datapath ID. */
512 p->datapath_id = pick_datapath_id(p);
513 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
515 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
522 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
524 uint64_t old_dpid = p->datapath_id;
525 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
526 if (p->datapath_id != old_dpid) {
527 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
529 /* Force all active connections to reconnect, since there is no way to
530 * notify a controller that the datapath ID has changed. */
531 ofproto_reconnect_controllers(p);
536 is_discovery_controller(const struct ofproto_controller *c)
538 return !strcmp(c->target, "discover");
542 is_in_band_controller(const struct ofproto_controller *c)
544 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
547 /* Creates a new controller in 'ofproto'. Some of the settings are initially
548 * drawn from 'c', but update_controller() needs to be called later to finish
549 * the new ofconn's configuration. */
551 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
553 struct discovery *discovery;
554 struct ofconn *ofconn;
556 if (is_discovery_controller(c)) {
557 int error = discovery_create(c->accept_re, c->update_resolv_conf,
558 ofproto->dpif, ofproto->switch_status,
567 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
568 ofconn->pktbuf = pktbuf_create();
569 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
571 ofconn->discovery = discovery;
573 char *name = ofconn_make_name(ofproto, c->target);
574 rconn_connect(ofconn->rconn, c->target, name);
577 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
578 hash_string(c->target, 0));
581 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
582 * target or turn discovery on or off (these are done by creating new ofconns
583 * and deleting old ones), but it can update the rest of an ofconn's
586 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
590 ofconn->band = (is_in_band_controller(c)
591 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
593 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
595 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
596 rconn_set_probe_interval(ofconn->rconn, probe_interval);
598 if (ofconn->discovery) {
599 discovery_set_update_resolv_conf(ofconn->discovery,
600 c->update_resolv_conf);
601 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
604 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
608 ofconn_get_target(const struct ofconn *ofconn)
610 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
613 static struct ofconn *
614 find_controller_by_target(struct ofproto *ofproto, const char *target)
616 struct ofconn *ofconn;
618 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
619 hash_string(target, 0), &ofproto->controllers) {
620 if (!strcmp(ofconn_get_target(ofconn), target)) {
628 update_in_band_remotes(struct ofproto *ofproto)
630 const struct ofconn *ofconn;
631 struct sockaddr_in *addrs;
632 size_t max_addrs, n_addrs;
636 /* Allocate enough memory for as many remotes as we could possibly have. */
637 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
638 addrs = xmalloc(max_addrs * sizeof *addrs);
641 /* Add all the remotes. */
643 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
644 struct sockaddr_in *sin = &addrs[n_addrs];
646 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
650 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
651 if (sin->sin_addr.s_addr) {
652 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
655 if (ofconn->discovery) {
659 for (i = 0; i < ofproto->n_extra_remotes; i++) {
660 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
663 /* Create or update or destroy in-band.
665 * Ordinarily we only enable in-band if there's at least one remote
666 * address, but discovery needs the in-band rules for DHCP to be installed
667 * even before we know any remote addresses. */
668 if (n_addrs || discovery) {
669 if (!ofproto->in_band) {
670 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
673 if (ofproto->in_band) {
674 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
676 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
677 ofproto->next_in_band_update = time_msec() + 1000;
679 in_band_destroy(ofproto->in_band);
680 ofproto->in_band = NULL;
688 update_fail_open(struct ofproto *p)
690 struct ofconn *ofconn;
692 if (!hmap_is_empty(&p->controllers)
693 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
694 struct rconn **rconns;
698 p->fail_open = fail_open_create(p, p->switch_status);
702 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
703 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
704 rconns[n++] = ofconn->rconn;
707 fail_open_set_controllers(p->fail_open, rconns, n);
708 /* p->fail_open takes ownership of 'rconns'. */
710 fail_open_destroy(p->fail_open);
716 ofproto_set_controllers(struct ofproto *p,
717 const struct ofproto_controller *controllers,
718 size_t n_controllers)
720 struct shash new_controllers;
721 struct ofconn *ofconn, *next_ofconn;
722 struct ofservice *ofservice, *next_ofservice;
726 /* Create newly configured controllers and services.
727 * Create a name to ofproto_controller mapping in 'new_controllers'. */
728 shash_init(&new_controllers);
729 for (i = 0; i < n_controllers; i++) {
730 const struct ofproto_controller *c = &controllers[i];
732 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
733 if (!find_controller_by_target(p, c->target)) {
734 add_controller(p, c);
736 } else if (!pvconn_verify_name(c->target)) {
737 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
741 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
742 dpif_name(p->dpif), c->target);
746 shash_add_once(&new_controllers, c->target, &controllers[i]);
749 /* Delete controllers that are no longer configured.
750 * Update configuration of all now-existing controllers. */
752 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
753 struct ofproto_controller *c;
755 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
757 ofconn_destroy(ofconn);
759 update_controller(ofconn, c);
766 /* Delete services that are no longer configured.
767 * Update configuration of all now-existing services. */
768 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
769 struct ofproto_controller *c;
771 c = shash_find_data(&new_controllers,
772 pvconn_get_name(ofservice->pvconn));
774 ofservice_destroy(p, ofservice);
776 ofservice_reconfigure(ofservice, c);
780 shash_destroy(&new_controllers);
782 update_in_band_remotes(p);
785 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
786 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
787 struct ofconn, hmap_node);
788 ofconn->ss = switch_status_register(p->switch_status, "remote",
789 rconn_status_cb, ofconn->rconn);
794 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
796 p->fail_mode = fail_mode;
800 /* Drops the connections between 'ofproto' and all of its controllers, forcing
801 * them to reconnect. */
803 ofproto_reconnect_controllers(struct ofproto *ofproto)
805 struct ofconn *ofconn;
807 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
808 rconn_reconnect(ofconn->rconn);
813 any_extras_changed(const struct ofproto *ofproto,
814 const struct sockaddr_in *extras, size_t n)
818 if (n != ofproto->n_extra_remotes) {
822 for (i = 0; i < n; i++) {
823 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
824 const struct sockaddr_in *new = &extras[i];
826 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
827 old->sin_port != new->sin_port) {
835 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
836 * in-band control should guarantee access, in the same way that in-band
837 * control guarantees access to OpenFlow controllers. */
839 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
840 const struct sockaddr_in *extras, size_t n)
842 if (!any_extras_changed(ofproto, extras, n)) {
846 free(ofproto->extra_in_band_remotes);
847 ofproto->n_extra_remotes = n;
848 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
850 update_in_band_remotes(ofproto);
853 /* Sets the OpenFlow queue used by flows set up by in-band control on
854 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
855 * flows will use the default queue. */
857 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
859 if (queue_id != ofproto->in_band_queue) {
860 ofproto->in_band_queue = queue_id;
861 update_in_band_remotes(ofproto);
866 ofproto_set_desc(struct ofproto *p,
867 const char *mfr_desc, const char *hw_desc,
868 const char *sw_desc, const char *serial_desc,
871 struct ofp_desc_stats *ods;
874 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
875 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
876 sizeof ods->mfr_desc);
879 p->mfr_desc = xstrdup(mfr_desc);
882 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
883 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
884 sizeof ods->hw_desc);
887 p->hw_desc = xstrdup(hw_desc);
890 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
891 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
892 sizeof ods->sw_desc);
895 p->sw_desc = xstrdup(sw_desc);
898 if (strlen(serial_desc) >= sizeof ods->serial_num) {
899 VLOG_WARN("truncating serial_desc, must be less than %zu "
901 sizeof ods->serial_num);
903 free(p->serial_desc);
904 p->serial_desc = xstrdup(serial_desc);
907 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
908 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
909 sizeof ods->dp_desc);
912 p->dp_desc = xstrdup(dp_desc);
917 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
918 const struct svec *svec)
920 struct pvconn **pvconns = *pvconnsp;
921 size_t n_pvconns = *n_pvconnsp;
925 for (i = 0; i < n_pvconns; i++) {
926 pvconn_close(pvconns[i]);
930 pvconns = xmalloc(svec->n * sizeof *pvconns);
932 for (i = 0; i < svec->n; i++) {
933 const char *name = svec->names[i];
934 struct pvconn *pvconn;
937 error = pvconn_open(name, &pvconn);
939 pvconns[n_pvconns++] = pvconn;
941 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
949 *n_pvconnsp = n_pvconns;
955 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
957 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
961 ofproto_set_netflow(struct ofproto *ofproto,
962 const struct netflow_options *nf_options)
964 if (nf_options && nf_options->collectors.n) {
965 if (!ofproto->netflow) {
966 ofproto->netflow = netflow_create();
968 return netflow_set_options(ofproto->netflow, nf_options);
970 netflow_destroy(ofproto->netflow);
971 ofproto->netflow = NULL;
977 ofproto_set_sflow(struct ofproto *ofproto,
978 const struct ofproto_sflow_options *oso)
980 struct ofproto_sflow *os = ofproto->sflow;
983 struct ofport *ofport;
985 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
986 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
987 ofproto_sflow_add_port(os, ofport->odp_port,
988 netdev_get_name(ofport->netdev));
991 ofproto_sflow_set_options(os, oso);
993 ofproto_sflow_destroy(os);
994 ofproto->sflow = NULL;
999 ofproto_get_datapath_id(const struct ofproto *ofproto)
1001 return ofproto->datapath_id;
1005 ofproto_has_primary_controller(const struct ofproto *ofproto)
1007 return !hmap_is_empty(&ofproto->controllers);
1010 enum ofproto_fail_mode
1011 ofproto_get_fail_mode(const struct ofproto *p)
1013 return p->fail_mode;
1017 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1021 for (i = 0; i < ofproto->n_snoops; i++) {
1022 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1027 ofproto_destroy(struct ofproto *p)
1029 struct ofservice *ofservice, *next_ofservice;
1030 struct ofconn *ofconn, *next_ofconn;
1031 struct ofport *ofport, *next_ofport;
1038 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1040 /* Destroy fail-open and in-band early, since they touch the classifier. */
1041 fail_open_destroy(p->fail_open);
1042 p->fail_open = NULL;
1044 in_band_destroy(p->in_band);
1046 free(p->extra_in_band_remotes);
1048 ofproto_flush_flows(p);
1049 classifier_destroy(&p->cls);
1050 hmap_destroy(&p->facets);
1052 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1053 ofconn_destroy(ofconn);
1055 hmap_destroy(&p->controllers);
1057 dpif_close(p->dpif);
1058 netdev_monitor_destroy(p->netdev_monitor);
1059 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1060 hmap_remove(&p->ports, &ofport->hmap_node);
1061 ofport_free(ofport);
1063 shash_destroy(&p->port_by_name);
1065 switch_status_destroy(p->switch_status);
1066 netflow_destroy(p->netflow);
1067 ofproto_sflow_destroy(p->sflow);
1069 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1070 ofservice_destroy(p, ofservice);
1072 hmap_destroy(&p->services);
1074 for (i = 0; i < p->n_snoops; i++) {
1075 pvconn_close(p->snoops[i]);
1079 mac_learning_destroy(p->ml);
1084 free(p->serial_desc);
1087 hmap_destroy(&p->ports);
1093 ofproto_run(struct ofproto *p)
1095 int error = ofproto_run1(p);
1097 error = ofproto_run2(p, false);
1103 process_port_change(struct ofproto *ofproto, int error, char *devname)
1105 if (error == ENOBUFS) {
1106 reinit_ports(ofproto);
1107 } else if (!error) {
1108 update_port(ofproto, devname);
1113 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1114 * means that 'ofconn' is more interesting for monitoring than a lower return
1117 snoop_preference(const struct ofconn *ofconn)
1119 switch (ofconn->role) {
1120 case NX_ROLE_MASTER:
1127 /* Shouldn't happen. */
1132 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1133 * Connects this vconn to a controller. */
1135 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1137 struct ofconn *ofconn, *best;
1139 /* Pick a controller for monitoring. */
1141 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1142 if (ofconn->type == OFCONN_PRIMARY
1143 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1149 rconn_add_monitor(best->rconn, vconn);
1151 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1157 ofproto_run1(struct ofproto *p)
1159 struct ofconn *ofconn, *next_ofconn;
1160 struct ofservice *ofservice;
1165 if (shash_is_empty(&p->port_by_name)) {
1169 for (i = 0; i < 50; i++) {
1170 struct dpif_upcall packet;
1172 error = dpif_recv(p->dpif, &packet);
1174 if (error == ENODEV) {
1175 /* Someone destroyed the datapath behind our back. The caller
1176 * better destroy us and give up, because we're just going to
1177 * spin from here on out. */
1178 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1179 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1180 dpif_name(p->dpif));
1186 handle_upcall(p, &packet);
1189 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1190 process_port_change(p, error, devname);
1192 while ((error = netdev_monitor_poll(p->netdev_monitor,
1193 &devname)) != EAGAIN) {
1194 process_port_change(p, error, devname);
1198 if (time_msec() >= p->next_in_band_update) {
1199 update_in_band_remotes(p);
1201 in_band_run(p->in_band);
1204 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1208 /* Fail-open maintenance. Do this after processing the ofconns since
1209 * fail-open checks the status of the controller rconn. */
1211 fail_open_run(p->fail_open);
1214 HMAP_FOR_EACH (ofservice, node, &p->services) {
1215 struct vconn *vconn;
1218 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1220 struct rconn *rconn;
1223 rconn = rconn_create(ofservice->probe_interval, 0);
1224 name = ofconn_make_name(p, vconn_get_name(vconn));
1225 rconn_connect_unreliably(rconn, vconn, name);
1228 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1229 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1230 ofservice->burst_limit);
1231 } else if (retval != EAGAIN) {
1232 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1236 for (i = 0; i < p->n_snoops; i++) {
1237 struct vconn *vconn;
1240 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1242 add_snooper(p, vconn);
1243 } else if (retval != EAGAIN) {
1244 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1248 if (time_msec() >= p->next_expiration) {
1249 int delay = ofproto_expire(p);
1250 p->next_expiration = time_msec() + delay;
1251 COVERAGE_INC(ofproto_expiration);
1255 netflow_run(p->netflow);
1258 ofproto_sflow_run(p->sflow);
1265 ofproto_run2(struct ofproto *p, bool revalidate_all)
1267 /* Figure out what we need to revalidate now, if anything. */
1268 struct tag_set revalidate_set = p->revalidate_set;
1269 if (p->need_revalidate) {
1270 revalidate_all = true;
1273 /* Clear the revalidation flags. */
1274 tag_set_init(&p->revalidate_set);
1275 p->need_revalidate = false;
1277 /* Now revalidate if there's anything to do. */
1278 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1279 struct facet *facet, *next;
1281 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1283 || tag_set_intersects(&revalidate_set, facet->tags)) {
1284 facet_revalidate(p, facet);
1293 ofproto_wait(struct ofproto *p)
1295 struct ofservice *ofservice;
1296 struct ofconn *ofconn;
1299 dpif_recv_wait(p->dpif);
1300 dpif_port_poll_wait(p->dpif);
1301 netdev_monitor_poll_wait(p->netdev_monitor);
1302 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1303 ofconn_wait(ofconn);
1306 poll_timer_wait_until(p->next_in_band_update);
1307 in_band_wait(p->in_band);
1310 fail_open_wait(p->fail_open);
1313 ofproto_sflow_wait(p->sflow);
1315 if (!tag_set_is_empty(&p->revalidate_set)) {
1316 poll_immediate_wake();
1318 if (p->need_revalidate) {
1319 /* Shouldn't happen, but if it does just go around again. */
1320 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1321 poll_immediate_wake();
1322 } else if (p->next_expiration != LLONG_MAX) {
1323 poll_timer_wait_until(p->next_expiration);
1325 HMAP_FOR_EACH (ofservice, node, &p->services) {
1326 pvconn_wait(ofservice->pvconn);
1328 for (i = 0; i < p->n_snoops; i++) {
1329 pvconn_wait(p->snoops[i]);
1334 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1336 tag_set_add(&ofproto->revalidate_set, tag);
1340 ofproto_get_revalidate_set(struct ofproto *ofproto)
1342 return &ofproto->revalidate_set;
1346 ofproto_is_alive(const struct ofproto *p)
1348 return !hmap_is_empty(&p->controllers);
1352 ofproto_get_ofproto_controller_info(const struct ofproto * ofproto,
1355 const struct ofconn *ofconn;
1359 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1360 const struct rconn *rconn = ofconn->rconn;
1361 const int last_error = rconn_get_last_error(rconn);
1362 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1364 shash_add(info, rconn_get_target(rconn), cinfo);
1366 cinfo->is_connected = rconn_is_connected(rconn);
1367 cinfo->role = ofconn->role;
1371 if (last_error == EOF) {
1372 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1373 cinfo->pairs.values[cinfo->pairs.n++] = xstrdup("End of file");
1374 } else if (last_error > 0) {
1375 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1376 cinfo->pairs.values[cinfo->pairs.n++] =
1377 xstrdup(strerror(last_error));
1380 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1381 cinfo->pairs.values[cinfo->pairs.n++] =
1382 xstrdup(rconn_get_state(rconn));
1384 cinfo->pairs.keys[cinfo->pairs.n] = "time_in_state";
1385 cinfo->pairs.values[cinfo->pairs.n++] =
1386 xasprintf("%u", rconn_get_state_elapsed(rconn));
1391 ofproto_free_ofproto_controller_info(struct shash *info)
1393 struct shash_node *node;
1395 SHASH_FOR_EACH (node, info) {
1396 struct ofproto_controller_info *cinfo = node->data;
1397 while (cinfo->pairs.n) {
1398 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1402 shash_destroy(info);
1405 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1407 * This is almost the same as calling dpif_port_del() directly on the
1408 * datapath, but it also makes 'ofproto' close its open netdev for the port
1409 * (if any). This makes it possible to create a new netdev of a different
1410 * type under the same name, which otherwise the netdev library would refuse
1411 * to do because of the conflict. (The netdev would eventually get closed on
1412 * the next trip through ofproto_run(), but this interface is more direct.)
1414 * Returns 0 if successful, otherwise a positive errno. */
1416 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1418 struct ofport *ofport = get_port(ofproto, odp_port);
1419 const char *name = ofport ? ofport->opp.name : "<unknown>";
1422 error = dpif_port_del(ofproto->dpif, odp_port);
1424 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1425 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1426 } else if (ofport) {
1427 /* 'name' is ofport->opp.name and update_port() is going to destroy
1428 * 'ofport'. Just in case update_port() refers to 'name' after it
1429 * destroys 'ofport', make a copy of it around the update_port()
1431 char *devname = xstrdup(name);
1432 update_port(ofproto, devname);
1438 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1439 * true if 'odp_port' exists and should be included, false otherwise. */
1441 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1443 struct ofport *ofport = get_port(ofproto, odp_port);
1444 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1448 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1449 const union ofp_action *actions, size_t n_actions,
1450 const struct ofpbuf *packet)
1452 struct action_xlate_ctx ctx;
1453 struct ofpbuf *odp_actions;
1455 action_xlate_ctx_init(&ctx, p, flow, packet);
1456 odp_actions = xlate_actions(&ctx, actions, n_actions);
1458 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1460 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
1462 ofpbuf_delete(odp_actions);
1467 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1468 * performs the 'n_actions' actions in 'actions'. The new flow will not
1471 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1472 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1473 * controllers; otherwise, it will be hidden.
1475 * The caller retains ownership of 'cls_rule' and 'actions'. */
1477 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1478 const union ofp_action *actions, size_t n_actions)
1481 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1482 rule_insert(p, rule);
1486 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1490 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1493 rule_remove(ofproto, rule);
1498 ofproto_flush_flows(struct ofproto *ofproto)
1500 struct facet *facet, *next_facet;
1501 struct rule *rule, *next_rule;
1502 struct cls_cursor cursor;
1504 COVERAGE_INC(ofproto_flush);
1506 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1507 /* Mark the facet as not installed so that facet_remove() doesn't
1508 * bother trying to uninstall it. There is no point in uninstalling it
1509 * individually since we are about to blow away all the facets with
1510 * dpif_flow_flush(). */
1511 facet->installed = false;
1512 facet_remove(ofproto, facet);
1515 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1516 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1517 rule_remove(ofproto, rule);
1520 dpif_flow_flush(ofproto->dpif);
1521 if (ofproto->in_band) {
1522 in_band_flushed(ofproto->in_band);
1524 if (ofproto->fail_open) {
1525 fail_open_flushed(ofproto->fail_open);
1530 reinit_ports(struct ofproto *p)
1532 struct dpif_port_dump dump;
1533 struct shash_node *node;
1534 struct shash devnames;
1535 struct ofport *ofport;
1536 struct dpif_port dpif_port;
1538 COVERAGE_INC(ofproto_reinit_ports);
1540 shash_init(&devnames);
1541 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1542 shash_add_once (&devnames, ofport->opp.name, NULL);
1544 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1545 shash_add_once (&devnames, dpif_port.name, NULL);
1548 SHASH_FOR_EACH (node, &devnames) {
1549 update_port(p, node->name);
1551 shash_destroy(&devnames);
1554 static struct ofport *
1555 make_ofport(const struct dpif_port *dpif_port)
1557 struct netdev_options netdev_options;
1558 enum netdev_flags flags;
1559 struct ofport *ofport;
1560 struct netdev *netdev;
1563 memset(&netdev_options, 0, sizeof netdev_options);
1564 netdev_options.name = dpif_port->name;
1565 netdev_options.type = dpif_port->type;
1566 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1568 error = netdev_open(&netdev_options, &netdev);
1570 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1571 "cannot be opened (%s)",
1572 dpif_port->name, dpif_port->port_no,
1573 dpif_port->name, strerror(error));
1577 ofport = xmalloc(sizeof *ofport);
1578 ofport->netdev = netdev;
1579 ofport->odp_port = dpif_port->port_no;
1580 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1581 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1582 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1584 netdev_get_flags(netdev, &flags);
1585 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1587 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1589 netdev_get_features(netdev,
1590 &ofport->opp.curr, &ofport->opp.advertised,
1591 &ofport->opp.supported, &ofport->opp.peer);
1596 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1598 if (get_port(p, dpif_port->port_no)) {
1599 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1600 dpif_port->port_no);
1602 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1603 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1612 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1614 const struct ofp_phy_port *a = &a_->opp;
1615 const struct ofp_phy_port *b = &b_->opp;
1617 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1618 return (a->port_no == b->port_no
1619 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1620 && !strcmp(a->name, b->name)
1621 && a->state == b->state
1622 && a->config == b->config
1623 && a->curr == b->curr
1624 && a->advertised == b->advertised
1625 && a->supported == b->supported
1626 && a->peer == b->peer);
1630 send_port_status(struct ofproto *p, const struct ofport *ofport,
1633 /* XXX Should limit the number of queued port status change messages. */
1634 struct ofconn *ofconn;
1635 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1636 struct ofp_port_status *ops;
1639 /* Primary controllers, even slaves, should always get port status
1640 updates. Otherwise obey ofconn_receives_async_msgs(). */
1641 if (ofconn->type != OFCONN_PRIMARY
1642 && !ofconn_receives_async_msgs(ofconn)) {
1646 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1647 ops->reason = reason;
1648 ops->desc = ofport->opp;
1649 hton_ofp_phy_port(&ops->desc);
1650 queue_tx(b, ofconn, NULL);
1655 ofport_install(struct ofproto *p, struct ofport *ofport)
1657 const char *netdev_name = ofport->opp.name;
1659 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1660 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1661 shash_add(&p->port_by_name, netdev_name, ofport);
1663 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1668 ofport_remove(struct ofproto *p, struct ofport *ofport)
1670 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1671 hmap_remove(&p->ports, &ofport->hmap_node);
1672 shash_delete(&p->port_by_name,
1673 shash_find(&p->port_by_name, ofport->opp.name));
1675 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1680 ofport_free(struct ofport *ofport)
1683 netdev_close(ofport->netdev);
1688 static struct ofport *
1689 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1691 struct ofport *port;
1693 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1694 hash_int(odp_port, 0), &ofproto->ports) {
1695 if (port->odp_port == odp_port) {
1703 update_port(struct ofproto *p, const char *devname)
1705 struct dpif_port dpif_port;
1706 struct ofport *old_ofport;
1707 struct ofport *new_ofport;
1710 COVERAGE_INC(ofproto_update_port);
1712 /* Query the datapath for port information. */
1713 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1715 /* Find the old ofport. */
1716 old_ofport = shash_find_data(&p->port_by_name, devname);
1719 /* There's no port named 'devname' but there might be a port with
1720 * the same port number. This could happen if a port is deleted
1721 * and then a new one added in its place very quickly, or if a port
1722 * is renamed. In the former case we want to send an OFPPR_DELETE
1723 * and an OFPPR_ADD, and in the latter case we want to send a
1724 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1725 * the old port's ifindex against the new port, or perhaps less
1726 * reliably but more portably by comparing the old port's MAC
1727 * against the new port's MAC. However, this code isn't that smart
1728 * and always sends an OFPPR_MODIFY (XXX). */
1729 old_ofport = get_port(p, dpif_port.port_no);
1731 } else if (error != ENOENT && error != ENODEV) {
1732 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1733 "%s", strerror(error));
1737 /* Create a new ofport. */
1738 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1740 /* Eliminate a few pathological cases. */
1741 if (!old_ofport && !new_ofport) {
1743 } else if (old_ofport && new_ofport) {
1744 /* Most of the 'config' bits are OpenFlow soft state, but
1745 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1746 * OpenFlow bits from old_ofport. (make_ofport() only sets
1747 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1748 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1750 if (ofport_equal(old_ofport, new_ofport)) {
1751 /* False alarm--no change. */
1752 ofport_free(new_ofport);
1757 /* Now deal with the normal cases. */
1759 ofport_remove(p, old_ofport);
1762 ofport_install(p, new_ofport);
1764 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1765 (!old_ofport ? OFPPR_ADD
1766 : !new_ofport ? OFPPR_DELETE
1768 ofport_free(old_ofport);
1771 dpif_port_destroy(&dpif_port);
1775 init_ports(struct ofproto *p)
1777 struct dpif_port_dump dump;
1778 struct dpif_port dpif_port;
1780 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1781 if (!ofport_conflicts(p, &dpif_port)) {
1782 struct ofport *ofport = make_ofport(&dpif_port);
1784 ofport_install(p, ofport);
1792 static struct ofconn *
1793 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1795 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1796 ofconn->ofproto = p;
1797 list_push_back(&p->all_conns, &ofconn->node);
1798 ofconn->rconn = rconn;
1799 ofconn->type = type;
1800 ofconn->flow_format = NXFF_OPENFLOW10;
1801 ofconn->role = NX_ROLE_OTHER;
1802 ofconn->packet_in_counter = rconn_packet_counter_create ();
1803 ofconn->pktbuf = NULL;
1804 ofconn->miss_send_len = 0;
1805 ofconn->reply_counter = rconn_packet_counter_create ();
1810 ofconn_destroy(struct ofconn *ofconn)
1812 if (ofconn->type == OFCONN_PRIMARY) {
1813 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1815 discovery_destroy(ofconn->discovery);
1817 list_remove(&ofconn->node);
1818 switch_status_unregister(ofconn->ss);
1819 rconn_destroy(ofconn->rconn);
1820 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1821 rconn_packet_counter_destroy(ofconn->reply_counter);
1822 pktbuf_destroy(ofconn->pktbuf);
1827 ofconn_run(struct ofconn *ofconn)
1829 struct ofproto *p = ofconn->ofproto;
1833 if (ofconn->discovery) {
1834 char *controller_name;
1835 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1836 discovery_question_connectivity(ofconn->discovery);
1838 if (discovery_run(ofconn->discovery, &controller_name)) {
1839 if (controller_name) {
1840 char *ofconn_name = ofconn_make_name(p, controller_name);
1841 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1844 rconn_disconnect(ofconn->rconn);
1849 for (i = 0; i < N_SCHEDULERS; i++) {
1850 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1853 rconn_run(ofconn->rconn);
1855 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1856 /* Limit the number of iterations to prevent other tasks from
1858 for (iteration = 0; iteration < 50; iteration++) {
1859 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1864 fail_open_maybe_recover(p->fail_open);
1866 handle_openflow(ofconn, of_msg);
1867 ofpbuf_delete(of_msg);
1871 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1872 ofconn_destroy(ofconn);
1877 ofconn_wait(struct ofconn *ofconn)
1881 if (ofconn->discovery) {
1882 discovery_wait(ofconn->discovery);
1884 for (i = 0; i < N_SCHEDULERS; i++) {
1885 pinsched_wait(ofconn->schedulers[i]);
1887 rconn_run_wait(ofconn->rconn);
1888 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1889 rconn_recv_wait(ofconn->rconn);
1891 COVERAGE_INC(ofproto_ofconn_stuck);
1895 /* Returns true if 'ofconn' should receive asynchronous messages. */
1897 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1899 if (ofconn->type == OFCONN_PRIMARY) {
1900 /* Primary controllers always get asynchronous messages unless they
1901 * have configured themselves as "slaves". */
1902 return ofconn->role != NX_ROLE_SLAVE;
1904 /* Service connections don't get asynchronous messages unless they have
1905 * explicitly asked for them by setting a nonzero miss send length. */
1906 return ofconn->miss_send_len > 0;
1910 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1911 * and 'target', suitable for use in log messages for identifying the
1914 * The name is dynamically allocated. The caller should free it (with free())
1915 * when it is no longer needed. */
1917 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1919 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1923 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1927 for (i = 0; i < N_SCHEDULERS; i++) {
1928 struct pinsched **s = &ofconn->schedulers[i];
1932 *s = pinsched_create(rate, burst,
1933 ofconn->ofproto->switch_status);
1935 pinsched_set_limits(*s, rate, burst);
1938 pinsched_destroy(*s);
1945 ofservice_reconfigure(struct ofservice *ofservice,
1946 const struct ofproto_controller *c)
1948 ofservice->probe_interval = c->probe_interval;
1949 ofservice->rate_limit = c->rate_limit;
1950 ofservice->burst_limit = c->burst_limit;
1953 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1954 * positive errno value. */
1956 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1958 struct ofservice *ofservice;
1959 struct pvconn *pvconn;
1962 error = pvconn_open(c->target, &pvconn);
1967 ofservice = xzalloc(sizeof *ofservice);
1968 hmap_insert(&ofproto->services, &ofservice->node,
1969 hash_string(c->target, 0));
1970 ofservice->pvconn = pvconn;
1972 ofservice_reconfigure(ofservice, c);
1978 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1980 hmap_remove(&ofproto->services, &ofservice->node);
1981 pvconn_close(ofservice->pvconn);
1985 /* Finds and returns the ofservice within 'ofproto' that has the given
1986 * 'target', or a null pointer if none exists. */
1987 static struct ofservice *
1988 ofservice_lookup(struct ofproto *ofproto, const char *target)
1990 struct ofservice *ofservice;
1992 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1993 &ofproto->services) {
1994 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2001 /* Returns true if 'rule' should be hidden from the controller.
2003 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2004 * (e.g. by in-band control) and are intentionally hidden from the
2007 rule_is_hidden(const struct rule *rule)
2009 return rule->cr.priority > UINT16_MAX;
2012 /* Creates and returns a new rule initialized as specified.
2014 * The caller is responsible for inserting the rule into the classifier (with
2015 * rule_insert()). */
2016 static struct rule *
2017 rule_create(const struct cls_rule *cls_rule,
2018 const union ofp_action *actions, size_t n_actions,
2019 uint16_t idle_timeout, uint16_t hard_timeout,
2020 ovs_be64 flow_cookie, bool send_flow_removed)
2022 struct rule *rule = xzalloc(sizeof *rule);
2023 rule->cr = *cls_rule;
2024 rule->idle_timeout = idle_timeout;
2025 rule->hard_timeout = hard_timeout;
2026 rule->flow_cookie = flow_cookie;
2027 rule->used = rule->created = time_msec();
2028 rule->send_flow_removed = send_flow_removed;
2029 list_init(&rule->facets);
2030 if (n_actions > 0) {
2031 rule->n_actions = n_actions;
2032 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2038 static struct rule *
2039 rule_from_cls_rule(const struct cls_rule *cls_rule)
2041 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2045 rule_free(struct rule *rule)
2047 free(rule->actions);
2051 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2052 * destroying any that no longer has a rule (which is probably all of them).
2054 * The caller must have already removed 'rule' from the classifier. */
2056 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2058 struct facet *facet, *next_facet;
2059 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2060 facet_revalidate(ofproto, facet);
2065 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2066 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2069 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2071 const union ofp_action *oa;
2072 struct actions_iterator i;
2074 if (out_port == htons(OFPP_NONE)) {
2077 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2078 oa = actions_next(&i)) {
2079 if (action_outputs_to_port(oa, out_port)) {
2086 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2087 * 'packet', which arrived on 'in_port'.
2089 * Takes ownership of 'packet'. */
2091 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2092 const struct nlattr *odp_actions, size_t actions_len,
2093 struct ofpbuf *packet)
2095 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2096 && odp_actions->nla_type == ODPAT_CONTROLLER) {
2097 /* As an optimization, avoid a round-trip from userspace to kernel to
2098 * userspace. This also avoids possibly filling up kernel packet
2099 * buffers along the way. */
2100 struct dpif_upcall upcall;
2102 upcall.type = _ODPL_ACTION_NR;
2103 upcall.packet = packet;
2106 upcall.userdata = nl_attr_get_u64(odp_actions);
2107 upcall.sample_pool = 0;
2108 upcall.actions = NULL;
2109 upcall.actions_len = 0;
2111 send_packet_in(ofproto, &upcall, flow, false);
2117 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2118 ofpbuf_delete(packet);
2123 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2124 * statistics appropriately. 'packet' must have at least sizeof(struct
2125 * ofp_packet_in) bytes of headroom.
2127 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2128 * applying flow_extract() to 'packet' would yield the same flow as
2131 * 'facet' must have accurately composed ODP actions; that is, it must not be
2132 * in need of revalidation.
2134 * Takes ownership of 'packet'. */
2136 facet_execute(struct ofproto *ofproto, struct facet *facet,
2137 struct ofpbuf *packet)
2139 struct dpif_flow_stats stats;
2141 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2143 flow_extract_stats(&facet->flow, packet, &stats);
2144 if (execute_odp_actions(ofproto, &facet->flow,
2145 facet->actions, facet->actions_len, packet)) {
2146 facet_update_stats(ofproto, facet, &stats);
2147 facet->used = time_msec();
2148 netflow_flow_update_time(ofproto->netflow,
2149 &facet->nf_flow, facet->used);
2153 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2154 * statistics (or the statistics for one of its facets) appropriately.
2155 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2157 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2158 * with statistics for 'packet' either way.
2160 * Takes ownership of 'packet'. */
2162 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2163 struct ofpbuf *packet)
2165 struct action_xlate_ctx ctx;
2166 struct ofpbuf *odp_actions;
2167 struct facet *facet;
2171 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2173 flow_extract(packet, 0, in_port, &flow);
2175 /* First look for a related facet. If we find one, account it to that. */
2176 facet = facet_lookup_valid(ofproto, &flow);
2177 if (facet && facet->rule == rule) {
2178 facet_execute(ofproto, facet, packet);
2182 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2183 * create a new facet for it and use that. */
2184 if (rule_lookup(ofproto, &flow) == rule) {
2185 facet = facet_create(ofproto, rule, &flow, packet);
2186 facet_execute(ofproto, facet, packet);
2187 facet_install(ofproto, facet, true);
2191 /* We can't account anything to a facet. If we were to try, then that
2192 * facet would have a non-matching rule, busting our invariants. */
2193 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2194 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2195 size = packet->size;
2196 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2197 odp_actions->size, packet)) {
2198 rule->used = time_msec();
2199 rule->packet_count++;
2200 rule->byte_count += size;
2202 ofpbuf_delete(odp_actions);
2205 /* Inserts 'rule' into 'p''s flow table. */
2207 rule_insert(struct ofproto *p, struct rule *rule)
2209 struct rule *displaced_rule;
2211 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2212 if (displaced_rule) {
2213 rule_destroy(p, displaced_rule);
2215 p->need_revalidate = true;
2218 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2219 * 'flow' and an example 'packet' within that flow.
2221 * The caller must already have determined that no facet with an identical
2222 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2223 * 'ofproto''s classifier table. */
2224 static struct facet *
2225 facet_create(struct ofproto *ofproto, struct rule *rule,
2226 const struct flow *flow, const struct ofpbuf *packet)
2228 struct facet *facet;
2230 facet = xzalloc(sizeof *facet);
2231 facet->used = time_msec();
2232 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2233 list_push_back(&rule->facets, &facet->list_node);
2235 facet->flow = *flow;
2236 netflow_flow_init(&facet->nf_flow);
2237 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2239 facet_make_actions(ofproto, facet, packet);
2245 facet_free(struct facet *facet)
2247 free(facet->actions);
2251 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2253 * - Removes 'rule' from the classifier.
2255 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2256 * destroys them), via rule_destroy().
2259 rule_remove(struct ofproto *ofproto, struct rule *rule)
2261 COVERAGE_INC(ofproto_del_rule);
2262 ofproto->need_revalidate = true;
2263 classifier_remove(&ofproto->cls, &rule->cr);
2264 rule_destroy(ofproto, rule);
2267 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2269 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2270 * rule's statistics, via facet_uninstall().
2272 * - Removes 'facet' from its rule and from ofproto->facets.
2275 facet_remove(struct ofproto *ofproto, struct facet *facet)
2277 facet_uninstall(ofproto, facet);
2278 facet_flush_stats(ofproto, facet);
2279 hmap_remove(&ofproto->facets, &facet->hmap_node);
2280 list_remove(&facet->list_node);
2284 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2286 facet_make_actions(struct ofproto *p, struct facet *facet,
2287 const struct ofpbuf *packet)
2289 const struct rule *rule = facet->rule;
2290 struct ofpbuf *odp_actions;
2291 struct action_xlate_ctx ctx;
2293 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2294 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2295 facet->tags = ctx.tags;
2296 facet->may_install = ctx.may_set_up_flow;
2297 facet->nf_flow.output_iface = ctx.nf_output_iface;
2299 if (facet->actions_len != odp_actions->size
2300 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2301 free(facet->actions);
2302 facet->actions_len = odp_actions->size;
2303 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2306 ofpbuf_delete(odp_actions);
2310 facet_put__(struct ofproto *ofproto, struct facet *facet,
2311 enum dpif_flow_put_flags flags)
2313 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2316 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2317 odp_flow_key_from_flow(&key, &facet->flow);
2318 assert(key.base == keybuf);
2320 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2321 facet->actions, facet->actions_len, NULL);
2324 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2325 * 'zero_stats' is true, clears any existing statistics from the datapath for
2328 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2330 if (facet->may_install) {
2331 enum dpif_flow_put_flags flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2333 flags |= DPIF_FP_ZERO_STATS;
2335 if (!facet_put__(p, facet, flags)) {
2336 facet->installed = true;
2341 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2342 * to the accounting hook function in the ofhooks structure. */
2344 facet_account(struct ofproto *ofproto,
2345 struct facet *facet, uint64_t extra_bytes)
2347 uint64_t total_bytes = facet->byte_count + extra_bytes;
2349 if (ofproto->ofhooks->account_flow_cb
2350 && total_bytes > facet->accounted_bytes)
2352 ofproto->ofhooks->account_flow_cb(
2353 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2354 total_bytes - facet->accounted_bytes, ofproto->aux);
2355 facet->accounted_bytes = total_bytes;
2359 /* If 'rule' is installed in the datapath, uninstalls it. */
2361 facet_uninstall(struct ofproto *p, struct facet *facet)
2363 if (facet->installed) {
2364 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2365 struct dpif_flow_stats stats;
2368 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2369 odp_flow_key_from_flow(&key, &facet->flow);
2370 assert(key.base == keybuf);
2372 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2373 facet_update_stats(p, facet, &stats);
2375 facet->installed = false;
2379 /* Returns true if the only action for 'facet' is to send to the controller.
2380 * (We don't report NetFlow expiration messages for such facets because they
2381 * are just part of the control logic for the network, not real traffic). */
2383 facet_is_controller_flow(struct facet *facet)
2386 && facet->rule->n_actions == 1
2387 && action_outputs_to_port(&facet->rule->actions[0],
2388 htons(OFPP_CONTROLLER)));
2391 /* Folds all of 'facet''s statistics into its rule. Also updates the
2392 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2394 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2396 facet_account(ofproto, facet, 0);
2398 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2399 struct ofexpired expired;
2400 expired.flow = facet->flow;
2401 expired.packet_count = facet->packet_count;
2402 expired.byte_count = facet->byte_count;
2403 expired.used = facet->used;
2404 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2407 facet->rule->packet_count += facet->packet_count;
2408 facet->rule->byte_count += facet->byte_count;
2410 /* Reset counters to prevent double counting if 'facet' ever gets
2412 facet->packet_count = 0;
2413 facet->byte_count = 0;
2414 facet->accounted_bytes = 0;
2416 netflow_flow_clear(&facet->nf_flow);
2419 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2420 * Returns it if found, otherwise a null pointer.
2422 * The returned facet might need revalidation; use facet_lookup_valid()
2423 * instead if that is important. */
2424 static struct facet *
2425 facet_find(struct ofproto *ofproto, const struct flow *flow)
2427 struct facet *facet;
2429 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2431 if (flow_equal(flow, &facet->flow)) {
2439 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2440 * Returns it if found, otherwise a null pointer.
2442 * The returned facet is guaranteed to be valid. */
2443 static struct facet *
2444 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2446 struct facet *facet = facet_find(ofproto, flow);
2448 /* The facet we found might not be valid, since we could be in need of
2449 * revalidation. If it is not valid, don't return it. */
2451 && ofproto->need_revalidate
2452 && !facet_revalidate(ofproto, facet)) {
2453 COVERAGE_INC(ofproto_invalidated);
2460 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2462 * - If the rule found is different from 'facet''s current rule, moves
2463 * 'facet' to the new rule and recompiles its actions.
2465 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2466 * where it is and recompiles its actions anyway.
2468 * - If there is none, destroys 'facet'.
2470 * Returns true if 'facet' still exists, false if it has been destroyed. */
2472 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2474 struct action_xlate_ctx ctx;
2475 struct ofpbuf *odp_actions;
2476 struct rule *new_rule;
2477 bool actions_changed;
2479 COVERAGE_INC(facet_revalidate);
2481 /* Determine the new rule. */
2482 new_rule = rule_lookup(ofproto, &facet->flow);
2484 /* No new rule, so delete the facet. */
2485 facet_remove(ofproto, facet);
2489 /* Calculate new ODP actions.
2491 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2492 * emit a NetFlow expiration and, if so, we need to have the old state
2493 * around to properly compose it. */
2494 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2495 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2496 actions_changed = (facet->actions_len != odp_actions->size
2497 || memcmp(facet->actions, odp_actions->data,
2498 facet->actions_len));
2500 /* If the ODP actions changed or the installability changed, then we need
2501 * to talk to the datapath. */
2502 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2503 if (ctx.may_set_up_flow) {
2504 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2505 struct dpif_flow_stats stats;
2508 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2509 odp_flow_key_from_flow(&key, &facet->flow);
2511 dpif_flow_put(ofproto->dpif,
2512 DPIF_FP_CREATE | DPIF_FP_MODIFY | DPIF_FP_ZERO_STATS,
2514 odp_actions->data, odp_actions->size, &stats);
2516 facet_update_stats(ofproto, facet, &stats);
2518 facet_uninstall(ofproto, facet);
2521 /* The datapath flow is gone or has zeroed stats, so push stats out of
2522 * 'facet' into 'rule'. */
2523 facet_flush_stats(ofproto, facet);
2526 /* Update 'facet' now that we've taken care of all the old state. */
2527 facet->tags = ctx.tags;
2528 facet->nf_flow.output_iface = ctx.nf_output_iface;
2529 facet->may_install = ctx.may_set_up_flow;
2530 if (actions_changed) {
2531 free(facet->actions);
2532 facet->actions_len = odp_actions->size;
2533 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2535 if (facet->rule != new_rule) {
2536 COVERAGE_INC(facet_changed_rule);
2537 list_remove(&facet->list_node);
2538 list_push_back(&new_rule->facets, &facet->list_node);
2539 facet->rule = new_rule;
2540 facet->used = new_rule->created;
2543 ofpbuf_delete(odp_actions);
2549 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2550 struct rconn_packet_counter *counter)
2552 update_openflow_length(msg);
2553 if (rconn_send(ofconn->rconn, msg, counter)) {
2559 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2562 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2564 COVERAGE_INC(ofproto_error);
2565 queue_tx(buf, ofconn, ofconn->reply_counter);
2570 hton_ofp_phy_port(struct ofp_phy_port *opp)
2572 opp->port_no = htons(opp->port_no);
2573 opp->config = htonl(opp->config);
2574 opp->state = htonl(opp->state);
2575 opp->curr = htonl(opp->curr);
2576 opp->advertised = htonl(opp->advertised);
2577 opp->supported = htonl(opp->supported);
2578 opp->peer = htonl(opp->peer);
2582 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2584 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2589 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2591 struct ofp_switch_features *osf;
2593 struct ofport *port;
2595 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2596 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2597 osf->n_buffers = htonl(pktbuf_capacity());
2599 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2600 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2601 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2602 (1u << OFPAT_SET_VLAN_VID) |
2603 (1u << OFPAT_SET_VLAN_PCP) |
2604 (1u << OFPAT_STRIP_VLAN) |
2605 (1u << OFPAT_SET_DL_SRC) |
2606 (1u << OFPAT_SET_DL_DST) |
2607 (1u << OFPAT_SET_NW_SRC) |
2608 (1u << OFPAT_SET_NW_DST) |
2609 (1u << OFPAT_SET_NW_TOS) |
2610 (1u << OFPAT_SET_TP_SRC) |
2611 (1u << OFPAT_SET_TP_DST) |
2612 (1u << OFPAT_ENQUEUE));
2614 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2615 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2618 queue_tx(buf, ofconn, ofconn->reply_counter);
2623 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2626 struct ofp_switch_config *osc;
2630 /* Figure out flags. */
2631 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2632 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2635 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2636 osc->flags = htons(flags);
2637 osc->miss_send_len = htons(ofconn->miss_send_len);
2638 queue_tx(buf, ofconn, ofconn->reply_counter);
2644 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2646 uint16_t flags = ntohs(osc->flags);
2648 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2649 switch (flags & OFPC_FRAG_MASK) {
2650 case OFPC_FRAG_NORMAL:
2651 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2653 case OFPC_FRAG_DROP:
2654 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2657 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2663 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2668 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2669 * flow translation. */
2670 #define MAX_RESUBMIT_RECURSION 16
2672 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2673 struct action_xlate_ctx *ctx);
2676 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2678 const struct ofport *ofport = get_port(ctx->ofproto, port);
2681 if (ofport->opp.config & OFPPC_NO_FWD) {
2682 /* Forwarding disabled on port. */
2687 * We don't have an ofport record for this port, but it doesn't hurt to
2688 * allow forwarding to it anyhow. Maybe such a port will appear later
2689 * and we're pre-populating the flow table.
2693 nl_msg_put_u32(ctx->odp_actions, ODPAT_OUTPUT, port);
2694 ctx->nf_output_iface = port;
2697 static struct rule *
2698 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2700 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2704 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2706 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2707 uint16_t old_in_port;
2710 /* Look up a flow with 'in_port' as the input port. Then restore the
2711 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2712 * have surprising behavior). */
2713 old_in_port = ctx->flow.in_port;
2714 ctx->flow.in_port = in_port;
2715 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2716 ctx->flow.in_port = old_in_port;
2718 if (ctx->resubmit_hook) {
2719 ctx->resubmit_hook(ctx, rule);
2724 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2728 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2730 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2731 MAX_RESUBMIT_RECURSION);
2736 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2737 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2739 struct ofport *ofport;
2741 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2742 uint16_t odp_port = ofport->odp_port;
2743 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2744 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, odp_port);
2747 *nf_output_iface = NF_OUT_FLOOD;
2751 xlate_output_action__(struct action_xlate_ctx *ctx,
2752 uint16_t port, uint16_t max_len)
2755 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2757 ctx->nf_output_iface = NF_OUT_DROP;
2761 add_output_action(ctx, ctx->flow.in_port);
2764 xlate_table_action(ctx, ctx->flow.in_port);
2767 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2768 ctx->odp_actions, &ctx->tags,
2769 &ctx->nf_output_iface,
2770 ctx->ofproto->aux)) {
2771 COVERAGE_INC(ofproto_uninstallable);
2772 ctx->may_set_up_flow = false;
2776 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2777 &ctx->nf_output_iface, ctx->odp_actions);
2780 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2781 &ctx->nf_output_iface, ctx->odp_actions);
2783 case OFPP_CONTROLLER:
2784 nl_msg_put_u64(ctx->odp_actions, ODPAT_CONTROLLER, max_len);
2787 add_output_action(ctx, ODPP_LOCAL);
2790 odp_port = ofp_port_to_odp_port(port);
2791 if (odp_port != ctx->flow.in_port) {
2792 add_output_action(ctx, odp_port);
2797 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2798 ctx->nf_output_iface = NF_OUT_FLOOD;
2799 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2800 ctx->nf_output_iface = prev_nf_output_iface;
2801 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2802 ctx->nf_output_iface != NF_OUT_FLOOD) {
2803 ctx->nf_output_iface = NF_OUT_MULTI;
2808 xlate_output_action(struct action_xlate_ctx *ctx,
2809 const struct ofp_action_output *oao)
2811 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2814 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2815 * optimization, because we're going to add another action that sets the
2816 * priority immediately after, or because there are no actions following the
2819 remove_pop_action(struct action_xlate_ctx *ctx)
2821 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2822 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2823 ctx->last_pop_priority = -1;
2828 add_pop_action(struct action_xlate_ctx *ctx)
2830 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2831 nl_msg_put_flag(ctx->odp_actions, ODPAT_POP_PRIORITY);
2832 ctx->last_pop_priority = ctx->odp_actions->size;
2837 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2838 const struct ofp_action_enqueue *oae)
2840 uint16_t ofp_port, odp_port;
2844 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2847 /* Fall back to ordinary output action. */
2848 xlate_output_action__(ctx, ntohs(oae->port), 0);
2852 /* Figure out ODP output port. */
2853 ofp_port = ntohs(oae->port);
2854 if (ofp_port != OFPP_IN_PORT) {
2855 odp_port = ofp_port_to_odp_port(ofp_port);
2857 odp_port = ctx->flow.in_port;
2860 /* Add ODP actions. */
2861 remove_pop_action(ctx);
2862 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2863 add_output_action(ctx, odp_port);
2864 add_pop_action(ctx);
2866 /* Update NetFlow output port. */
2867 if (ctx->nf_output_iface == NF_OUT_DROP) {
2868 ctx->nf_output_iface = odp_port;
2869 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2870 ctx->nf_output_iface = NF_OUT_MULTI;
2875 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2876 const struct nx_action_set_queue *nasq)
2881 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2884 /* Couldn't translate queue to a priority, so ignore. A warning
2885 * has already been logged. */
2889 remove_pop_action(ctx);
2890 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2894 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2896 ovs_be16 tci = ctx->flow.vlan_tci;
2897 if (!(tci & htons(VLAN_CFI))) {
2898 nl_msg_put_flag(ctx->odp_actions, ODPAT_STRIP_VLAN);
2900 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_DL_TCI,
2901 tci & ~htons(VLAN_CFI));
2905 struct xlate_reg_state {
2911 save_reg_state(const struct action_xlate_ctx *ctx,
2912 struct xlate_reg_state *state)
2914 state->vlan_tci = ctx->flow.vlan_tci;
2915 state->tun_id = ctx->flow.tun_id;
2919 update_reg_state(struct action_xlate_ctx *ctx,
2920 const struct xlate_reg_state *state)
2922 if (ctx->flow.vlan_tci != state->vlan_tci) {
2923 xlate_set_dl_tci(ctx);
2925 if (ctx->flow.tun_id != state->tun_id) {
2926 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, ctx->flow.tun_id);
2931 xlate_nicira_action(struct action_xlate_ctx *ctx,
2932 const struct nx_action_header *nah)
2934 const struct nx_action_resubmit *nar;
2935 const struct nx_action_set_tunnel *nast;
2936 const struct nx_action_set_queue *nasq;
2937 const struct nx_action_multipath *nam;
2938 enum nx_action_subtype subtype = ntohs(nah->subtype);
2939 struct xlate_reg_state state;
2942 assert(nah->vendor == htonl(NX_VENDOR_ID));
2944 case NXAST_RESUBMIT:
2945 nar = (const struct nx_action_resubmit *) nah;
2946 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2949 case NXAST_SET_TUNNEL:
2950 nast = (const struct nx_action_set_tunnel *) nah;
2951 tun_id = htonll(ntohl(nast->tun_id));
2952 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id);
2953 ctx->flow.tun_id = tun_id;
2956 case NXAST_DROP_SPOOFED_ARP:
2957 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2958 nl_msg_put_flag(ctx->odp_actions, ODPAT_DROP_SPOOFED_ARP);
2962 case NXAST_SET_QUEUE:
2963 nasq = (const struct nx_action_set_queue *) nah;
2964 xlate_set_queue_action(ctx, nasq);
2967 case NXAST_POP_QUEUE:
2968 add_pop_action(ctx);
2971 case NXAST_REG_MOVE:
2972 save_reg_state(ctx, &state);
2973 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2975 update_reg_state(ctx, &state);
2978 case NXAST_REG_LOAD:
2979 save_reg_state(ctx, &state);
2980 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2982 update_reg_state(ctx, &state);
2986 /* Nothing to do. */
2989 case NXAST_SET_TUNNEL64:
2990 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2991 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id);
2992 ctx->flow.tun_id = tun_id;
2995 case NXAST_MULTIPATH:
2996 nam = (const struct nx_action_multipath *) nah;
2997 multipath_execute(nam, &ctx->flow);
3000 /* If you add a new action here that modifies flow data, don't forget to
3001 * update the flow key in ctx->flow at the same time. */
3003 case NXAST_SNAT__OBSOLETE:
3005 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3011 do_xlate_actions(const union ofp_action *in, size_t n_in,
3012 struct action_xlate_ctx *ctx)
3014 struct actions_iterator iter;
3015 const union ofp_action *ia;
3016 const struct ofport *port;
3018 port = get_port(ctx->ofproto, ctx->flow.in_port);
3019 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3020 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3021 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3022 /* Drop this flow. */
3026 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3027 enum ofp_action_type type = ntohs(ia->type);
3028 const struct ofp_action_dl_addr *oada;
3032 xlate_output_action(ctx, &ia->output);
3035 case OFPAT_SET_VLAN_VID:
3036 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3037 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3038 xlate_set_dl_tci(ctx);
3041 case OFPAT_SET_VLAN_PCP:
3042 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3043 ctx->flow.vlan_tci |= htons(
3044 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3045 xlate_set_dl_tci(ctx);
3048 case OFPAT_STRIP_VLAN:
3049 ctx->flow.vlan_tci = htons(0);
3050 xlate_set_dl_tci(ctx);
3053 case OFPAT_SET_DL_SRC:
3054 oada = ((struct ofp_action_dl_addr *) ia);
3055 nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_SRC,
3056 oada->dl_addr, ETH_ADDR_LEN);
3057 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3060 case OFPAT_SET_DL_DST:
3061 oada = ((struct ofp_action_dl_addr *) ia);
3062 nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_DST,
3063 oada->dl_addr, ETH_ADDR_LEN);
3064 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3067 case OFPAT_SET_NW_SRC:
3068 nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_SRC,
3069 ia->nw_addr.nw_addr);
3070 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3073 case OFPAT_SET_NW_DST:
3074 nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_DST,
3075 ia->nw_addr.nw_addr);
3076 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3079 case OFPAT_SET_NW_TOS:
3080 nl_msg_put_u8(ctx->odp_actions, ODPAT_SET_NW_TOS,
3082 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3085 case OFPAT_SET_TP_SRC:
3086 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_SRC,
3087 ia->tp_port.tp_port);
3088 ctx->flow.tp_src = ia->tp_port.tp_port;
3091 case OFPAT_SET_TP_DST:
3092 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_DST,
3093 ia->tp_port.tp_port);
3094 ctx->flow.tp_dst = ia->tp_port.tp_port;
3098 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3102 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3106 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3113 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3114 struct ofproto *ofproto, const struct flow *flow,
3115 const struct ofpbuf *packet)
3117 ctx->ofproto = ofproto;
3119 ctx->packet = packet;
3120 ctx->resubmit_hook = NULL;
3123 static struct ofpbuf *
3124 xlate_actions(struct action_xlate_ctx *ctx,
3125 const union ofp_action *in, size_t n_in)
3127 COVERAGE_INC(ofproto_ofp2odp);
3129 ctx->odp_actions = ofpbuf_new(512);
3131 ctx->may_set_up_flow = true;
3132 ctx->nf_output_iface = NF_OUT_DROP;
3134 ctx->last_pop_priority = -1;
3135 do_xlate_actions(in, n_in, ctx);
3136 remove_pop_action(ctx);
3138 /* Check with in-band control to see if we're allowed to set up this
3140 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3141 ctx->odp_actions->data, ctx->odp_actions->size)) {
3142 ctx->may_set_up_flow = false;
3145 return ctx->odp_actions;
3148 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3149 * error message code (composed with ofp_mkerr()) for the caller to propagate
3150 * upward. Otherwise, returns 0.
3152 * The log message mentions 'msg_type'. */
3154 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3156 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3157 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3158 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3161 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3168 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3170 struct ofproto *p = ofconn->ofproto;
3171 struct ofp_packet_out *opo;
3172 struct ofpbuf payload, *buffer;
3173 union ofp_action *ofp_actions;
3174 struct action_xlate_ctx ctx;
3175 struct ofpbuf *odp_actions;
3176 struct ofpbuf request;
3178 size_t n_ofp_actions;
3182 COVERAGE_INC(ofproto_packet_out);
3184 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3189 /* Get ofp_packet_out. */
3190 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3191 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3194 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3195 &ofp_actions, &n_ofp_actions);
3201 if (opo->buffer_id != htonl(UINT32_MAX)) {
3202 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3204 if (error || !buffer) {
3213 /* Extract flow, check actions. */
3214 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3216 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3222 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3223 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3224 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3225 ofpbuf_delete(odp_actions);
3228 ofpbuf_delete(buffer);
3233 update_port_config(struct ofproto *p, struct ofport *port,
3234 uint32_t config, uint32_t mask)
3236 mask &= config ^ port->opp.config;
3237 if (mask & OFPPC_PORT_DOWN) {
3238 if (config & OFPPC_PORT_DOWN) {
3239 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3241 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3244 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3245 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3246 if (mask & REVALIDATE_BITS) {
3247 COVERAGE_INC(ofproto_costly_flags);
3248 port->opp.config ^= mask & REVALIDATE_BITS;
3249 p->need_revalidate = true;
3251 #undef REVALIDATE_BITS
3252 if (mask & OFPPC_NO_PACKET_IN) {
3253 port->opp.config ^= OFPPC_NO_PACKET_IN;
3258 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3260 struct ofproto *p = ofconn->ofproto;
3261 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3262 struct ofport *port;
3265 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3270 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3272 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3273 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3274 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3276 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3277 if (opm->advertise) {
3278 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3284 static struct ofpbuf *
3285 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3287 struct ofp_stats_reply *osr;
3290 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3291 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3293 osr->flags = htons(0);
3297 static struct ofpbuf *
3298 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3300 const struct ofp_stats_request *osr
3301 = (const struct ofp_stats_request *) request;
3302 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3306 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3307 struct ofpbuf **msgp)
3309 struct ofpbuf *msg = *msgp;
3310 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3311 if (nbytes + msg->size > UINT16_MAX) {
3312 struct ofp_stats_reply *reply = msg->data;
3313 reply->flags = htons(OFPSF_REPLY_MORE);
3314 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3315 queue_tx(msg, ofconn, ofconn->reply_counter);
3317 return ofpbuf_put_uninit(*msgp, nbytes);
3320 static struct ofpbuf *
3321 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3323 struct nicira_stats_msg *nsm;
3326 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3327 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3328 nsm->type = htons(OFPST_VENDOR);
3329 nsm->flags = htons(0);
3330 nsm->vendor = htonl(NX_VENDOR_ID);
3331 nsm->subtype = subtype;
3335 static struct ofpbuf *
3336 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3338 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3342 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3343 struct ofpbuf **msgp)
3345 struct ofpbuf *msg = *msgp;
3346 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3347 if (nbytes + msg->size > UINT16_MAX) {
3348 struct nicira_stats_msg *reply = msg->data;
3349 reply->flags = htons(OFPSF_REPLY_MORE);
3350 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3351 queue_tx(msg, ofconn, ofconn->reply_counter);
3353 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3357 handle_desc_stats_request(struct ofconn *ofconn,
3358 const struct ofp_header *request)
3360 struct ofproto *p = ofconn->ofproto;
3361 struct ofp_desc_stats *ods;
3364 msg = start_ofp_stats_reply(request, sizeof *ods);
3365 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3366 memset(ods, 0, sizeof *ods);
3367 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3368 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3369 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3370 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3371 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3372 queue_tx(msg, ofconn, ofconn->reply_counter);
3378 handle_table_stats_request(struct ofconn *ofconn,
3379 const struct ofp_header *request)
3381 struct ofproto *p = ofconn->ofproto;
3382 struct ofp_table_stats *ots;
3385 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3387 /* Classifier table. */
3388 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3389 memset(ots, 0, sizeof *ots);
3390 strcpy(ots->name, "classifier");
3391 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3392 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3393 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3394 ots->active_count = htonl(classifier_count(&p->cls));
3395 ots->lookup_count = htonll(0); /* XXX */
3396 ots->matched_count = htonll(0); /* XXX */
3398 queue_tx(msg, ofconn, ofconn->reply_counter);
3403 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3404 struct ofpbuf **msgp)
3406 struct netdev_stats stats;
3407 struct ofp_port_stats *ops;
3409 /* Intentionally ignore return value, since errors will set
3410 * 'stats' to all-1s, which is correct for OpenFlow, and
3411 * netdev_get_stats() will log errors. */
3412 netdev_get_stats(port->netdev, &stats);
3414 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3415 ops->port_no = htons(port->opp.port_no);
3416 memset(ops->pad, 0, sizeof ops->pad);
3417 ops->rx_packets = htonll(stats.rx_packets);
3418 ops->tx_packets = htonll(stats.tx_packets);
3419 ops->rx_bytes = htonll(stats.rx_bytes);
3420 ops->tx_bytes = htonll(stats.tx_bytes);
3421 ops->rx_dropped = htonll(stats.rx_dropped);
3422 ops->tx_dropped = htonll(stats.tx_dropped);
3423 ops->rx_errors = htonll(stats.rx_errors);
3424 ops->tx_errors = htonll(stats.tx_errors);
3425 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3426 ops->rx_over_err = htonll(stats.rx_over_errors);
3427 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3428 ops->collisions = htonll(stats.collisions);
3432 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3434 struct ofproto *p = ofconn->ofproto;
3435 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3436 struct ofp_port_stats *ops;
3438 struct ofport *port;
3440 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3441 if (psr->port_no != htons(OFPP_NONE)) {
3442 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3444 append_port_stat(port, ofconn, &msg);
3447 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3448 append_port_stat(port, ofconn, &msg);
3452 queue_tx(msg, ofconn, ofconn->reply_counter);
3456 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3457 * '*packet_countp' and '*byte_countp'. The returned statistics include
3458 * statistics for all of 'rule''s facets. */
3460 query_stats(struct ofproto *p, struct rule *rule,
3461 uint64_t *packet_countp, uint64_t *byte_countp)
3463 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
3464 uint64_t packet_count, byte_count;
3465 struct facet *facet;
3468 /* Start from historical data for 'rule' itself that are no longer tracked
3469 * by the datapath. This counts, for example, facets that have expired. */
3470 packet_count = rule->packet_count;
3471 byte_count = rule->byte_count;
3473 /* Ask the datapath for statistics on all of the rule's facets.
3475 * Also, add any statistics that are not tracked by the datapath for each
3476 * facet. This includes, for example, statistics for packets that were
3477 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3479 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
3480 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3481 struct dpif_flow_stats stats;
3484 odp_flow_key_from_flow(&key, &facet->flow);
3485 dpif_flow_get(p->dpif, 0, key.data, key.size, NULL, &stats);
3487 packet_count += stats.n_packets + facet->packet_count;
3488 byte_count += stats.n_bytes + facet->byte_count;
3491 /* Return the stats to the caller. */
3492 *packet_countp = packet_count;
3493 *byte_countp = byte_count;
3497 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3499 long long int msecs = time_msec() - start;
3500 *sec = htonl(msecs / 1000);
3501 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3505 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3506 ovs_be16 out_port, struct ofpbuf **replyp)
3508 struct ofp_flow_stats *ofs;
3509 uint64_t packet_count, byte_count;
3510 size_t act_len, len;
3512 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3516 act_len = sizeof *rule->actions * rule->n_actions;
3517 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3519 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3521 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3522 ofs->length = htons(len);
3525 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3526 rule->flow_cookie, &ofs->cookie);
3527 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3528 ofs->priority = htons(rule->cr.priority);
3529 ofs->idle_timeout = htons(rule->idle_timeout);
3530 ofs->hard_timeout = htons(rule->hard_timeout);
3531 memset(ofs->pad2, 0, sizeof ofs->pad2);
3532 ofs->packet_count = htonll(packet_count);
3533 ofs->byte_count = htonll(byte_count);
3534 if (rule->n_actions > 0) {
3535 memcpy(ofs->actions, rule->actions, act_len);
3540 is_valid_table(uint8_t table_id)
3542 return table_id == 0 || table_id == 0xff;
3546 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3548 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3549 struct ofpbuf *reply;
3551 COVERAGE_INC(ofproto_flows_req);
3552 reply = start_ofp_stats_reply(oh, 1024);
3553 if (is_valid_table(fsr->table_id)) {
3554 struct cls_cursor cursor;
3555 struct cls_rule target;
3558 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3560 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3561 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3562 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3565 queue_tx(reply, ofconn, ofconn->reply_counter);
3571 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3572 ovs_be16 out_port, struct ofpbuf **replyp)
3574 struct nx_flow_stats *nfs;
3575 uint64_t packet_count, byte_count;
3576 size_t act_len, start_len;
3577 struct ofpbuf *reply;
3579 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3583 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3585 act_len = sizeof *rule->actions * rule->n_actions;
3587 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3588 start_len = (*replyp)->size;
3591 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3594 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3595 nfs->cookie = rule->flow_cookie;
3596 nfs->priority = htons(rule->cr.priority);
3597 nfs->idle_timeout = htons(rule->idle_timeout);
3598 nfs->hard_timeout = htons(rule->hard_timeout);
3599 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3600 memset(nfs->pad2, 0, sizeof nfs->pad2);
3601 nfs->packet_count = htonll(packet_count);
3602 nfs->byte_count = htonll(byte_count);
3603 if (rule->n_actions > 0) {
3604 ofpbuf_put(reply, rule->actions, act_len);
3606 nfs->length = htons(reply->size - start_len);
3610 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3612 struct nx_flow_stats_request *nfsr;
3613 struct cls_rule target;
3614 struct ofpbuf *reply;
3618 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3620 /* Dissect the message. */
3621 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3622 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3627 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3630 COVERAGE_INC(ofproto_flows_req);
3631 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3632 if (is_valid_table(nfsr->table_id)) {
3633 struct cls_cursor cursor;
3636 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3637 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3638 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3641 queue_tx(reply, ofconn, ofconn->reply_counter);
3647 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3649 uint64_t packet_count, byte_count;
3650 size_t act_len = sizeof *rule->actions * rule->n_actions;
3652 query_stats(ofproto, rule, &packet_count, &byte_count);
3654 ds_put_format(results, "duration=%llds, ",
3655 (time_msec() - rule->created) / 1000);
3656 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3657 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3658 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3659 cls_rule_format(&rule->cr, results);
3661 ofp_print_actions(results, &rule->actions->header, act_len);
3663 ds_put_cstr(results, "drop");
3665 ds_put_cstr(results, "\n");
3668 /* Adds a pretty-printed description of all flows to 'results', including
3669 * those marked hidden by secchan (e.g., by in-band control). */
3671 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3673 struct cls_cursor cursor;
3676 cls_cursor_init(&cursor, &p->cls, NULL);
3677 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3678 flow_stats_ds(p, rule, results);
3683 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3684 ovs_be16 out_port, uint8_t table_id,
3685 struct ofp_aggregate_stats_reply *oasr)
3687 uint64_t total_packets = 0;
3688 uint64_t total_bytes = 0;
3691 COVERAGE_INC(ofproto_agg_request);
3693 if (is_valid_table(table_id)) {
3694 struct cls_cursor cursor;
3697 cls_cursor_init(&cursor, &ofproto->cls, target);
3698 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3699 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3700 uint64_t packet_count;
3701 uint64_t byte_count;
3703 query_stats(ofproto, rule, &packet_count, &byte_count);
3705 total_packets += packet_count;
3706 total_bytes += byte_count;
3712 oasr->flow_count = htonl(n_flows);
3713 oasr->packet_count = htonll(total_packets);
3714 oasr->byte_count = htonll(total_bytes);
3715 memset(oasr->pad, 0, sizeof oasr->pad);
3719 handle_aggregate_stats_request(struct ofconn *ofconn,
3720 const struct ofp_header *oh)
3722 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3723 struct ofp_aggregate_stats_reply *reply;
3724 struct cls_rule target;
3727 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3730 msg = start_ofp_stats_reply(oh, sizeof *reply);
3731 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3732 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3733 request->table_id, reply);
3734 queue_tx(msg, ofconn, ofconn->reply_counter);
3739 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3741 struct nx_aggregate_stats_request *request;
3742 struct ofp_aggregate_stats_reply *reply;
3743 struct cls_rule target;
3748 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3750 /* Dissect the message. */
3751 request = ofpbuf_pull(&b, sizeof *request);
3752 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3757 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3761 COVERAGE_INC(ofproto_flows_req);
3762 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3763 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3764 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3765 request->table_id, reply);
3766 queue_tx(buf, ofconn, ofconn->reply_counter);
3771 struct queue_stats_cbdata {
3772 struct ofconn *ofconn;
3773 struct ofport *ofport;
3778 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3779 const struct netdev_queue_stats *stats)
3781 struct ofp_queue_stats *reply;
3783 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3784 reply->port_no = htons(cbdata->ofport->opp.port_no);
3785 memset(reply->pad, 0, sizeof reply->pad);
3786 reply->queue_id = htonl(queue_id);
3787 reply->tx_bytes = htonll(stats->tx_bytes);
3788 reply->tx_packets = htonll(stats->tx_packets);
3789 reply->tx_errors = htonll(stats->tx_errors);
3793 handle_queue_stats_dump_cb(uint32_t queue_id,
3794 struct netdev_queue_stats *stats,
3797 struct queue_stats_cbdata *cbdata = cbdata_;
3799 put_queue_stats(cbdata, queue_id, stats);
3803 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3804 struct queue_stats_cbdata *cbdata)
3806 cbdata->ofport = port;
3807 if (queue_id == OFPQ_ALL) {
3808 netdev_dump_queue_stats(port->netdev,
3809 handle_queue_stats_dump_cb, cbdata);
3811 struct netdev_queue_stats stats;
3813 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3814 put_queue_stats(cbdata, queue_id, &stats);
3820 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3822 struct ofproto *ofproto = ofconn->ofproto;
3823 const struct ofp_queue_stats_request *qsr;
3824 struct queue_stats_cbdata cbdata;
3825 struct ofport *port;
3826 unsigned int port_no;
3829 qsr = ofputil_stats_body(oh);
3831 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3834 COVERAGE_INC(ofproto_queue_req);
3836 cbdata.ofconn = ofconn;
3837 cbdata.msg = start_ofp_stats_reply(oh, 128);
3839 port_no = ntohs(qsr->port_no);
3840 queue_id = ntohl(qsr->queue_id);
3841 if (port_no == OFPP_ALL) {
3842 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3843 handle_queue_stats_for_port(port, queue_id, &cbdata);
3845 } else if (port_no < ofproto->max_ports) {
3846 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3848 handle_queue_stats_for_port(port, queue_id, &cbdata);
3851 ofpbuf_delete(cbdata.msg);
3852 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3854 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3860 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3861 const struct dpif_flow_stats *stats)
3863 long long int used = stats->used;
3864 if (used > facet->used) {
3866 if (used > facet->rule->used) {
3867 facet->rule->used = used;
3869 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3873 /* Folds the statistics from 'stats' into the counters in 'facet'.
3875 * Because of the meaning of a facet's counters, it only makes sense to do this
3876 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3877 * packet that was sent by hand or if it represents statistics that have been
3878 * cleared out of the datapath. */
3880 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3881 const struct dpif_flow_stats *stats)
3883 if (stats->n_packets) {
3884 facet_update_time(ofproto, facet, stats);
3885 facet->packet_count += stats->n_packets;
3886 facet->byte_count += stats->n_bytes;
3887 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3891 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3892 * in which no matching flow already exists in the flow table.
3894 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3895 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3896 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3898 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3901 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3903 struct ofproto *p = ofconn->ofproto;
3904 struct ofpbuf *packet;
3909 if (fm->flags & OFPFF_CHECK_OVERLAP
3910 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3911 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3915 if (fm->buffer_id != UINT32_MAX) {
3916 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3920 in_port = UINT16_MAX;
3923 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3924 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3925 fm->flags & OFPFF_SEND_FLOW_REM);
3926 rule_insert(p, rule);
3928 rule_execute(p, rule, in_port, packet);
3933 static struct rule *
3934 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3936 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3940 send_buffered_packet(struct ofconn *ofconn,
3941 struct rule *rule, uint32_t buffer_id)
3943 struct ofpbuf *packet;
3947 if (buffer_id == UINT32_MAX) {
3951 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3956 rule_execute(ofconn->ofproto, rule, in_port, packet);
3961 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3963 struct modify_flows_cbdata {
3964 struct ofproto *ofproto;
3965 const struct flow_mod *fm;
3969 static int modify_flow(struct ofproto *, const struct flow_mod *,
3972 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3973 * encoded by ofp_mkerr() on failure.
3975 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3978 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3980 struct ofproto *p = ofconn->ofproto;
3981 struct rule *match = NULL;
3982 struct cls_cursor cursor;
3985 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3986 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3987 if (!rule_is_hidden(rule)) {
3989 modify_flow(p, fm, rule);
3994 /* This credits the packet to whichever flow happened to match last.
3995 * That's weird. Maybe we should do a lookup for the flow that
3996 * actually matches the packet? Who knows. */
3997 send_buffered_packet(ofconn, match, fm->buffer_id);
4000 return add_flow(ofconn, fm);
4004 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4005 * code as encoded by ofp_mkerr() on failure.
4007 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4010 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4012 struct ofproto *p = ofconn->ofproto;
4013 struct rule *rule = find_flow_strict(p, fm);
4014 if (rule && !rule_is_hidden(rule)) {
4015 modify_flow(p, fm, rule);
4016 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4018 return add_flow(ofconn, fm);
4022 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4023 * been identified as a flow in 'p''s flow table to be modified, by changing
4024 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4025 * ofp_action[] structures). */
4027 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4029 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4031 rule->flow_cookie = fm->cookie;
4033 /* If the actions are the same, do nothing. */
4034 if (fm->n_actions == rule->n_actions
4036 || !memcmp(fm->actions, rule->actions, actions_len))) {
4040 /* Replace actions. */
4041 free(rule->actions);
4042 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4043 rule->n_actions = fm->n_actions;
4045 p->need_revalidate = true;
4050 /* OFPFC_DELETE implementation. */
4052 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4054 /* Implements OFPFC_DELETE. */
4056 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4058 struct rule *rule, *next_rule;
4059 struct cls_cursor cursor;
4061 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4062 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4063 delete_flow(p, rule, htons(fm->out_port));
4067 /* Implements OFPFC_DELETE_STRICT. */
4069 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4071 struct rule *rule = find_flow_strict(p, fm);
4073 delete_flow(p, rule, htons(fm->out_port));
4077 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4078 * been identified as a flow to delete from 'p''s flow table, by deleting the
4079 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4082 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4083 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4084 * specified 'out_port'. */
4086 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4088 if (rule_is_hidden(rule)) {
4092 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4096 rule_send_removed(p, rule, OFPRR_DELETE);
4097 rule_remove(p, rule);
4101 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4103 struct ofproto *p = ofconn->ofproto;
4107 error = reject_slave_controller(ofconn, "flow_mod");
4112 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4117 /* We do not support the emergency flow cache. It will hopefully get
4118 * dropped from OpenFlow in the near future. */
4119 if (fm.flags & OFPFF_EMERG) {
4120 /* There isn't a good fit for an error code, so just state that the
4121 * flow table is full. */
4122 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4125 error = validate_actions(fm.actions, fm.n_actions,
4126 &fm.cr.flow, p->max_ports);
4131 switch (fm.command) {
4133 return add_flow(ofconn, &fm);
4136 return modify_flows_loose(ofconn, &fm);
4138 case OFPFC_MODIFY_STRICT:
4139 return modify_flow_strict(ofconn, &fm);
4142 delete_flows_loose(p, &fm);
4145 case OFPFC_DELETE_STRICT:
4146 delete_flow_strict(p, &fm);
4150 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4155 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4157 const struct nxt_tun_id_cookie *msg
4158 = (const struct nxt_tun_id_cookie *) oh;
4160 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4165 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4167 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4168 struct nx_role_request *reply;
4172 if (ofconn->type != OFCONN_PRIMARY) {
4173 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4175 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4178 role = ntohl(nrr->role);
4179 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4180 && role != NX_ROLE_SLAVE) {
4181 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4183 /* There's no good error code for this. */
4184 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4187 if (role == NX_ROLE_MASTER) {
4188 struct ofconn *other;
4190 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4191 if (other->role == NX_ROLE_MASTER) {
4192 other->role = NX_ROLE_SLAVE;
4196 ofconn->role = role;
4198 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4199 reply->role = htonl(role);
4200 queue_tx(buf, ofconn, ofconn->reply_counter);
4206 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4208 const struct nxt_set_flow_format *msg
4209 = (const struct nxt_set_flow_format *) oh;
4212 format = ntohl(msg->format);
4213 if (format == NXFF_OPENFLOW10
4214 || format == NXFF_TUN_ID_FROM_COOKIE
4215 || format == NXFF_NXM) {
4216 ofconn->flow_format = format;
4219 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4224 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4226 struct ofp_header *ob;
4229 /* Currently, everything executes synchronously, so we can just
4230 * immediately send the barrier reply. */
4231 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4232 queue_tx(buf, ofconn, ofconn->reply_counter);
4237 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4239 const struct ofp_header *oh = msg->data;
4240 const struct ofputil_msg_type *type;
4243 error = ofputil_decode_msg_type(oh, &type);
4248 switch (ofputil_msg_type_code(type)) {
4249 /* OpenFlow requests. */
4250 case OFPUTIL_OFPT_ECHO_REQUEST:
4251 return handle_echo_request(ofconn, oh);
4253 case OFPUTIL_OFPT_FEATURES_REQUEST:
4254 return handle_features_request(ofconn, oh);
4256 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4257 return handle_get_config_request(ofconn, oh);
4259 case OFPUTIL_OFPT_SET_CONFIG:
4260 return handle_set_config(ofconn, msg->data);
4262 case OFPUTIL_OFPT_PACKET_OUT:
4263 return handle_packet_out(ofconn, oh);
4265 case OFPUTIL_OFPT_PORT_MOD:
4266 return handle_port_mod(ofconn, oh);
4268 case OFPUTIL_OFPT_FLOW_MOD:
4269 return handle_flow_mod(ofconn, oh);
4271 case OFPUTIL_OFPT_BARRIER_REQUEST:
4272 return handle_barrier_request(ofconn, oh);
4274 /* OpenFlow replies. */
4275 case OFPUTIL_OFPT_ECHO_REPLY:
4278 /* Nicira extension requests. */
4279 case OFPUTIL_NXT_STATUS_REQUEST:
4280 return switch_status_handle_request(
4281 ofconn->ofproto->switch_status, ofconn->rconn, oh);
4283 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4284 return handle_tun_id_from_cookie(ofconn, oh);
4286 case OFPUTIL_NXT_ROLE_REQUEST:
4287 return handle_role_request(ofconn, oh);
4289 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4290 return handle_nxt_set_flow_format(ofconn, oh);
4292 case OFPUTIL_NXT_FLOW_MOD:
4293 return handle_flow_mod(ofconn, oh);
4295 /* OpenFlow statistics requests. */
4296 case OFPUTIL_OFPST_DESC_REQUEST:
4297 return handle_desc_stats_request(ofconn, oh);
4299 case OFPUTIL_OFPST_FLOW_REQUEST:
4300 return handle_flow_stats_request(ofconn, oh);
4302 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4303 return handle_aggregate_stats_request(ofconn, oh);
4305 case OFPUTIL_OFPST_TABLE_REQUEST:
4306 return handle_table_stats_request(ofconn, oh);
4308 case OFPUTIL_OFPST_PORT_REQUEST:
4309 return handle_port_stats_request(ofconn, oh);
4311 case OFPUTIL_OFPST_QUEUE_REQUEST:
4312 return handle_queue_stats_request(ofconn, oh);
4314 /* Nicira extension statistics requests. */
4315 case OFPUTIL_NXST_FLOW_REQUEST:
4316 return handle_nxst_flow(ofconn, oh);
4318 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4319 return handle_nxst_aggregate(ofconn, oh);
4321 case OFPUTIL_INVALID:
4322 case OFPUTIL_OFPT_HELLO:
4323 case OFPUTIL_OFPT_ERROR:
4324 case OFPUTIL_OFPT_FEATURES_REPLY:
4325 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4326 case OFPUTIL_OFPT_PACKET_IN:
4327 case OFPUTIL_OFPT_FLOW_REMOVED:
4328 case OFPUTIL_OFPT_PORT_STATUS:
4329 case OFPUTIL_OFPT_BARRIER_REPLY:
4330 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4331 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4332 case OFPUTIL_OFPST_DESC_REPLY:
4333 case OFPUTIL_OFPST_FLOW_REPLY:
4334 case OFPUTIL_OFPST_QUEUE_REPLY:
4335 case OFPUTIL_OFPST_PORT_REPLY:
4336 case OFPUTIL_OFPST_TABLE_REPLY:
4337 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4338 case OFPUTIL_NXT_STATUS_REPLY:
4339 case OFPUTIL_NXT_ROLE_REPLY:
4340 case OFPUTIL_NXT_FLOW_REMOVED:
4341 case OFPUTIL_NXST_FLOW_REPLY:
4342 case OFPUTIL_NXST_AGGREGATE_REPLY:
4344 if (VLOG_IS_WARN_ENABLED()) {
4345 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4346 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4349 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4350 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4352 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4358 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4360 int error = handle_openflow__(ofconn, ofp_msg);
4362 send_error_oh(ofconn, ofp_msg->data, error);
4364 COVERAGE_INC(ofproto_recv_openflow);
4368 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4370 struct facet *facet;
4373 /* Obtain in_port and tun_id, at least. */
4374 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4376 /* Set header pointers in 'flow'. */
4377 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4379 /* Check with in-band control to see if this packet should be sent
4380 * to the local port regardless of the flow table. */
4381 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4382 struct ofpbuf odp_actions;
4384 ofpbuf_init(&odp_actions, 32);
4385 nl_msg_put_u32(&odp_actions, ODPAT_OUTPUT, ODPP_LOCAL);
4386 dpif_execute(p->dpif, odp_actions.data, odp_actions.size,
4388 ofpbuf_uninit(&odp_actions);
4391 facet = facet_lookup_valid(p, &flow);
4393 struct rule *rule = rule_lookup(p, &flow);
4395 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4396 struct ofport *port = get_port(p, flow.in_port);
4398 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4399 COVERAGE_INC(ofproto_no_packet_in);
4400 /* XXX install 'drop' flow entry */
4401 ofpbuf_delete(upcall->packet);
4405 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4409 COVERAGE_INC(ofproto_packet_in);
4410 send_packet_in(p, upcall, &flow, false);
4414 facet = facet_create(p, rule, &flow, upcall->packet);
4415 } else if (!facet->may_install) {
4416 /* The facet is not installable, that is, we need to process every
4417 * packet, so process the current packet's actions into 'facet'. */
4418 facet_make_actions(p, facet, upcall->packet);
4421 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4423 * Extra-special case for fail-open mode.
4425 * We are in fail-open mode and the packet matched the fail-open rule,
4426 * but we are connected to a controller too. We should send the packet
4427 * up to the controller in the hope that it will try to set up a flow
4428 * and thereby allow us to exit fail-open.
4430 * See the top-level comment in fail-open.c for more information.
4432 send_packet_in(p, upcall, &flow, true);
4435 facet_execute(p, facet, upcall->packet);
4436 facet_install(p, facet, false);
4440 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4444 switch (upcall->type) {
4445 case _ODPL_ACTION_NR:
4446 COVERAGE_INC(ofproto_ctlr_action);
4447 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4448 send_packet_in(p, upcall, &flow, false);
4451 case _ODPL_SFLOW_NR:
4453 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4454 ofproto_sflow_received(p->sflow, upcall, &flow);
4456 ofpbuf_delete(upcall->packet);
4460 handle_miss_upcall(p, upcall);
4464 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4469 /* Flow expiration. */
4471 static int ofproto_dp_max_idle(const struct ofproto *);
4472 static void ofproto_update_used(struct ofproto *);
4473 static void rule_expire(struct ofproto *, struct rule *);
4474 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4476 /* This function is called periodically by ofproto_run(). Its job is to
4477 * collect updates for the flows that have been installed into the datapath,
4478 * most importantly when they last were used, and then use that information to
4479 * expire flows that have not been used recently.
4481 * Returns the number of milliseconds after which it should be called again. */
4483 ofproto_expire(struct ofproto *ofproto)
4485 struct rule *rule, *next_rule;
4486 struct cls_cursor cursor;
4489 /* Update 'used' for each flow in the datapath. */
4490 ofproto_update_used(ofproto);
4492 /* Expire facets that have been idle too long. */
4493 dp_max_idle = ofproto_dp_max_idle(ofproto);
4494 ofproto_expire_facets(ofproto, dp_max_idle);
4496 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4497 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4498 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4499 rule_expire(ofproto, rule);
4502 /* Let the hook know that we're at a stable point: all outstanding data
4503 * in existing flows has been accounted to the account_cb. Thus, the
4504 * hook can now reasonably do operations that depend on having accurate
4505 * flow volume accounting (currently, that's just bond rebalancing). */
4506 if (ofproto->ofhooks->account_checkpoint_cb) {
4507 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4510 return MIN(dp_max_idle, 1000);
4513 /* Update 'used' member of installed facets. */
4515 ofproto_update_used(struct ofproto *p)
4517 const struct dpif_flow_stats *stats;
4518 struct dpif_flow_dump dump;
4519 const struct nlattr *key;
4522 dpif_flow_dump_start(&dump, p->dpif);
4523 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4524 struct facet *facet;
4527 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4531 odp_flow_key_format(key, key_len, &s);
4532 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4538 facet = facet_find(p, &flow);
4540 if (facet && facet->installed) {
4541 facet_update_time(p, facet, stats);
4542 facet_account(p, facet, stats->n_bytes);
4544 /* There's a flow in the datapath that we know nothing about.
4546 COVERAGE_INC(ofproto_unexpected_rule);
4547 dpif_flow_del(p->dpif, key, key_len, NULL);
4550 dpif_flow_dump_done(&dump);
4553 /* Calculates and returns the number of milliseconds of idle time after which
4554 * facets should expire from the datapath and we should fold their statistics
4555 * into their parent rules in userspace. */
4557 ofproto_dp_max_idle(const struct ofproto *ofproto)
4560 * Idle time histogram.
4562 * Most of the time a switch has a relatively small number of facets. When
4563 * this is the case we might as well keep statistics for all of them in
4564 * userspace and to cache them in the kernel datapath for performance as
4567 * As the number of facets increases, the memory required to maintain
4568 * statistics about them in userspace and in the kernel becomes
4569 * significant. However, with a large number of facets it is likely that
4570 * only a few of them are "heavy hitters" that consume a large amount of
4571 * bandwidth. At this point, only heavy hitters are worth caching in the
4572 * kernel and maintaining in userspaces; other facets we can discard.
4574 * The technique used to compute the idle time is to build a histogram with
4575 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4576 * that is installed in the kernel gets dropped in the appropriate bucket.
4577 * After the histogram has been built, we compute the cutoff so that only
4578 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4579 * cached. At least the most-recently-used bucket of facets is kept, so
4580 * actually an arbitrary number of facets can be kept in any given
4581 * expiration run (though the next run will delete most of those unless
4582 * they receive additional data).
4584 * This requires a second pass through the facets, in addition to the pass
4585 * made by ofproto_update_used(), because the former function never looks
4586 * at uninstallable facets.
4588 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4589 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4590 int buckets[N_BUCKETS] = { 0 };
4591 struct facet *facet;
4596 total = hmap_count(&ofproto->facets);
4597 if (total <= 1000) {
4598 return N_BUCKETS * BUCKET_WIDTH;
4601 /* Build histogram. */
4603 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4604 long long int idle = now - facet->used;
4605 int bucket = (idle <= 0 ? 0
4606 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4607 : (unsigned int) idle / BUCKET_WIDTH);
4611 /* Find the first bucket whose flows should be expired. */
4612 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4613 if (buckets[bucket]) {
4616 subtotal += buckets[bucket++];
4617 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4622 if (VLOG_IS_DBG_ENABLED()) {
4626 ds_put_cstr(&s, "keep");
4627 for (i = 0; i < N_BUCKETS; i++) {
4629 ds_put_cstr(&s, ", drop");
4632 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4635 VLOG_INFO("%s: %s (msec:count)",
4636 dpif_name(ofproto->dpif), ds_cstr(&s));
4640 return bucket * BUCKET_WIDTH;
4644 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4646 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4647 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4648 struct ofexpired expired;
4650 expired.flow = facet->flow;
4651 expired.packet_count = facet->packet_count;
4652 expired.byte_count = facet->byte_count;
4653 expired.used = facet->used;
4655 /* Get updated flow stats.
4657 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4658 * updated TCP flags and (2) the dpif_flow_list_all() in
4659 * ofproto_update_used() zeroed TCP flags. */
4660 if (facet->installed) {
4661 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
4662 struct dpif_flow_stats stats;
4665 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
4666 odp_flow_key_from_flow(&key, &facet->flow);
4668 if (!dpif_flow_get(ofproto->dpif, ODPFF_ZERO_TCP_FLAGS,
4669 key.data, key.size, NULL, &stats)) {
4670 expired.packet_count += stats.n_packets;
4671 expired.byte_count += stats.n_bytes;
4672 if (stats.n_packets) {
4673 facet_update_time(ofproto, facet, &stats);
4674 netflow_flow_update_flags(&facet->nf_flow,
4680 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4685 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4687 long long int cutoff = time_msec() - dp_max_idle;
4688 struct facet *facet, *next_facet;
4690 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4691 facet_active_timeout(ofproto, facet);
4692 if (facet->used < cutoff) {
4693 facet_remove(ofproto, facet);
4698 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4699 * then delete it entirely. */
4701 rule_expire(struct ofproto *ofproto, struct rule *rule)
4703 struct facet *facet, *next_facet;
4707 /* Has 'rule' expired? */
4709 if (rule->hard_timeout
4710 && now > rule->created + rule->hard_timeout * 1000) {
4711 reason = OFPRR_HARD_TIMEOUT;
4712 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4713 && now >rule->used + rule->idle_timeout * 1000) {
4714 reason = OFPRR_IDLE_TIMEOUT;
4719 COVERAGE_INC(ofproto_expired);
4721 /* Update stats. (This is a no-op if the rule expired due to an idle
4722 * timeout, because that only happens when the rule has no facets left.) */
4723 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4724 facet_remove(ofproto, facet);
4727 /* Get rid of the rule. */
4728 if (!rule_is_hidden(rule)) {
4729 rule_send_removed(ofproto, rule, reason);
4731 rule_remove(ofproto, rule);
4734 static struct ofpbuf *
4735 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4738 struct ofp_flow_removed *ofr;
4741 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4742 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4743 rule->flow_cookie, &ofr->cookie);
4744 ofr->priority = htons(rule->cr.priority);
4745 ofr->reason = reason;
4746 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4747 ofr->idle_timeout = htons(rule->idle_timeout);
4748 ofr->packet_count = htonll(rule->packet_count);
4749 ofr->byte_count = htonll(rule->byte_count);
4754 static struct ofpbuf *
4755 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4757 struct nx_flow_removed *nfr;
4761 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4762 match_len = nx_put_match(buf, &rule->cr);
4765 nfr->cookie = rule->flow_cookie;
4766 nfr->priority = htons(rule->cr.priority);
4767 nfr->reason = reason;
4768 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4769 nfr->idle_timeout = htons(rule->idle_timeout);
4770 nfr->match_len = htons(match_len);
4771 nfr->packet_count = htonll(rule->packet_count);
4772 nfr->byte_count = htonll(rule->byte_count);
4778 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4780 struct ofconn *ofconn;
4782 if (!rule->send_flow_removed) {
4786 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4789 if (!rconn_is_connected(ofconn->rconn)
4790 || !ofconn_receives_async_msgs(ofconn)) {
4794 msg = (ofconn->flow_format == NXFF_NXM
4795 ? compose_nx_flow_removed(rule, reason)
4796 : compose_ofp_flow_removed(ofconn, rule, reason));
4798 /* Account flow expirations under ofconn->reply_counter, the counter
4799 * for replies to OpenFlow requests. That works because preventing
4800 * OpenFlow requests from being processed also prevents new flows from
4801 * being added (and expiring). (It also prevents processing OpenFlow
4802 * requests that would not add new flows, so it is imperfect.) */
4803 queue_tx(msg, ofconn, ofconn->reply_counter);
4807 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4809 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4811 struct ofconn *ofconn = ofconn_;
4813 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4814 ofconn->packet_in_counter, 100);
4817 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4818 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4819 * scheduler for sending.
4821 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4822 * Otherwise, ownership is transferred to this function. */
4824 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4825 const struct flow *flow, bool clone)
4827 enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
4828 struct ofproto *ofproto = ofconn->ofproto;
4829 struct ofp_packet_in *opi;
4830 int total_len, send_len;
4831 struct ofpbuf *packet;
4834 /* Get OpenFlow buffer_id. */
4835 if (upcall->type == _ODPL_ACTION_NR) {
4836 buffer_id = UINT32_MAX;
4837 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4838 buffer_id = pktbuf_get_null();
4839 } else if (!ofconn->pktbuf) {
4840 buffer_id = UINT32_MAX;
4842 buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
4845 /* Figure out how much of the packet to send. */
4846 total_len = send_len = upcall->packet->size;
4847 if (buffer_id != UINT32_MAX) {
4848 send_len = MIN(send_len, ofconn->miss_send_len);
4850 if (upcall->type == _ODPL_ACTION_NR) {
4851 send_len = MIN(send_len, upcall->userdata);
4854 /* Copy or steal buffer for OFPT_PACKET_IN. */
4856 packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
4857 send_len, OPI_SIZE);
4859 packet = upcall->packet;
4860 packet->size = send_len;
4863 /* Add OFPT_PACKET_IN. */
4864 opi = ofpbuf_push_zeros(packet, OPI_SIZE);
4865 opi->header.version = OFP_VERSION;
4866 opi->header.type = OFPT_PACKET_IN;
4867 opi->total_len = htons(total_len);
4868 opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
4869 opi->reason = upcall->type == _ODPL_MISS_NR ? OFPR_NO_MATCH : OFPR_ACTION;
4870 opi->buffer_id = htonl(buffer_id);
4871 update_openflow_length(packet);
4873 /* Hand over to packet scheduler. It might immediately call into
4874 * do_send_packet_in() or it might buffer it for a while (until a later
4875 * call to pinsched_run()). */
4876 pinsched_send(ofconn->schedulers[opi->reason], flow->in_port,
4877 packet, do_send_packet_in, ofconn);
4880 /* Given 'upcall', of type _ODPL_ACTION_NR or _ODPL_MISS_NR, sends an
4881 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4882 * their individual configurations.
4884 * Takes ownership of 'packet'. */
4886 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4887 const struct flow *flow, bool clone)
4889 struct ofconn *ofconn, *prev;
4892 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4893 if (ofconn_receives_async_msgs(ofconn)) {
4895 schedule_packet_in(prev, upcall, flow, true);
4901 schedule_packet_in(prev, upcall, flow, clone);
4902 } else if (!clone) {
4903 ofpbuf_delete(upcall->packet);
4908 pick_datapath_id(const struct ofproto *ofproto)
4910 const struct ofport *port;
4912 port = get_port(ofproto, ODPP_LOCAL);
4914 uint8_t ea[ETH_ADDR_LEN];
4917 error = netdev_get_etheraddr(port->netdev, ea);
4919 return eth_addr_to_uint64(ea);
4921 VLOG_WARN("could not get MAC address for %s (%s)",
4922 netdev_get_name(port->netdev), strerror(error));
4924 return ofproto->fallback_dpid;
4928 pick_fallback_dpid(void)
4930 uint8_t ea[ETH_ADDR_LEN];
4931 eth_addr_nicira_random(ea);
4932 return eth_addr_to_uint64(ea);
4936 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4937 void *aux OVS_UNUSED)
4939 const struct shash_node *node;
4943 SHASH_FOR_EACH (node, &all_ofprotos) {
4944 ds_put_format(&results, "%s\n", node->name);
4946 unixctl_command_reply(conn, 200, ds_cstr(&results));
4947 ds_destroy(&results);
4950 struct ofproto_trace {
4951 struct action_xlate_ctx ctx;
4957 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4959 ds_put_char_multiple(result, '\t', level);
4961 ds_put_cstr(result, "No match\n");
4965 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4966 ntohll(rule->flow_cookie));
4967 cls_rule_format(&rule->cr, result);
4968 ds_put_char(result, '\n');
4970 ds_put_char_multiple(result, '\t', level);
4971 ds_put_cstr(result, "OpenFlow ");
4972 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4973 rule->n_actions * sizeof *rule->actions);
4974 ds_put_char(result, '\n');
4978 trace_format_flow(struct ds *result, int level, const char *title,
4979 struct ofproto_trace *trace)
4981 ds_put_char_multiple(result, '\t', level);
4982 ds_put_format(result, "%s: ", title);
4983 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4984 ds_put_cstr(result, "unchanged");
4986 flow_format(result, &trace->ctx.flow);
4987 trace->flow = trace->ctx.flow;
4989 ds_put_char(result, '\n');
4993 trace_resubmit(struct action_xlate_ctx *ctx, const struct rule *rule)
4995 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4996 struct ds *result = trace->result;
4998 ds_put_char(result, '\n');
4999 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5000 trace_format_rule(result, ctx->recurse + 1, rule);
5004 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5005 void *aux OVS_UNUSED)
5007 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5008 char *args = xstrdup(args_);
5009 char *save_ptr = NULL;
5010 struct ofproto *ofproto;
5011 struct ofpbuf packet;
5019 ofpbuf_init(&packet, strlen(args) / 2);
5022 dpname = strtok_r(args, " ", &save_ptr);
5023 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5024 in_port_s = strtok_r(NULL, " ", &save_ptr);
5025 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5026 if (!dpname || !in_port_s || !packet_s) {
5027 unixctl_command_reply(conn, 501, "Bad command syntax");
5031 ofproto = shash_find_data(&all_ofprotos, dpname);
5033 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5038 tun_id = htonll(strtoull(tun_id_s, NULL, 10));
5039 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5041 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5042 packet_s += strspn(packet_s, " ");
5043 if (*packet_s != '\0') {
5044 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5047 if (packet.size < ETH_HEADER_LEN) {
5048 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5052 ds_put_cstr(&result, "Packet: ");
5053 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5054 ds_put_cstr(&result, s);
5057 flow_extract(&packet, tun_id, in_port, &flow);
5058 ds_put_cstr(&result, "Flow: ");
5059 flow_format(&result, &flow);
5060 ds_put_char(&result, '\n');
5062 rule = rule_lookup(ofproto, &flow);
5063 trace_format_rule(&result, 0, rule);
5065 struct ofproto_trace trace;
5066 struct ofpbuf *odp_actions;
5068 trace.result = &result;
5070 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5071 trace.ctx.resubmit_hook = trace_resubmit;
5072 odp_actions = xlate_actions(&trace.ctx,
5073 rule->actions, rule->n_actions);
5075 ds_put_char(&result, '\n');
5076 trace_format_flow(&result, 0, "Final flow", &trace);
5077 ds_put_cstr(&result, "Datapath actions: ");
5078 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5079 ofpbuf_delete(odp_actions);
5082 unixctl_command_reply(conn, 200, ds_cstr(&result));
5085 ds_destroy(&result);
5086 ofpbuf_uninit(&packet);
5091 ofproto_unixctl_init(void)
5093 static bool registered;
5099 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5100 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5104 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5105 struct ofpbuf *odp_actions, tag_type *tags,
5106 uint16_t *nf_output_iface, void *ofproto_)
5108 struct ofproto *ofproto = ofproto_;
5111 /* Drop frames for reserved multicast addresses. */
5112 if (eth_addr_is_reserved(flow->dl_dst)) {
5116 /* Learn source MAC (but don't try to learn from revalidation). */
5117 if (packet != NULL) {
5118 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5120 GRAT_ARP_LOCK_NONE);
5122 /* The log messages here could actually be useful in debugging,
5123 * so keep the rate limit relatively high. */
5124 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5125 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5126 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5127 ofproto_revalidate(ofproto, rev_tag);
5131 /* Determine output port. */
5132 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5135 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5136 nf_output_iface, odp_actions);
5137 } else if (out_port != flow->in_port) {
5138 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, out_port);
5139 *nf_output_iface = out_port;
5147 static const struct ofhooks default_ofhooks = {
5148 default_normal_ofhook_cb,