2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
34 #include "mac-learning.h"
38 #include "ofp-print.h"
40 #include "ofproto-sflow.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
44 #include "openvswitch/datapath-protocol.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
54 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto)
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER = 1
73 struct netdev *netdev;
74 struct ofp_phy_port opp; /* In host byte order. */
77 static void ofport_free(struct ofport *);
78 static void hton_ofp_phy_port(struct ofp_phy_port *);
80 static int xlate_actions(const union ofp_action *in, size_t n_in,
81 const flow_t *flow, struct ofproto *ofproto,
82 const struct ofpbuf *packet,
83 struct odp_actions *out, tag_type *tags,
84 bool *may_set_up_flow, uint16_t *nf_output_iface);
89 uint64_t flow_cookie; /* Controller-issued identifier.
90 (Kept in network-byte order.) */
91 uint16_t idle_timeout; /* In seconds from time of last use. */
92 uint16_t hard_timeout; /* In seconds from time of creation. */
93 bool send_flow_removed; /* Send a flow removed message? */
94 long long int used; /* Last-used time (0 if never used). */
95 long long int created; /* Creation time. */
96 uint64_t packet_count; /* Number of packets received. */
97 uint64_t byte_count; /* Number of bytes received. */
98 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
99 tag_type tags; /* Tags (set only by hooks). */
100 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
102 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
103 * exact-match rule (having cr.wc.wildcards of 0) generated from the
104 * wildcard rule 'super'. In this case, 'list' is an element of the
107 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
108 * a list of subrules. A super-rule with no wildcards (where
109 * cr.wc.wildcards is 0) will never have any subrules. */
115 * 'n_actions' is the number of elements in the 'actions' array. A single
116 * action may take up more more than one element's worth of space.
118 * A subrule has no actions (it uses the super-rule's actions). */
120 union ofp_action *actions;
124 * A super-rule with wildcard fields never has ODP actions (since the
125 * datapath only supports exact-match flows). */
126 bool installed; /* Installed in datapath? */
127 bool may_install; /* True ordinarily; false if actions must
128 * be reassessed for every packet. */
130 union odp_action *odp_actions;
134 rule_is_hidden(const struct rule *rule)
136 /* Subrules are merely an implementation detail, so hide them from the
138 if (rule->super != NULL) {
142 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
143 * (e.g. by in-band control) and are intentionally hidden from the
145 if (rule->cr.priority > UINT16_MAX) {
152 static struct rule *rule_create(struct ofproto *, struct rule *super,
153 const union ofp_action *, size_t n_actions,
154 uint16_t idle_timeout, uint16_t hard_timeout,
155 uint64_t flow_cookie, bool send_flow_removed);
156 static void rule_free(struct rule *);
157 static void rule_destroy(struct ofproto *, struct rule *);
158 static struct rule *rule_from_cls_rule(const struct cls_rule *);
159 static void rule_insert(struct ofproto *, struct rule *,
160 struct ofpbuf *packet, uint16_t in_port);
161 static void rule_remove(struct ofproto *, struct rule *);
162 static bool rule_make_actions(struct ofproto *, struct rule *,
163 const struct ofpbuf *packet);
164 static void rule_install(struct ofproto *, struct rule *,
165 struct rule *displaced_rule);
166 static void rule_uninstall(struct ofproto *, struct rule *);
167 static void rule_post_uninstall(struct ofproto *, struct rule *);
168 static void send_flow_removed(struct ofproto *p, struct rule *rule,
169 long long int now, uint8_t reason);
171 /* ofproto supports two kinds of OpenFlow connections:
173 * - "Controller connections": Connections to ordinary OpenFlow controllers.
174 * ofproto maintains persistent connections to these controllers and by
175 * default sends them asynchronous messages such as packet-ins.
177 * - "Transient connections", e.g. from ovs-ofctl. When these connections
178 * drop, it is the other side's responsibility to reconnect them if
179 * necessary. ofproto does not send them asynchronous messages by default.
182 OFCONN_CONTROLLER, /* An OpenFlow controller. */
183 OFCONN_TRANSIENT /* A transient connection. */
186 /* An OpenFlow connection. */
188 struct ofproto *ofproto; /* The ofproto that owns this connection. */
189 struct list node; /* In struct ofproto's "all_conns" list. */
190 struct rconn *rconn; /* OpenFlow connection. */
191 enum ofconn_type type; /* Type. */
193 /* OFPT_PACKET_IN related data. */
194 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
195 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
196 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
197 int miss_send_len; /* Bytes to send of buffered packets. */
199 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
200 * requests, and the maximum number before we stop reading OpenFlow
202 #define OFCONN_REPLY_MAX 100
203 struct rconn_packet_counter *reply_counter;
205 /* type == OFCONN_CONTROLLER only. */
206 enum nx_role role; /* Role. */
207 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
208 struct discovery *discovery; /* Controller discovery object, if enabled. */
209 struct status_category *ss; /* Switch status category. */
210 enum ofproto_band band; /* In-band or out-of-band? */
213 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
214 * "schedulers" array. Their values are 0 and 1, and their meanings and values
215 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
216 * case anything ever changes, check their values here. */
217 #define N_SCHEDULERS 2
218 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
219 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
220 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
221 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
223 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
225 static void ofconn_destroy(struct ofconn *);
226 static void ofconn_run(struct ofconn *, struct ofproto *);
227 static void ofconn_wait(struct ofconn *);
228 static bool ofconn_receives_async_msgs(const struct ofconn *);
229 static char *ofconn_make_name(const struct ofproto *, const char *target);
231 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
232 struct rconn_packet_counter *counter);
234 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
235 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
239 uint64_t datapath_id; /* Datapath ID. */
240 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
241 char *mfr_desc; /* Manufacturer. */
242 char *hw_desc; /* Hardware. */
243 char *sw_desc; /* Software version. */
244 char *serial_desc; /* Serial number. */
245 char *dp_desc; /* Datapath description. */
249 struct netdev_monitor *netdev_monitor;
250 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
252 struct shash port_by_name;
256 struct switch_status *switch_status;
257 struct fail_open *fail_open;
258 struct netflow *netflow;
259 struct ofproto_sflow *sflow;
261 /* In-band control. */
262 struct in_band *in_band;
263 long long int next_in_band_update;
264 struct sockaddr_in *extra_in_band_remotes;
265 size_t n_extra_remotes;
268 struct classifier cls;
269 bool need_revalidate;
270 long long int next_expiration;
271 struct tag_set revalidate_set;
272 bool tun_id_from_cookie;
274 /* OpenFlow connections. */
275 struct hmap controllers; /* Controller "struct ofconn"s. */
276 struct list all_conns; /* Contains "struct ofconn"s. */
277 enum ofproto_fail_mode fail_mode;
278 struct pvconn **listeners;
280 struct pvconn **snoops;
283 /* Hooks for ovs-vswitchd. */
284 const struct ofhooks *ofhooks;
287 /* Used by default ofhooks. */
288 struct mac_learning *ml;
291 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
293 static const struct ofhooks default_ofhooks;
295 static uint64_t pick_datapath_id(const struct ofproto *);
296 static uint64_t pick_fallback_dpid(void);
298 static void update_used(struct ofproto *);
299 static void update_stats(struct ofproto *, struct rule *,
300 const struct odp_flow_stats *);
301 static void expire_rule(struct cls_rule *, void *ofproto);
302 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
303 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
304 static void revalidate_cb(struct cls_rule *rule_, void *p_);
306 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
308 static void handle_openflow(struct ofconn *, struct ofproto *,
311 static void refresh_port_groups(struct ofproto *);
313 static void update_port(struct ofproto *, const char *devname);
314 static int init_ports(struct ofproto *);
315 static void reinit_ports(struct ofproto *);
318 ofproto_create(const char *datapath, const char *datapath_type,
319 const struct ofhooks *ofhooks, void *aux,
320 struct ofproto **ofprotop)
322 struct odp_stats stats;
329 /* Connect to datapath and start listening for messages. */
330 error = dpif_open(datapath, datapath_type, &dpif);
332 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
335 error = dpif_get_dp_stats(dpif, &stats);
337 VLOG_ERR("failed to obtain stats for datapath %s: %s",
338 datapath, strerror(error));
342 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
344 VLOG_ERR("failed to listen on datapath %s: %s",
345 datapath, strerror(error));
349 dpif_flow_flush(dpif);
350 dpif_recv_purge(dpif);
352 /* Initialize settings. */
353 p = xzalloc(sizeof *p);
354 p->fallback_dpid = pick_fallback_dpid();
355 p->datapath_id = p->fallback_dpid;
356 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
357 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
358 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
359 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
360 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
362 /* Initialize datapath. */
364 p->netdev_monitor = netdev_monitor_create();
365 port_array_init(&p->ports);
366 shash_init(&p->port_by_name);
367 p->max_ports = stats.max_ports;
369 /* Initialize submodules. */
370 p->switch_status = switch_status_create(p);
376 /* Initialize flow table. */
377 classifier_init(&p->cls);
378 p->need_revalidate = false;
379 p->next_expiration = time_msec() + 1000;
380 tag_set_init(&p->revalidate_set);
382 /* Initialize OpenFlow connections. */
383 list_init(&p->all_conns);
384 hmap_init(&p->controllers);
390 /* Initialize hooks. */
392 p->ofhooks = ofhooks;
396 p->ofhooks = &default_ofhooks;
398 p->ml = mac_learning_create();
401 /* Pick final datapath ID. */
402 p->datapath_id = pick_datapath_id(p);
403 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
410 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
412 uint64_t old_dpid = p->datapath_id;
413 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
414 if (p->datapath_id != old_dpid) {
415 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
417 /* Force all active connections to reconnect, since there is no way to
418 * notify a controller that the datapath ID has changed. */
419 ofproto_reconnect_controllers(p);
424 is_discovery_controller(const struct ofproto_controller *c)
426 return !strcmp(c->target, "discover");
430 is_in_band_controller(const struct ofproto_controller *c)
432 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
435 /* Creates a new controller in 'ofproto'. Some of the settings are initially
436 * drawn from 'c', but update_controller() needs to be called later to finish
437 * the new ofconn's configuration. */
439 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
441 struct discovery *discovery;
442 struct ofconn *ofconn;
444 if (is_discovery_controller(c)) {
445 int error = discovery_create(c->accept_re, c->update_resolv_conf,
446 ofproto->dpif, ofproto->switch_status,
455 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
456 ofconn->pktbuf = pktbuf_create();
457 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
459 ofconn->discovery = discovery;
461 char *name = ofconn_make_name(ofproto, c->target);
462 rconn_connect(ofconn->rconn, c->target, name);
465 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
466 hash_string(c->target, 0));
469 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
470 * target or turn discovery on or off (these are done by creating new ofconns
471 * and deleting old ones), but it can update the rest of an ofconn's
474 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
476 struct ofproto *ofproto = ofconn->ofproto;
480 ofconn->band = (is_in_band_controller(c)
481 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
483 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
485 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
486 rconn_set_probe_interval(ofconn->rconn, probe_interval);
488 if (ofconn->discovery) {
489 discovery_set_update_resolv_conf(ofconn->discovery,
490 c->update_resolv_conf);
491 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
494 for (i = 0; i < N_SCHEDULERS; i++) {
495 struct pinsched **s = &ofconn->schedulers[i];
497 if (c->rate_limit > 0) {
499 *s = pinsched_create(c->rate_limit, c->burst_limit,
500 ofproto->switch_status);
502 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
505 pinsched_destroy(*s);
512 ofconn_get_target(const struct ofconn *ofconn)
514 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
517 static struct ofconn *
518 find_controller_by_target(struct ofproto *ofproto, const char *target)
520 struct ofconn *ofconn;
522 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
523 hash_string(target, 0), &ofproto->controllers) {
524 if (!strcmp(ofconn_get_target(ofconn), target)) {
532 update_in_band_remotes(struct ofproto *ofproto)
534 const struct ofconn *ofconn;
535 struct sockaddr_in *addrs;
536 size_t max_addrs, n_addrs;
540 /* Allocate enough memory for as many remotes as we could possibly have. */
541 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
542 addrs = xmalloc(max_addrs * sizeof *addrs);
545 /* Add all the remotes. */
547 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
548 struct sockaddr_in *sin = &addrs[n_addrs];
550 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
554 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
555 if (sin->sin_addr.s_addr) {
556 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
559 if (ofconn->discovery) {
563 for (i = 0; i < ofproto->n_extra_remotes; i++) {
564 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
567 /* Create or update or destroy in-band.
569 * Ordinarily we only enable in-band if there's at least one remote
570 * address, but discovery needs the in-band rules for DHCP to be installed
571 * even before we know any remote addresses. */
572 if (n_addrs || discovery) {
573 if (!ofproto->in_band) {
574 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
577 if (ofproto->in_band) {
578 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
580 ofproto->next_in_band_update = time_msec() + 1000;
582 in_band_destroy(ofproto->in_band);
583 ofproto->in_band = NULL;
591 update_fail_open(struct ofproto *p)
593 struct ofconn *ofconn;
595 if (!hmap_is_empty(&p->controllers)
596 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
597 struct rconn **rconns;
601 p->fail_open = fail_open_create(p, p->switch_status);
605 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
606 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
607 rconns[n++] = ofconn->rconn;
610 fail_open_set_controllers(p->fail_open, rconns, n);
611 /* p->fail_open takes ownership of 'rconns'. */
613 fail_open_destroy(p->fail_open);
619 ofproto_set_controllers(struct ofproto *p,
620 const struct ofproto_controller *controllers,
621 size_t n_controllers)
623 struct shash new_controllers;
624 struct ofconn *ofconn, *next;
628 shash_init(&new_controllers);
629 for (i = 0; i < n_controllers; i++) {
630 const struct ofproto_controller *c = &controllers[i];
632 shash_add_once(&new_controllers, c->target, &controllers[i]);
633 if (!find_controller_by_target(p, c->target)) {
634 add_controller(p, c);
639 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
641 struct ofproto_controller *c;
643 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
645 ofconn_destroy(ofconn);
647 update_controller(ofconn, c);
653 shash_destroy(&new_controllers);
655 update_in_band_remotes(p);
659 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
660 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
661 struct ofconn, hmap_node);
662 ofconn->ss = switch_status_register(p->switch_status, "remote",
663 rconn_status_cb, ofconn->rconn);
668 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
670 p->fail_mode = fail_mode;
674 /* Drops the connections between 'ofproto' and all of its controllers, forcing
675 * them to reconnect. */
677 ofproto_reconnect_controllers(struct ofproto *ofproto)
679 struct ofconn *ofconn;
681 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
682 rconn_reconnect(ofconn->rconn);
687 any_extras_changed(const struct ofproto *ofproto,
688 const struct sockaddr_in *extras, size_t n)
692 if (n != ofproto->n_extra_remotes) {
696 for (i = 0; i < n; i++) {
697 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
698 const struct sockaddr_in *new = &extras[i];
700 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
701 old->sin_port != new->sin_port) {
709 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
710 * in-band control should guarantee access, in the same way that in-band
711 * control guarantees access to OpenFlow controllers. */
713 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
714 const struct sockaddr_in *extras, size_t n)
716 if (!any_extras_changed(ofproto, extras, n)) {
720 free(ofproto->extra_in_band_remotes);
721 ofproto->n_extra_remotes = n;
722 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
724 update_in_band_remotes(ofproto);
728 ofproto_set_desc(struct ofproto *p,
729 const char *mfr_desc, const char *hw_desc,
730 const char *sw_desc, const char *serial_desc,
733 struct ofp_desc_stats *ods;
736 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
737 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
738 sizeof ods->mfr_desc);
741 p->mfr_desc = xstrdup(mfr_desc);
744 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
745 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
746 sizeof ods->hw_desc);
749 p->hw_desc = xstrdup(hw_desc);
752 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
753 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
754 sizeof ods->sw_desc);
757 p->sw_desc = xstrdup(sw_desc);
760 if (strlen(serial_desc) >= sizeof ods->serial_num) {
761 VLOG_WARN("truncating serial_desc, must be less than %zu "
763 sizeof ods->serial_num);
765 free(p->serial_desc);
766 p->serial_desc = xstrdup(serial_desc);
769 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
770 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
771 sizeof ods->dp_desc);
774 p->dp_desc = xstrdup(dp_desc);
779 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
780 const struct svec *svec)
782 struct pvconn **pvconns = *pvconnsp;
783 size_t n_pvconns = *n_pvconnsp;
787 for (i = 0; i < n_pvconns; i++) {
788 pvconn_close(pvconns[i]);
792 pvconns = xmalloc(svec->n * sizeof *pvconns);
794 for (i = 0; i < svec->n; i++) {
795 const char *name = svec->names[i];
796 struct pvconn *pvconn;
799 error = pvconn_open(name, &pvconn);
801 pvconns[n_pvconns++] = pvconn;
803 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
811 *n_pvconnsp = n_pvconns;
817 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
819 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
823 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
825 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
829 ofproto_set_netflow(struct ofproto *ofproto,
830 const struct netflow_options *nf_options)
832 if (nf_options && nf_options->collectors.n) {
833 if (!ofproto->netflow) {
834 ofproto->netflow = netflow_create();
836 return netflow_set_options(ofproto->netflow, nf_options);
838 netflow_destroy(ofproto->netflow);
839 ofproto->netflow = NULL;
845 ofproto_set_sflow(struct ofproto *ofproto,
846 const struct ofproto_sflow_options *oso)
848 struct ofproto_sflow *os = ofproto->sflow;
851 struct ofport *ofport;
852 unsigned int odp_port;
854 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
855 refresh_port_groups(ofproto);
856 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
857 ofproto_sflow_add_port(os, odp_port,
858 netdev_get_name(ofport->netdev));
861 ofproto_sflow_set_options(os, oso);
863 ofproto_sflow_destroy(os);
864 ofproto->sflow = NULL;
869 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
873 VLOG_WARN("STP is not yet implemented");
881 ofproto_get_datapath_id(const struct ofproto *ofproto)
883 return ofproto->datapath_id;
887 ofproto_has_controller(const struct ofproto *ofproto)
889 return !hmap_is_empty(&ofproto->controllers);
893 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
897 for (i = 0; i < ofproto->n_listeners; i++) {
898 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
903 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
907 for (i = 0; i < ofproto->n_snoops; i++) {
908 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
913 ofproto_destroy(struct ofproto *p)
915 struct ofconn *ofconn, *next_ofconn;
916 struct ofport *ofport;
917 unsigned int port_no;
924 /* Destroy fail-open and in-band early, since they touch the classifier. */
925 fail_open_destroy(p->fail_open);
928 in_band_destroy(p->in_band);
930 free(p->extra_in_band_remotes);
932 ofproto_flush_flows(p);
933 classifier_destroy(&p->cls);
935 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
937 ofconn_destroy(ofconn);
939 hmap_destroy(&p->controllers);
942 netdev_monitor_destroy(p->netdev_monitor);
943 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
946 shash_destroy(&p->port_by_name);
948 switch_status_destroy(p->switch_status);
949 netflow_destroy(p->netflow);
950 ofproto_sflow_destroy(p->sflow);
952 for (i = 0; i < p->n_listeners; i++) {
953 pvconn_close(p->listeners[i]);
957 for (i = 0; i < p->n_snoops; i++) {
958 pvconn_close(p->snoops[i]);
962 mac_learning_destroy(p->ml);
967 free(p->serial_desc);
970 port_array_destroy(&p->ports);
976 ofproto_run(struct ofproto *p)
978 int error = ofproto_run1(p);
980 error = ofproto_run2(p, false);
986 process_port_change(struct ofproto *ofproto, int error, char *devname)
988 if (error == ENOBUFS) {
989 reinit_ports(ofproto);
991 update_port(ofproto, devname);
996 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
997 * means that 'ofconn' is more interesting for monitoring than a lower return
1000 snoop_preference(const struct ofconn *ofconn)
1002 switch (ofconn->role) {
1003 case NX_ROLE_MASTER:
1010 /* Shouldn't happen. */
1015 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1016 * Connects this vconn to a controller. */
1018 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1020 struct ofconn *ofconn, *best;
1022 /* Pick a controller for monitoring. */
1024 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1025 if (ofconn->type == OFCONN_CONTROLLER
1026 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1032 rconn_add_monitor(best->rconn, vconn);
1034 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1040 ofproto_run1(struct ofproto *p)
1042 struct ofconn *ofconn, *next_ofconn;
1047 if (shash_is_empty(&p->port_by_name)) {
1051 for (i = 0; i < 50; i++) {
1055 error = dpif_recv(p->dpif, &buf);
1057 if (error == ENODEV) {
1058 /* Someone destroyed the datapath behind our back. The caller
1059 * better destroy us and give up, because we're just going to
1060 * spin from here on out. */
1061 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1062 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1063 dpif_name(p->dpif));
1069 handle_odp_msg(p, buf);
1072 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1073 process_port_change(p, error, devname);
1075 while ((error = netdev_monitor_poll(p->netdev_monitor,
1076 &devname)) != EAGAIN) {
1077 process_port_change(p, error, devname);
1081 if (time_msec() >= p->next_in_band_update) {
1082 update_in_band_remotes(p);
1084 in_band_run(p->in_band);
1087 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1089 ofconn_run(ofconn, p);
1092 /* Fail-open maintenance. Do this after processing the ofconns since
1093 * fail-open checks the status of the controller rconn. */
1095 fail_open_run(p->fail_open);
1098 for (i = 0; i < p->n_listeners; i++) {
1099 struct vconn *vconn;
1102 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1104 struct rconn *rconn;
1107 rconn = rconn_create(60, 0);
1108 name = ofconn_make_name(p, vconn_get_name(vconn));
1109 rconn_connect_unreliably(rconn, vconn, name);
1112 ofconn_create(p, rconn, OFCONN_TRANSIENT);
1113 } else if (retval != EAGAIN) {
1114 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1118 for (i = 0; i < p->n_snoops; i++) {
1119 struct vconn *vconn;
1122 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1124 add_snooper(p, vconn);
1125 } else if (retval != EAGAIN) {
1126 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1130 if (time_msec() >= p->next_expiration) {
1131 COVERAGE_INC(ofproto_expiration);
1132 p->next_expiration = time_msec() + 1000;
1135 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1137 /* Let the hook know that we're at a stable point: all outstanding data
1138 * in existing flows has been accounted to the account_cb. Thus, the
1139 * hook can now reasonably do operations that depend on having accurate
1140 * flow volume accounting (currently, that's just bond rebalancing). */
1141 if (p->ofhooks->account_checkpoint_cb) {
1142 p->ofhooks->account_checkpoint_cb(p->aux);
1147 netflow_run(p->netflow);
1150 ofproto_sflow_run(p->sflow);
1156 struct revalidate_cbdata {
1157 struct ofproto *ofproto;
1158 bool revalidate_all; /* Revalidate all exact-match rules? */
1159 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1160 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1164 ofproto_run2(struct ofproto *p, bool revalidate_all)
1166 if (p->need_revalidate || revalidate_all
1167 || !tag_set_is_empty(&p->revalidate_set)) {
1168 struct revalidate_cbdata cbdata;
1170 cbdata.revalidate_all = revalidate_all;
1171 cbdata.revalidate_subrules = p->need_revalidate;
1172 cbdata.revalidate_set = p->revalidate_set;
1173 tag_set_init(&p->revalidate_set);
1174 COVERAGE_INC(ofproto_revalidate);
1175 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1176 p->need_revalidate = false;
1183 ofproto_wait(struct ofproto *p)
1185 struct ofconn *ofconn;
1188 dpif_recv_wait(p->dpif);
1189 dpif_port_poll_wait(p->dpif);
1190 netdev_monitor_poll_wait(p->netdev_monitor);
1191 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1192 ofconn_wait(ofconn);
1195 poll_timer_wait_until(p->next_in_band_update);
1196 in_band_wait(p->in_band);
1199 fail_open_wait(p->fail_open);
1202 ofproto_sflow_wait(p->sflow);
1204 if (!tag_set_is_empty(&p->revalidate_set)) {
1205 poll_immediate_wake();
1207 if (p->need_revalidate) {
1208 /* Shouldn't happen, but if it does just go around again. */
1209 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1210 poll_immediate_wake();
1211 } else if (p->next_expiration != LLONG_MAX) {
1212 poll_timer_wait_until(p->next_expiration);
1214 for (i = 0; i < p->n_listeners; i++) {
1215 pvconn_wait(p->listeners[i]);
1217 for (i = 0; i < p->n_snoops; i++) {
1218 pvconn_wait(p->snoops[i]);
1223 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1225 tag_set_add(&ofproto->revalidate_set, tag);
1229 ofproto_get_revalidate_set(struct ofproto *ofproto)
1231 return &ofproto->revalidate_set;
1235 ofproto_is_alive(const struct ofproto *p)
1237 return !hmap_is_empty(&p->controllers);
1241 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1242 const union ofp_action *actions, size_t n_actions,
1243 const struct ofpbuf *packet)
1245 struct odp_actions odp_actions;
1248 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1254 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1256 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1257 odp_actions.n_actions, packet);
1262 ofproto_add_flow(struct ofproto *p,
1263 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1264 const union ofp_action *actions, size_t n_actions,
1268 rule = rule_create(p, NULL, actions, n_actions,
1269 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1271 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1272 rule_insert(p, rule, NULL, 0);
1276 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1277 uint32_t wildcards, unsigned int priority)
1281 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1285 rule_remove(ofproto, rule);
1290 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1292 struct rule *rule = rule_from_cls_rule(rule_);
1293 struct ofproto *ofproto = ofproto_;
1295 /* Mark the flow as not installed, even though it might really be
1296 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1297 * There is no point in uninstalling it individually since we are about to
1298 * blow away all the flows with dpif_flow_flush(). */
1299 rule->installed = false;
1301 rule_remove(ofproto, rule);
1305 ofproto_flush_flows(struct ofproto *ofproto)
1307 COVERAGE_INC(ofproto_flush);
1308 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1309 dpif_flow_flush(ofproto->dpif);
1310 if (ofproto->in_band) {
1311 in_band_flushed(ofproto->in_band);
1313 if (ofproto->fail_open) {
1314 fail_open_flushed(ofproto->fail_open);
1319 reinit_ports(struct ofproto *p)
1321 struct svec devnames;
1322 struct ofport *ofport;
1323 unsigned int port_no;
1324 struct odp_port *odp_ports;
1328 svec_init(&devnames);
1329 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1330 svec_add (&devnames, (char *) ofport->opp.name);
1332 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1333 for (i = 0; i < n_odp_ports; i++) {
1334 svec_add (&devnames, odp_ports[i].devname);
1338 svec_sort_unique(&devnames);
1339 for (i = 0; i < devnames.n; i++) {
1340 update_port(p, devnames.names[i]);
1342 svec_destroy(&devnames);
1346 refresh_port_group(struct ofproto *p, unsigned int group)
1350 struct ofport *port;
1351 unsigned int port_no;
1353 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1355 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1357 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1358 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1359 ports[n_ports++] = port_no;
1362 dpif_port_group_set(p->dpif, group, ports, n_ports);
1369 refresh_port_groups(struct ofproto *p)
1371 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1372 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1374 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1378 static struct ofport *
1379 make_ofport(const struct odp_port *odp_port)
1381 struct netdev_options netdev_options;
1382 enum netdev_flags flags;
1383 struct ofport *ofport;
1384 struct netdev *netdev;
1388 memset(&netdev_options, 0, sizeof netdev_options);
1389 netdev_options.name = odp_port->devname;
1390 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1392 error = netdev_open(&netdev_options, &netdev);
1394 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1395 "cannot be opened (%s)",
1396 odp_port->devname, odp_port->port,
1397 odp_port->devname, strerror(error));
1401 ofport = xmalloc(sizeof *ofport);
1402 ofport->netdev = netdev;
1403 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1404 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1405 memcpy(ofport->opp.name, odp_port->devname,
1406 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1407 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1409 netdev_get_flags(netdev, &flags);
1410 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1412 netdev_get_carrier(netdev, &carrier);
1413 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1415 netdev_get_features(netdev,
1416 &ofport->opp.curr, &ofport->opp.advertised,
1417 &ofport->opp.supported, &ofport->opp.peer);
1422 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1424 if (port_array_get(&p->ports, odp_port->port)) {
1425 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1428 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1429 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1438 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1440 const struct ofp_phy_port *a = &a_->opp;
1441 const struct ofp_phy_port *b = &b_->opp;
1443 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1444 return (a->port_no == b->port_no
1445 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1446 && !strcmp((char *) a->name, (char *) b->name)
1447 && a->state == b->state
1448 && a->config == b->config
1449 && a->curr == b->curr
1450 && a->advertised == b->advertised
1451 && a->supported == b->supported
1452 && a->peer == b->peer);
1456 send_port_status(struct ofproto *p, const struct ofport *ofport,
1459 /* XXX Should limit the number of queued port status change messages. */
1460 struct ofconn *ofconn;
1461 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1462 struct ofp_port_status *ops;
1465 if (!ofconn_receives_async_msgs(ofconn)) {
1469 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1470 ops->reason = reason;
1471 ops->desc = ofport->opp;
1472 hton_ofp_phy_port(&ops->desc);
1473 queue_tx(b, ofconn, NULL);
1475 if (p->ofhooks->port_changed_cb) {
1476 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1481 ofport_install(struct ofproto *p, struct ofport *ofport)
1483 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1484 const char *netdev_name = (const char *) ofport->opp.name;
1486 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1487 port_array_set(&p->ports, odp_port, ofport);
1488 shash_add(&p->port_by_name, netdev_name, ofport);
1490 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1495 ofport_remove(struct ofproto *p, struct ofport *ofport)
1497 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1499 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1500 port_array_delete(&p->ports, odp_port);
1501 shash_delete(&p->port_by_name,
1502 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1504 ofproto_sflow_del_port(p->sflow, odp_port);
1509 ofport_free(struct ofport *ofport)
1512 netdev_close(ofport->netdev);
1518 update_port(struct ofproto *p, const char *devname)
1520 struct odp_port odp_port;
1521 struct ofport *old_ofport;
1522 struct ofport *new_ofport;
1525 COVERAGE_INC(ofproto_update_port);
1527 /* Query the datapath for port information. */
1528 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1530 /* Find the old ofport. */
1531 old_ofport = shash_find_data(&p->port_by_name, devname);
1534 /* There's no port named 'devname' but there might be a port with
1535 * the same port number. This could happen if a port is deleted
1536 * and then a new one added in its place very quickly, or if a port
1537 * is renamed. In the former case we want to send an OFPPR_DELETE
1538 * and an OFPPR_ADD, and in the latter case we want to send a
1539 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1540 * the old port's ifindex against the new port, or perhaps less
1541 * reliably but more portably by comparing the old port's MAC
1542 * against the new port's MAC. However, this code isn't that smart
1543 * and always sends an OFPPR_MODIFY (XXX). */
1544 old_ofport = port_array_get(&p->ports, odp_port.port);
1546 } else if (error != ENOENT && error != ENODEV) {
1547 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1548 "%s", strerror(error));
1552 /* Create a new ofport. */
1553 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1555 /* Eliminate a few pathological cases. */
1556 if (!old_ofport && !new_ofport) {
1558 } else if (old_ofport && new_ofport) {
1559 /* Most of the 'config' bits are OpenFlow soft state, but
1560 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1561 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1562 * leaves the other bits 0.) */
1563 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1565 if (ofport_equal(old_ofport, new_ofport)) {
1566 /* False alarm--no change. */
1567 ofport_free(new_ofport);
1572 /* Now deal with the normal cases. */
1574 ofport_remove(p, old_ofport);
1577 ofport_install(p, new_ofport);
1579 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1580 (!old_ofport ? OFPPR_ADD
1581 : !new_ofport ? OFPPR_DELETE
1583 ofport_free(old_ofport);
1585 /* Update port groups. */
1586 refresh_port_groups(p);
1590 init_ports(struct ofproto *p)
1592 struct odp_port *ports;
1597 error = dpif_port_list(p->dpif, &ports, &n_ports);
1602 for (i = 0; i < n_ports; i++) {
1603 const struct odp_port *odp_port = &ports[i];
1604 if (!ofport_conflicts(p, odp_port)) {
1605 struct ofport *ofport = make_ofport(odp_port);
1607 ofport_install(p, ofport);
1612 refresh_port_groups(p);
1616 static struct ofconn *
1617 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1619 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1620 ofconn->ofproto = p;
1621 list_push_back(&p->all_conns, &ofconn->node);
1622 ofconn->rconn = rconn;
1623 ofconn->type = type;
1624 ofconn->role = NX_ROLE_OTHER;
1625 ofconn->packet_in_counter = rconn_packet_counter_create ();
1626 ofconn->pktbuf = NULL;
1627 ofconn->miss_send_len = 0;
1628 ofconn->reply_counter = rconn_packet_counter_create ();
1633 ofconn_destroy(struct ofconn *ofconn)
1635 if (ofconn->type == OFCONN_CONTROLLER) {
1636 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1638 discovery_destroy(ofconn->discovery);
1640 list_remove(&ofconn->node);
1641 switch_status_unregister(ofconn->ss);
1642 rconn_destroy(ofconn->rconn);
1643 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1644 rconn_packet_counter_destroy(ofconn->reply_counter);
1645 pktbuf_destroy(ofconn->pktbuf);
1650 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1655 if (ofconn->discovery) {
1656 char *controller_name;
1657 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1658 discovery_question_connectivity(ofconn->discovery);
1660 if (discovery_run(ofconn->discovery, &controller_name)) {
1661 if (controller_name) {
1662 char *ofconn_name = ofconn_make_name(p, controller_name);
1663 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1666 rconn_disconnect(ofconn->rconn);
1671 for (i = 0; i < N_SCHEDULERS; i++) {
1672 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1675 rconn_run(ofconn->rconn);
1677 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1678 /* Limit the number of iterations to prevent other tasks from
1680 for (iteration = 0; iteration < 50; iteration++) {
1681 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1686 fail_open_maybe_recover(p->fail_open);
1688 handle_openflow(ofconn, p, of_msg);
1689 ofpbuf_delete(of_msg);
1693 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1694 ofconn_destroy(ofconn);
1699 ofconn_wait(struct ofconn *ofconn)
1703 if (ofconn->discovery) {
1704 discovery_wait(ofconn->discovery);
1706 for (i = 0; i < N_SCHEDULERS; i++) {
1707 pinsched_wait(ofconn->schedulers[i]);
1709 rconn_run_wait(ofconn->rconn);
1710 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1711 rconn_recv_wait(ofconn->rconn);
1713 COVERAGE_INC(ofproto_ofconn_stuck);
1717 /* Returns true if 'ofconn' should receive asynchronous messages. */
1719 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1721 if (ofconn->type == OFCONN_CONTROLLER) {
1722 /* Ordinary controllers always get asynchronous messages unless they
1723 * have configured themselves as "slaves". */
1724 return ofconn->role != NX_ROLE_SLAVE;
1726 /* Transient connections don't get asynchronous messages unless they
1727 * have explicitly asked for them by setting a nonzero miss send
1729 return ofconn->miss_send_len > 0;
1733 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1734 * and 'target', suitable for use in log messages for identifying the
1737 * The name is dynamically allocated. The caller should free it (with free())
1738 * when it is no longer needed. */
1740 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1742 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1745 /* Caller is responsible for initializing the 'cr' member of the returned
1747 static struct rule *
1748 rule_create(struct ofproto *ofproto, struct rule *super,
1749 const union ofp_action *actions, size_t n_actions,
1750 uint16_t idle_timeout, uint16_t hard_timeout,
1751 uint64_t flow_cookie, bool send_flow_removed)
1753 struct rule *rule = xzalloc(sizeof *rule);
1754 rule->idle_timeout = idle_timeout;
1755 rule->hard_timeout = hard_timeout;
1756 rule->flow_cookie = flow_cookie;
1757 rule->used = rule->created = time_msec();
1758 rule->send_flow_removed = send_flow_removed;
1759 rule->super = super;
1761 list_push_back(&super->list, &rule->list);
1763 list_init(&rule->list);
1765 rule->n_actions = n_actions;
1766 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1767 netflow_flow_clear(&rule->nf_flow);
1768 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1773 static struct rule *
1774 rule_from_cls_rule(const struct cls_rule *cls_rule)
1776 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1780 rule_free(struct rule *rule)
1782 free(rule->actions);
1783 free(rule->odp_actions);
1787 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1788 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1789 * through all of its subrules and revalidates them, destroying any that no
1790 * longer has a super-rule (which is probably all of them).
1792 * Before calling this function, the caller must make have removed 'rule' from
1793 * the classifier. If 'rule' is an exact-match rule, the caller is also
1794 * responsible for ensuring that it has been uninstalled from the datapath. */
1796 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1799 struct rule *subrule, *next;
1800 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1801 revalidate_rule(ofproto, subrule);
1804 list_remove(&rule->list);
1810 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1812 const union ofp_action *oa;
1813 struct actions_iterator i;
1815 if (out_port == htons(OFPP_NONE)) {
1818 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1819 oa = actions_next(&i)) {
1820 if (action_outputs_to_port(oa, out_port)) {
1827 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1828 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1830 * The flow that 'packet' actually contains does not need to actually match
1831 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1832 * the packet and byte counters for 'rule' will be credited for the packet sent
1833 * out whether or not the packet actually matches 'rule'.
1835 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1836 * the caller must already have accurately composed ODP actions for it given
1837 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1838 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1839 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1840 * actions and apply them to 'packet'. */
1842 rule_execute(struct ofproto *ofproto, struct rule *rule,
1843 struct ofpbuf *packet, const flow_t *flow)
1845 const union odp_action *actions;
1847 struct odp_actions a;
1849 /* Grab or compose the ODP actions.
1851 * The special case for an exact-match 'rule' where 'flow' is not the
1852 * rule's flow is important to avoid, e.g., sending a packet out its input
1853 * port simply because the ODP actions were composed for the wrong
1855 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1856 struct rule *super = rule->super ? rule->super : rule;
1857 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1858 packet, &a, NULL, 0, NULL)) {
1861 actions = a.actions;
1862 n_actions = a.n_actions;
1864 actions = rule->odp_actions;
1865 n_actions = rule->n_odp_actions;
1868 /* Execute the ODP actions. */
1869 if (!dpif_execute(ofproto->dpif, flow->in_port,
1870 actions, n_actions, packet)) {
1871 struct odp_flow_stats stats;
1872 flow_extract_stats(flow, packet, &stats);
1873 update_stats(ofproto, rule, &stats);
1874 rule->used = time_msec();
1875 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1880 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1883 struct rule *displaced_rule;
1885 /* Insert the rule in the classifier. */
1886 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1887 if (!rule->cr.wc.wildcards) {
1888 rule_make_actions(p, rule, packet);
1891 /* Send the packet and credit it to the rule. */
1894 flow_extract(packet, 0, in_port, &flow);
1895 rule_execute(p, rule, packet, &flow);
1898 /* Install the rule in the datapath only after sending the packet, to
1899 * avoid packet reordering. */
1900 if (rule->cr.wc.wildcards) {
1901 COVERAGE_INC(ofproto_add_wc_flow);
1902 p->need_revalidate = true;
1904 rule_install(p, rule, displaced_rule);
1907 /* Free the rule that was displaced, if any. */
1908 if (displaced_rule) {
1909 rule_destroy(p, displaced_rule);
1913 static struct rule *
1914 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1917 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1918 rule->idle_timeout, rule->hard_timeout,
1920 COVERAGE_INC(ofproto_subrule_create);
1921 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1922 : rule->cr.priority), &subrule->cr);
1923 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1929 rule_remove(struct ofproto *ofproto, struct rule *rule)
1931 if (rule->cr.wc.wildcards) {
1932 COVERAGE_INC(ofproto_del_wc_flow);
1933 ofproto->need_revalidate = true;
1935 rule_uninstall(ofproto, rule);
1937 classifier_remove(&ofproto->cls, &rule->cr);
1938 rule_destroy(ofproto, rule);
1941 /* Returns true if the actions changed, false otherwise. */
1943 rule_make_actions(struct ofproto *p, struct rule *rule,
1944 const struct ofpbuf *packet)
1946 const struct rule *super;
1947 struct odp_actions a;
1950 assert(!rule->cr.wc.wildcards);
1952 super = rule->super ? rule->super : rule;
1954 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1955 packet, &a, &rule->tags, &rule->may_install,
1956 &rule->nf_flow.output_iface);
1958 actions_len = a.n_actions * sizeof *a.actions;
1959 if (rule->n_odp_actions != a.n_actions
1960 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1961 COVERAGE_INC(ofproto_odp_unchanged);
1962 free(rule->odp_actions);
1963 rule->n_odp_actions = a.n_actions;
1964 rule->odp_actions = xmemdup(a.actions, actions_len);
1972 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1973 struct odp_flow_put *put)
1975 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1976 put->flow.key = rule->cr.flow;
1977 put->flow.actions = rule->odp_actions;
1978 put->flow.n_actions = rule->n_odp_actions;
1979 put->flow.flags = 0;
1981 return dpif_flow_put(ofproto->dpif, put);
1985 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1987 assert(!rule->cr.wc.wildcards);
1989 if (rule->may_install) {
1990 struct odp_flow_put put;
1991 if (!do_put_flow(p, rule,
1992 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1994 rule->installed = true;
1995 if (displaced_rule) {
1996 update_stats(p, displaced_rule, &put.flow.stats);
1997 rule_post_uninstall(p, displaced_rule);
2000 } else if (displaced_rule) {
2001 rule_uninstall(p, displaced_rule);
2006 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2008 if (rule->installed) {
2009 struct odp_flow_put put;
2010 COVERAGE_INC(ofproto_dp_missed);
2011 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2013 rule_install(ofproto, rule, NULL);
2018 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2020 bool actions_changed;
2021 uint16_t new_out_iface, old_out_iface;
2023 old_out_iface = rule->nf_flow.output_iface;
2024 actions_changed = rule_make_actions(ofproto, rule, NULL);
2026 if (rule->may_install) {
2027 if (rule->installed) {
2028 if (actions_changed) {
2029 struct odp_flow_put put;
2030 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2031 | ODPPF_ZERO_STATS, &put);
2032 update_stats(ofproto, rule, &put.flow.stats);
2034 /* Temporarily set the old output iface so that NetFlow
2035 * messages have the correct output interface for the old
2037 new_out_iface = rule->nf_flow.output_iface;
2038 rule->nf_flow.output_iface = old_out_iface;
2039 rule_post_uninstall(ofproto, rule);
2040 rule->nf_flow.output_iface = new_out_iface;
2043 rule_install(ofproto, rule, NULL);
2046 rule_uninstall(ofproto, rule);
2051 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2053 uint64_t total_bytes = rule->byte_count + extra_bytes;
2055 if (ofproto->ofhooks->account_flow_cb
2056 && total_bytes > rule->accounted_bytes)
2058 ofproto->ofhooks->account_flow_cb(
2059 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2060 total_bytes - rule->accounted_bytes, ofproto->aux);
2061 rule->accounted_bytes = total_bytes;
2066 rule_uninstall(struct ofproto *p, struct rule *rule)
2068 assert(!rule->cr.wc.wildcards);
2069 if (rule->installed) {
2070 struct odp_flow odp_flow;
2072 odp_flow.key = rule->cr.flow;
2073 odp_flow.actions = NULL;
2074 odp_flow.n_actions = 0;
2076 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2077 update_stats(p, rule, &odp_flow.stats);
2079 rule->installed = false;
2081 rule_post_uninstall(p, rule);
2086 is_controller_rule(struct rule *rule)
2088 /* If the only action is send to the controller then don't report
2089 * NetFlow expiration messages since it is just part of the control
2090 * logic for the network and not real traffic. */
2094 && rule->super->n_actions == 1
2095 && action_outputs_to_port(&rule->super->actions[0],
2096 htons(OFPP_CONTROLLER)));
2100 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2102 struct rule *super = rule->super;
2104 rule_account(ofproto, rule, 0);
2106 if (ofproto->netflow && !is_controller_rule(rule)) {
2107 struct ofexpired expired;
2108 expired.flow = rule->cr.flow;
2109 expired.packet_count = rule->packet_count;
2110 expired.byte_count = rule->byte_count;
2111 expired.used = rule->used;
2112 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2115 super->packet_count += rule->packet_count;
2116 super->byte_count += rule->byte_count;
2118 /* Reset counters to prevent double counting if the rule ever gets
2120 rule->packet_count = 0;
2121 rule->byte_count = 0;
2122 rule->accounted_bytes = 0;
2124 netflow_flow_clear(&rule->nf_flow);
2129 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2130 struct rconn_packet_counter *counter)
2132 update_openflow_length(msg);
2133 if (rconn_send(ofconn->rconn, msg, counter)) {
2139 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2140 int error, const void *data, size_t len)
2143 struct ofp_error_msg *oem;
2145 if (!(error >> 16)) {
2146 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2151 COVERAGE_INC(ofproto_error);
2152 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2153 oh ? oh->xid : 0, &buf);
2154 oem->type = htons((unsigned int) error >> 16);
2155 oem->code = htons(error & 0xffff);
2156 memcpy(oem->data, data, len);
2157 queue_tx(buf, ofconn, ofconn->reply_counter);
2161 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2164 size_t oh_length = ntohs(oh->length);
2165 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2169 hton_ofp_phy_port(struct ofp_phy_port *opp)
2171 opp->port_no = htons(opp->port_no);
2172 opp->config = htonl(opp->config);
2173 opp->state = htonl(opp->state);
2174 opp->curr = htonl(opp->curr);
2175 opp->advertised = htonl(opp->advertised);
2176 opp->supported = htonl(opp->supported);
2177 opp->peer = htonl(opp->peer);
2181 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2183 struct ofp_header *rq = oh;
2184 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2189 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2190 struct ofp_header *oh)
2192 struct ofp_switch_features *osf;
2194 unsigned int port_no;
2195 struct ofport *port;
2197 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2198 osf->datapath_id = htonll(p->datapath_id);
2199 osf->n_buffers = htonl(pktbuf_capacity());
2201 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2202 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2203 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2204 (1u << OFPAT_SET_VLAN_VID) |
2205 (1u << OFPAT_SET_VLAN_PCP) |
2206 (1u << OFPAT_STRIP_VLAN) |
2207 (1u << OFPAT_SET_DL_SRC) |
2208 (1u << OFPAT_SET_DL_DST) |
2209 (1u << OFPAT_SET_NW_SRC) |
2210 (1u << OFPAT_SET_NW_DST) |
2211 (1u << OFPAT_SET_NW_TOS) |
2212 (1u << OFPAT_SET_TP_SRC) |
2213 (1u << OFPAT_SET_TP_DST) |
2214 (1u << OFPAT_ENQUEUE));
2216 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2217 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2220 queue_tx(buf, ofconn, ofconn->reply_counter);
2225 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2226 struct ofp_header *oh)
2229 struct ofp_switch_config *osc;
2233 /* Figure out flags. */
2234 dpif_get_drop_frags(p->dpif, &drop_frags);
2235 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2238 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2239 osc->flags = htons(flags);
2240 osc->miss_send_len = htons(ofconn->miss_send_len);
2241 queue_tx(buf, ofconn, ofconn->reply_counter);
2247 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2248 struct ofp_switch_config *osc)
2253 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2257 flags = ntohs(osc->flags);
2259 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2260 switch (flags & OFPC_FRAG_MASK) {
2261 case OFPC_FRAG_NORMAL:
2262 dpif_set_drop_frags(p->dpif, false);
2264 case OFPC_FRAG_DROP:
2265 dpif_set_drop_frags(p->dpif, true);
2268 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2274 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2280 add_output_group_action(struct odp_actions *actions, uint16_t group,
2281 uint16_t *nf_output_iface)
2283 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2285 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2286 *nf_output_iface = NF_OUT_FLOOD;
2291 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2293 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2294 a->controller.arg = max_len;
2297 struct action_xlate_ctx {
2299 flow_t flow; /* Flow to which these actions correspond. */
2300 int recurse; /* Recursion level, via xlate_table_action. */
2301 struct ofproto *ofproto;
2302 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2303 * null pointer if we are revalidating
2304 * without a packet to refer to. */
2307 struct odp_actions *out; /* Datapath actions. */
2308 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2309 bool may_set_up_flow; /* True ordinarily; false if the actions must
2310 * be reassessed for every packet. */
2311 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2314 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2315 struct action_xlate_ctx *ctx);
2318 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2320 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2323 if (ofport->opp.config & OFPPC_NO_FWD) {
2324 /* Forwarding disabled on port. */
2329 * We don't have an ofport record for this port, but it doesn't hurt to
2330 * allow forwarding to it anyhow. Maybe such a port will appear later
2331 * and we're pre-populating the flow table.
2335 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2336 ctx->nf_output_iface = port;
2339 static struct rule *
2340 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2343 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2345 /* The rule we found might not be valid, since we could be in need of
2346 * revalidation. If it is not valid, don't return it. */
2349 && ofproto->need_revalidate
2350 && !revalidate_rule(ofproto, rule)) {
2351 COVERAGE_INC(ofproto_invalidated);
2359 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2361 if (!ctx->recurse) {
2362 uint16_t old_in_port;
2365 /* Look up a flow with 'in_port' as the input port. Then restore the
2366 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2367 * have surprising behavior). */
2368 old_in_port = ctx->flow.in_port;
2369 ctx->flow.in_port = in_port;
2370 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2371 ctx->flow.in_port = old_in_port;
2379 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2386 xlate_output_action__(struct action_xlate_ctx *ctx,
2387 uint16_t port, uint16_t max_len)
2390 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2392 ctx->nf_output_iface = NF_OUT_DROP;
2396 add_output_action(ctx, ctx->flow.in_port);
2399 xlate_table_action(ctx, ctx->flow.in_port);
2402 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2403 ctx->out, ctx->tags,
2404 &ctx->nf_output_iface,
2405 ctx->ofproto->aux)) {
2406 COVERAGE_INC(ofproto_uninstallable);
2407 ctx->may_set_up_flow = false;
2411 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2412 &ctx->nf_output_iface);
2415 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2417 case OFPP_CONTROLLER:
2418 add_controller_action(ctx->out, max_len);
2421 add_output_action(ctx, ODPP_LOCAL);
2424 odp_port = ofp_port_to_odp_port(port);
2425 if (odp_port != ctx->flow.in_port) {
2426 add_output_action(ctx, odp_port);
2431 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2432 ctx->nf_output_iface = NF_OUT_FLOOD;
2433 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2434 ctx->nf_output_iface = prev_nf_output_iface;
2435 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2436 ctx->nf_output_iface != NF_OUT_FLOOD) {
2437 ctx->nf_output_iface = NF_OUT_MULTI;
2442 xlate_output_action(struct action_xlate_ctx *ctx,
2443 const struct ofp_action_output *oao)
2445 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2448 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2449 * optimization, because we're going to add another action that sets the
2450 * priority immediately after, or because there are no actions following the
2453 remove_pop_action(struct action_xlate_ctx *ctx)
2455 size_t n = ctx->out->n_actions;
2456 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2457 ctx->out->n_actions--;
2462 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2463 const struct ofp_action_enqueue *oae)
2465 uint16_t ofp_port, odp_port;
2469 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2472 /* Fall back to ordinary output action. */
2473 xlate_output_action__(ctx, ntohs(oae->port), 0);
2477 /* Figure out ODP output port. */
2478 ofp_port = ntohs(oae->port);
2479 if (ofp_port != OFPP_IN_PORT) {
2480 odp_port = ofp_port_to_odp_port(ofp_port);
2482 odp_port = ctx->flow.in_port;
2485 /* Add ODP actions. */
2486 remove_pop_action(ctx);
2487 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2489 add_output_action(ctx, odp_port);
2490 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2492 /* Update NetFlow output port. */
2493 if (ctx->nf_output_iface == NF_OUT_DROP) {
2494 ctx->nf_output_iface = odp_port;
2495 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2496 ctx->nf_output_iface = NF_OUT_MULTI;
2501 xlate_nicira_action(struct action_xlate_ctx *ctx,
2502 const struct nx_action_header *nah)
2504 const struct nx_action_resubmit *nar;
2505 const struct nx_action_set_tunnel *nast;
2506 union odp_action *oa;
2507 int subtype = ntohs(nah->subtype);
2509 assert(nah->vendor == htonl(NX_VENDOR_ID));
2511 case NXAST_RESUBMIT:
2512 nar = (const struct nx_action_resubmit *) nah;
2513 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2516 case NXAST_SET_TUNNEL:
2517 nast = (const struct nx_action_set_tunnel *) nah;
2518 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2519 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2522 /* If you add a new action here that modifies flow data, don't forget to
2523 * update the flow key in ctx->flow at the same time. */
2526 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2532 do_xlate_actions(const union ofp_action *in, size_t n_in,
2533 struct action_xlate_ctx *ctx)
2535 struct actions_iterator iter;
2536 const union ofp_action *ia;
2537 const struct ofport *port;
2539 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2540 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2541 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2542 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2543 /* Drop this flow. */
2547 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2548 uint16_t type = ntohs(ia->type);
2549 union odp_action *oa;
2553 xlate_output_action(ctx, &ia->output);
2556 case OFPAT_SET_VLAN_VID:
2557 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2558 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2561 case OFPAT_SET_VLAN_PCP:
2562 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2563 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2566 case OFPAT_STRIP_VLAN:
2567 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2568 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2569 ctx->flow.dl_vlan_pcp = 0;
2572 case OFPAT_SET_DL_SRC:
2573 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2574 memcpy(oa->dl_addr.dl_addr,
2575 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2576 memcpy(ctx->flow.dl_src,
2577 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2580 case OFPAT_SET_DL_DST:
2581 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2582 memcpy(oa->dl_addr.dl_addr,
2583 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2584 memcpy(ctx->flow.dl_dst,
2585 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2588 case OFPAT_SET_NW_SRC:
2589 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2590 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2593 case OFPAT_SET_NW_DST:
2594 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2595 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2598 case OFPAT_SET_NW_TOS:
2599 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2600 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2603 case OFPAT_SET_TP_SRC:
2604 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2605 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2608 case OFPAT_SET_TP_DST:
2609 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2610 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2614 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2618 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2622 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2629 xlate_actions(const union ofp_action *in, size_t n_in,
2630 const flow_t *flow, struct ofproto *ofproto,
2631 const struct ofpbuf *packet,
2632 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2633 uint16_t *nf_output_iface)
2635 tag_type no_tags = 0;
2636 struct action_xlate_ctx ctx;
2637 COVERAGE_INC(ofproto_ofp2odp);
2638 odp_actions_init(out);
2641 ctx.ofproto = ofproto;
2642 ctx.packet = packet;
2644 ctx.tags = tags ? tags : &no_tags;
2645 ctx.may_set_up_flow = true;
2646 ctx.nf_output_iface = NF_OUT_DROP;
2647 do_xlate_actions(in, n_in, &ctx);
2648 remove_pop_action(&ctx);
2650 /* Check with in-band control to see if we're allowed to set up this
2652 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2653 ctx.may_set_up_flow = false;
2656 if (may_set_up_flow) {
2657 *may_set_up_flow = ctx.may_set_up_flow;
2659 if (nf_output_iface) {
2660 *nf_output_iface = ctx.nf_output_iface;
2662 if (odp_actions_overflow(out)) {
2663 odp_actions_init(out);
2664 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2669 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2670 * error message code (composed with ofp_mkerr()) for the caller to propagate
2671 * upward. Otherwise, returns 0.
2673 * 'oh' is used to make log messages more informative. */
2675 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2677 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2678 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2681 type_name = ofp_message_type_to_string(oh->type);
2682 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2686 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2693 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2694 struct ofp_header *oh)
2696 struct ofp_packet_out *opo;
2697 struct ofpbuf payload, *buffer;
2698 struct odp_actions actions;
2704 error = reject_slave_controller(ofconn, oh);
2709 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2713 opo = (struct ofp_packet_out *) oh;
2715 COVERAGE_INC(ofproto_packet_out);
2716 if (opo->buffer_id != htonl(UINT32_MAX)) {
2717 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2719 if (error || !buffer) {
2727 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2728 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2729 &flow, p, &payload, &actions, NULL, NULL, NULL);
2734 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2736 ofpbuf_delete(buffer);
2742 update_port_config(struct ofproto *p, struct ofport *port,
2743 uint32_t config, uint32_t mask)
2745 mask &= config ^ port->opp.config;
2746 if (mask & OFPPC_PORT_DOWN) {
2747 if (config & OFPPC_PORT_DOWN) {
2748 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2750 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2753 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2754 if (mask & REVALIDATE_BITS) {
2755 COVERAGE_INC(ofproto_costly_flags);
2756 port->opp.config ^= mask & REVALIDATE_BITS;
2757 p->need_revalidate = true;
2759 #undef REVALIDATE_BITS
2760 if (mask & OFPPC_NO_FLOOD) {
2761 port->opp.config ^= OFPPC_NO_FLOOD;
2762 refresh_port_groups(p);
2764 if (mask & OFPPC_NO_PACKET_IN) {
2765 port->opp.config ^= OFPPC_NO_PACKET_IN;
2770 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2771 struct ofp_header *oh)
2773 const struct ofp_port_mod *opm;
2774 struct ofport *port;
2777 error = reject_slave_controller(ofconn, oh);
2781 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2785 opm = (struct ofp_port_mod *) oh;
2787 port = port_array_get(&p->ports,
2788 ofp_port_to_odp_port(ntohs(opm->port_no)));
2790 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2791 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2792 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2794 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2795 if (opm->advertise) {
2796 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2802 static struct ofpbuf *
2803 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2805 struct ofp_stats_reply *osr;
2808 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2809 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2811 osr->flags = htons(0);
2815 static struct ofpbuf *
2816 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2818 return make_stats_reply(request->header.xid, request->type, body_len);
2822 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2824 struct ofpbuf *msg = *msgp;
2825 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2826 if (nbytes + msg->size > UINT16_MAX) {
2827 struct ofp_stats_reply *reply = msg->data;
2828 reply->flags = htons(OFPSF_REPLY_MORE);
2829 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2830 queue_tx(msg, ofconn, ofconn->reply_counter);
2832 return ofpbuf_put_uninit(*msgp, nbytes);
2836 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2837 struct ofp_stats_request *request)
2839 struct ofp_desc_stats *ods;
2842 msg = start_stats_reply(request, sizeof *ods);
2843 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2844 memset(ods, 0, sizeof *ods);
2845 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2846 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2847 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2848 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2849 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2850 queue_tx(msg, ofconn, ofconn->reply_counter);
2856 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2858 struct rule *rule = rule_from_cls_rule(cls_rule);
2859 int *n_subrules = n_subrules_;
2867 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2868 struct ofp_stats_request *request)
2870 struct ofp_table_stats *ots;
2872 struct odp_stats dpstats;
2873 int n_exact, n_subrules, n_wild;
2875 msg = start_stats_reply(request, sizeof *ots * 2);
2877 /* Count rules of various kinds. */
2879 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2880 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2881 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2884 dpif_get_dp_stats(p->dpif, &dpstats);
2885 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2886 memset(ots, 0, sizeof *ots);
2887 ots->table_id = TABLEID_HASH;
2888 strcpy(ots->name, "hash");
2889 ots->wildcards = htonl(0);
2890 ots->max_entries = htonl(dpstats.max_capacity);
2891 ots->active_count = htonl(n_exact);
2892 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2894 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2896 /* Classifier table. */
2897 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2898 memset(ots, 0, sizeof *ots);
2899 ots->table_id = TABLEID_CLASSIFIER;
2900 strcpy(ots->name, "classifier");
2901 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2903 ots->max_entries = htonl(65536);
2904 ots->active_count = htonl(n_wild);
2905 ots->lookup_count = htonll(0); /* XXX */
2906 ots->matched_count = htonll(0); /* XXX */
2908 queue_tx(msg, ofconn, ofconn->reply_counter);
2913 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2914 struct ofpbuf **msgp)
2916 struct netdev_stats stats;
2917 struct ofp_port_stats *ops;
2919 /* Intentionally ignore return value, since errors will set
2920 * 'stats' to all-1s, which is correct for OpenFlow, and
2921 * netdev_get_stats() will log errors. */
2922 netdev_get_stats(port->netdev, &stats);
2924 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2925 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2926 memset(ops->pad, 0, sizeof ops->pad);
2927 ops->rx_packets = htonll(stats.rx_packets);
2928 ops->tx_packets = htonll(stats.tx_packets);
2929 ops->rx_bytes = htonll(stats.rx_bytes);
2930 ops->tx_bytes = htonll(stats.tx_bytes);
2931 ops->rx_dropped = htonll(stats.rx_dropped);
2932 ops->tx_dropped = htonll(stats.tx_dropped);
2933 ops->rx_errors = htonll(stats.rx_errors);
2934 ops->tx_errors = htonll(stats.tx_errors);
2935 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2936 ops->rx_over_err = htonll(stats.rx_over_errors);
2937 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2938 ops->collisions = htonll(stats.collisions);
2942 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2943 struct ofp_stats_request *osr,
2946 struct ofp_port_stats_request *psr;
2947 struct ofp_port_stats *ops;
2949 struct ofport *port;
2950 unsigned int port_no;
2952 if (arg_size != sizeof *psr) {
2953 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2955 psr = (struct ofp_port_stats_request *) osr->body;
2957 msg = start_stats_reply(osr, sizeof *ops * 16);
2958 if (psr->port_no != htons(OFPP_NONE)) {
2959 port = port_array_get(&p->ports,
2960 ofp_port_to_odp_port(ntohs(psr->port_no)));
2962 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2965 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2966 append_port_stat(port, port_no, ofconn, &msg);
2970 queue_tx(msg, ofconn, ofconn->reply_counter);
2974 struct flow_stats_cbdata {
2975 struct ofproto *ofproto;
2976 struct ofconn *ofconn;
2981 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2982 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2983 * returned statistic include statistics for all of 'rule''s subrules. */
2985 query_stats(struct ofproto *p, struct rule *rule,
2986 uint64_t *packet_countp, uint64_t *byte_countp)
2988 uint64_t packet_count, byte_count;
2989 struct rule *subrule;
2990 struct odp_flow *odp_flows;
2993 /* Start from historical data for 'rule' itself that are no longer tracked
2994 * by the datapath. This counts, for example, subrules that have
2996 packet_count = rule->packet_count;
2997 byte_count = rule->byte_count;
2999 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3000 * wildcarded then on all of its subrules.
3002 * Also, add any statistics that are not tracked by the datapath for each
3003 * subrule. This includes, for example, statistics for packets that were
3004 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3006 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3007 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3008 if (rule->cr.wc.wildcards) {
3010 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3011 odp_flows[i++].key = subrule->cr.flow;
3012 packet_count += subrule->packet_count;
3013 byte_count += subrule->byte_count;
3016 odp_flows[0].key = rule->cr.flow;
3019 /* Fetch up-to-date statistics from the datapath and add them in. */
3020 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3022 for (i = 0; i < n_odp_flows; i++) {
3023 struct odp_flow *odp_flow = &odp_flows[i];
3024 packet_count += odp_flow->stats.n_packets;
3025 byte_count += odp_flow->stats.n_bytes;
3030 /* Return the stats to the caller. */
3031 *packet_countp = packet_count;
3032 *byte_countp = byte_count;
3036 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3038 struct rule *rule = rule_from_cls_rule(rule_);
3039 struct flow_stats_cbdata *cbdata = cbdata_;
3040 struct ofp_flow_stats *ofs;
3041 uint64_t packet_count, byte_count;
3042 size_t act_len, len;
3043 long long int tdiff = time_msec() - rule->created;
3044 uint32_t sec = tdiff / 1000;
3045 uint32_t msec = tdiff - (sec * 1000);
3047 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3051 act_len = sizeof *rule->actions * rule->n_actions;
3052 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3054 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3056 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3057 ofs->length = htons(len);
3058 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3060 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3061 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3062 ofs->duration_sec = htonl(sec);
3063 ofs->duration_nsec = htonl(msec * 1000000);
3064 ofs->cookie = rule->flow_cookie;
3065 ofs->priority = htons(rule->cr.priority);
3066 ofs->idle_timeout = htons(rule->idle_timeout);
3067 ofs->hard_timeout = htons(rule->hard_timeout);
3068 memset(ofs->pad2, 0, sizeof ofs->pad2);
3069 ofs->packet_count = htonll(packet_count);
3070 ofs->byte_count = htonll(byte_count);
3071 memcpy(ofs->actions, rule->actions, act_len);
3075 table_id_to_include(uint8_t table_id)
3077 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3078 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3079 : table_id == 0xff ? CLS_INC_ALL
3084 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3085 const struct ofp_stats_request *osr,
3088 struct ofp_flow_stats_request *fsr;
3089 struct flow_stats_cbdata cbdata;
3090 struct cls_rule target;
3092 if (arg_size != sizeof *fsr) {
3093 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3095 fsr = (struct ofp_flow_stats_request *) osr->body;
3097 COVERAGE_INC(ofproto_flows_req);
3099 cbdata.ofconn = ofconn;
3100 cbdata.out_port = fsr->out_port;
3101 cbdata.msg = start_stats_reply(osr, 1024);
3102 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3103 classifier_for_each_match(&p->cls, &target,
3104 table_id_to_include(fsr->table_id),
3105 flow_stats_cb, &cbdata);
3106 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3110 struct flow_stats_ds_cbdata {
3111 struct ofproto *ofproto;
3116 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3118 struct rule *rule = rule_from_cls_rule(rule_);
3119 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3120 struct ds *results = cbdata->results;
3121 struct ofp_match match;
3122 uint64_t packet_count, byte_count;
3123 size_t act_len = sizeof *rule->actions * rule->n_actions;
3125 /* Don't report on subrules. */
3126 if (rule->super != NULL) {
3130 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3131 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3132 cbdata->ofproto->tun_id_from_cookie, &match);
3134 ds_put_format(results, "duration=%llds, ",
3135 (time_msec() - rule->created) / 1000);
3136 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3137 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3138 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3139 ofp_print_match(results, &match, true);
3140 ofp_print_actions(results, &rule->actions->header, act_len);
3141 ds_put_cstr(results, "\n");
3144 /* Adds a pretty-printed description of all flows to 'results', including
3145 * those marked hidden by secchan (e.g., by in-band control). */
3147 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3149 struct ofp_match match;
3150 struct cls_rule target;
3151 struct flow_stats_ds_cbdata cbdata;
3153 memset(&match, 0, sizeof match);
3154 match.wildcards = htonl(OVSFW_ALL);
3157 cbdata.results = results;
3159 cls_rule_from_match(&match, 0, false, 0, &target);
3160 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3161 flow_stats_ds_cb, &cbdata);
3164 struct aggregate_stats_cbdata {
3165 struct ofproto *ofproto;
3167 uint64_t packet_count;
3168 uint64_t byte_count;
3173 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3175 struct rule *rule = rule_from_cls_rule(rule_);
3176 struct aggregate_stats_cbdata *cbdata = cbdata_;
3177 uint64_t packet_count, byte_count;
3179 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3183 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3185 cbdata->packet_count += packet_count;
3186 cbdata->byte_count += byte_count;
3191 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3192 const struct ofp_stats_request *osr,
3195 struct ofp_aggregate_stats_request *asr;
3196 struct ofp_aggregate_stats_reply *reply;
3197 struct aggregate_stats_cbdata cbdata;
3198 struct cls_rule target;
3201 if (arg_size != sizeof *asr) {
3202 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3204 asr = (struct ofp_aggregate_stats_request *) osr->body;
3206 COVERAGE_INC(ofproto_agg_request);
3208 cbdata.out_port = asr->out_port;
3209 cbdata.packet_count = 0;
3210 cbdata.byte_count = 0;
3212 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3213 classifier_for_each_match(&p->cls, &target,
3214 table_id_to_include(asr->table_id),
3215 aggregate_stats_cb, &cbdata);
3217 msg = start_stats_reply(osr, sizeof *reply);
3218 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3219 reply->flow_count = htonl(cbdata.n_flows);
3220 reply->packet_count = htonll(cbdata.packet_count);
3221 reply->byte_count = htonll(cbdata.byte_count);
3222 queue_tx(msg, ofconn, ofconn->reply_counter);
3226 struct queue_stats_cbdata {
3227 struct ofconn *ofconn;
3233 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3234 const struct netdev_queue_stats *stats)
3236 struct ofp_queue_stats *reply;
3238 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3239 reply->port_no = htons(cbdata->port_no);
3240 memset(reply->pad, 0, sizeof reply->pad);
3241 reply->queue_id = htonl(queue_id);
3242 reply->tx_bytes = htonll(stats->tx_bytes);
3243 reply->tx_packets = htonll(stats->tx_packets);
3244 reply->tx_errors = htonll(stats->tx_errors);
3248 handle_queue_stats_dump_cb(uint32_t queue_id,
3249 struct netdev_queue_stats *stats,
3252 struct queue_stats_cbdata *cbdata = cbdata_;
3254 put_queue_stats(cbdata, queue_id, stats);
3258 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3260 struct queue_stats_cbdata *cbdata)
3262 cbdata->port_no = port_no;
3263 if (queue_id == OFPQ_ALL) {
3264 netdev_dump_queue_stats(port->netdev,
3265 handle_queue_stats_dump_cb, cbdata);
3267 struct netdev_queue_stats stats;
3269 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3270 put_queue_stats(cbdata, queue_id, &stats);
3275 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3276 const struct ofp_stats_request *osr,
3279 struct ofp_queue_stats_request *qsr;
3280 struct queue_stats_cbdata cbdata;
3281 struct ofport *port;
3282 unsigned int port_no;
3285 if (arg_size != sizeof *qsr) {
3286 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3288 qsr = (struct ofp_queue_stats_request *) osr->body;
3290 COVERAGE_INC(ofproto_queue_req);
3292 cbdata.ofconn = ofconn;
3293 cbdata.msg = start_stats_reply(osr, 128);
3295 port_no = ntohs(qsr->port_no);
3296 queue_id = ntohl(qsr->queue_id);
3297 if (port_no == OFPP_ALL) {
3298 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3299 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3301 } else if (port_no < ofproto->max_ports) {
3302 port = port_array_get(&ofproto->ports, port_no);
3304 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3307 ofpbuf_delete(cbdata.msg);
3308 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3310 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3316 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3317 struct ofp_header *oh)
3319 struct ofp_stats_request *osr;
3323 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3328 osr = (struct ofp_stats_request *) oh;
3330 switch (ntohs(osr->type)) {
3332 return handle_desc_stats_request(p, ofconn, osr);
3335 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3337 case OFPST_AGGREGATE:
3338 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3341 return handle_table_stats_request(p, ofconn, osr);
3344 return handle_port_stats_request(p, ofconn, osr, arg_size);
3347 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3350 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3353 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3357 static long long int
3358 msec_from_nsec(uint64_t sec, uint32_t nsec)
3360 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3364 update_time(struct ofproto *ofproto, struct rule *rule,
3365 const struct odp_flow_stats *stats)
3367 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3368 if (used > rule->used) {
3370 if (rule->super && used > rule->super->used) {
3371 rule->super->used = used;
3373 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3378 update_stats(struct ofproto *ofproto, struct rule *rule,
3379 const struct odp_flow_stats *stats)
3381 if (stats->n_packets) {
3382 update_time(ofproto, rule, stats);
3383 rule->packet_count += stats->n_packets;
3384 rule->byte_count += stats->n_bytes;
3385 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3390 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3391 * in which no matching flow already exists in the flow table.
3393 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3394 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3395 * code as encoded by ofp_mkerr() on failure.
3397 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3400 add_flow(struct ofproto *p, struct ofconn *ofconn,
3401 const struct ofp_flow_mod *ofm, size_t n_actions)
3403 struct ofpbuf *packet;
3408 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3412 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3414 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3415 ntohs(ofm->priority))) {
3416 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3420 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3421 n_actions, ntohs(ofm->idle_timeout),
3422 ntohs(ofm->hard_timeout), ofm->cookie,
3423 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3424 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3425 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3428 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3429 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3433 in_port = UINT16_MAX;
3436 rule_insert(p, rule, packet, in_port);
3437 ofpbuf_delete(packet);
3441 static struct rule *
3442 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3447 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3449 return rule_from_cls_rule(classifier_find_rule_exactly(
3450 &p->cls, &flow, wildcards,
3451 ntohs(ofm->priority)));
3455 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3456 struct rule *rule, const struct ofp_flow_mod *ofm)
3458 struct ofpbuf *packet;
3463 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3467 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3473 flow_extract(packet, 0, in_port, &flow);
3474 rule_execute(ofproto, rule, packet, &flow);
3475 ofpbuf_delete(packet);
3480 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3482 struct modify_flows_cbdata {
3483 struct ofproto *ofproto;
3484 const struct ofp_flow_mod *ofm;
3489 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3490 size_t n_actions, struct rule *);
3491 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3493 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3494 * encoded by ofp_mkerr() on failure.
3496 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3499 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3500 const struct ofp_flow_mod *ofm, size_t n_actions)
3502 struct modify_flows_cbdata cbdata;
3503 struct cls_rule target;
3507 cbdata.n_actions = n_actions;
3508 cbdata.match = NULL;
3510 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3513 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3514 modify_flows_cb, &cbdata);
3516 /* This credits the packet to whichever flow happened to happened to
3517 * match last. That's weird. Maybe we should do a lookup for the
3518 * flow that actually matches the packet? Who knows. */
3519 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3522 return add_flow(p, ofconn, ofm, n_actions);
3526 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3527 * code as encoded by ofp_mkerr() on failure.
3529 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3532 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3533 struct ofp_flow_mod *ofm, size_t n_actions)
3535 struct rule *rule = find_flow_strict(p, ofm);
3536 if (rule && !rule_is_hidden(rule)) {
3537 modify_flow(p, ofm, n_actions, rule);
3538 return send_buffered_packet(p, ofconn, rule, ofm);
3540 return add_flow(p, ofconn, ofm, n_actions);
3544 /* Callback for modify_flows_loose(). */
3546 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3548 struct rule *rule = rule_from_cls_rule(rule_);
3549 struct modify_flows_cbdata *cbdata = cbdata_;
3551 if (!rule_is_hidden(rule)) {
3552 cbdata->match = rule;
3553 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3557 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3558 * been identified as a flow in 'p''s flow table to be modified, by changing
3559 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3560 * ofp_action[] structures). */
3562 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3563 size_t n_actions, struct rule *rule)
3565 size_t actions_len = n_actions * sizeof *rule->actions;
3567 rule->flow_cookie = ofm->cookie;
3569 /* If the actions are the same, do nothing. */
3570 if (n_actions == rule->n_actions
3571 && !memcmp(ofm->actions, rule->actions, actions_len))
3576 /* Replace actions. */
3577 free(rule->actions);
3578 rule->actions = xmemdup(ofm->actions, actions_len);
3579 rule->n_actions = n_actions;
3581 /* Make sure that the datapath gets updated properly. */
3582 if (rule->cr.wc.wildcards) {
3583 COVERAGE_INC(ofproto_mod_wc_flow);
3584 p->need_revalidate = true;
3586 rule_update_actions(p, rule);
3592 /* OFPFC_DELETE implementation. */
3594 struct delete_flows_cbdata {
3595 struct ofproto *ofproto;
3599 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3600 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3602 /* Implements OFPFC_DELETE. */
3604 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3606 struct delete_flows_cbdata cbdata;
3607 struct cls_rule target;
3610 cbdata.out_port = ofm->out_port;
3612 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3615 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3616 delete_flows_cb, &cbdata);
3619 /* Implements OFPFC_DELETE_STRICT. */
3621 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3623 struct rule *rule = find_flow_strict(p, ofm);
3625 delete_flow(p, rule, ofm->out_port);
3629 /* Callback for delete_flows_loose(). */
3631 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3633 struct rule *rule = rule_from_cls_rule(rule_);
3634 struct delete_flows_cbdata *cbdata = cbdata_;
3636 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3639 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3640 * been identified as a flow to delete from 'p''s flow table, by deleting the
3641 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3644 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3645 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3646 * specified 'out_port'. */
3648 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3650 if (rule_is_hidden(rule)) {
3654 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3658 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3659 rule_remove(p, rule);
3663 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3664 struct ofp_flow_mod *ofm)
3666 struct ofp_match orig_match;
3670 error = reject_slave_controller(ofconn, &ofm->header);
3674 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3675 sizeof *ofm->actions, &n_actions);
3680 /* We do not support the emergency flow cache. It will hopefully
3681 * get dropped from OpenFlow in the near future. */
3682 if (ofm->flags & htons(OFPFF_EMERG)) {
3683 /* There isn't a good fit for an error code, so just state that the
3684 * flow table is full. */
3685 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3688 /* Normalize ofp->match. If normalization actually changes anything, then
3689 * log the differences. */
3690 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3691 orig_match = ofm->match;
3692 normalize_match(&ofm->match);
3693 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3694 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3695 if (!VLOG_DROP_INFO(&normal_rl)) {
3696 char *old = ofp_match_to_literal_string(&orig_match);
3697 char *new = ofp_match_to_literal_string(&ofm->match);
3698 VLOG_INFO("%s: normalization changed ofp_match, details:",
3699 rconn_get_name(ofconn->rconn));
3700 VLOG_INFO(" pre: %s", old);
3701 VLOG_INFO("post: %s", new);
3707 if (!ofm->match.wildcards) {
3708 ofm->priority = htons(UINT16_MAX);
3711 error = validate_actions((const union ofp_action *) ofm->actions,
3712 n_actions, p->max_ports);
3717 switch (ntohs(ofm->command)) {
3719 return add_flow(p, ofconn, ofm, n_actions);
3722 return modify_flows_loose(p, ofconn, ofm, n_actions);
3724 case OFPFC_MODIFY_STRICT:
3725 return modify_flow_strict(p, ofconn, ofm, n_actions);
3728 delete_flows_loose(p, ofm);
3731 case OFPFC_DELETE_STRICT:
3732 delete_flow_strict(p, ofm);
3736 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3741 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3745 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3750 p->tun_id_from_cookie = !!msg->set;
3755 handle_role_request(struct ofproto *ofproto,
3756 struct ofconn *ofconn, struct nicira_header *msg)
3758 struct nx_role_request *nrr;
3759 struct nx_role_request *reply;
3763 if (ntohs(msg->header.length) != sizeof *nrr) {
3764 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3765 ntohs(msg->header.length), sizeof *nrr);
3766 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3768 nrr = (struct nx_role_request *) msg;
3770 if (ofconn->type != OFCONN_CONTROLLER) {
3771 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3773 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3776 role = ntohl(nrr->role);
3777 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3778 && role != NX_ROLE_SLAVE) {
3779 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3781 /* There's no good error code for this. */
3782 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3785 if (role == NX_ROLE_MASTER) {
3786 struct ofconn *other;
3788 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3789 &ofproto->controllers) {
3790 if (other->role == NX_ROLE_MASTER) {
3791 other->role = NX_ROLE_SLAVE;
3795 ofconn->role = role;
3797 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3799 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3800 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3801 reply->role = htonl(role);
3802 queue_tx(buf, ofconn, ofconn->reply_counter);
3808 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3810 struct ofp_vendor_header *ovh = msg;
3811 struct nicira_header *nh;
3813 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3814 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3815 "(expected at least %zu)",
3816 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3817 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3819 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3820 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3822 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3823 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3824 "(expected at least %zu)",
3825 ntohs(ovh->header.length), sizeof(struct nicira_header));
3826 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3830 switch (ntohl(nh->subtype)) {
3831 case NXT_STATUS_REQUEST:
3832 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3835 case NXT_TUN_ID_FROM_COOKIE:
3836 return handle_tun_id_from_cookie(p, msg);
3838 case NXT_ROLE_REQUEST:
3839 return handle_role_request(p, ofconn, msg);
3842 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3846 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3848 struct ofp_header *ob;
3851 /* Currently, everything executes synchronously, so we can just
3852 * immediately send the barrier reply. */
3853 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3854 queue_tx(buf, ofconn, ofconn->reply_counter);
3859 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3860 struct ofpbuf *ofp_msg)
3862 struct ofp_header *oh = ofp_msg->data;
3865 COVERAGE_INC(ofproto_recv_openflow);
3867 case OFPT_ECHO_REQUEST:
3868 error = handle_echo_request(ofconn, oh);
3871 case OFPT_ECHO_REPLY:
3875 case OFPT_FEATURES_REQUEST:
3876 error = handle_features_request(p, ofconn, oh);
3879 case OFPT_GET_CONFIG_REQUEST:
3880 error = handle_get_config_request(p, ofconn, oh);
3883 case OFPT_SET_CONFIG:
3884 error = handle_set_config(p, ofconn, ofp_msg->data);
3887 case OFPT_PACKET_OUT:
3888 error = handle_packet_out(p, ofconn, ofp_msg->data);
3892 error = handle_port_mod(p, ofconn, oh);
3896 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3899 case OFPT_STATS_REQUEST:
3900 error = handle_stats_request(p, ofconn, oh);
3904 error = handle_vendor(p, ofconn, ofp_msg->data);
3907 case OFPT_BARRIER_REQUEST:
3908 error = handle_barrier_request(ofconn, oh);
3912 if (VLOG_IS_WARN_ENABLED()) {
3913 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3914 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3917 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3922 send_error_oh(ofconn, ofp_msg->data, error);
3927 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3929 struct odp_msg *msg = packet->data;
3931 struct ofpbuf payload;
3934 payload.data = msg + 1;
3935 payload.size = msg->length - sizeof *msg;
3936 flow_extract(&payload, msg->arg, msg->port, &flow);
3938 /* Check with in-band control to see if this packet should be sent
3939 * to the local port regardless of the flow table. */
3940 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3941 union odp_action action;
3943 memset(&action, 0, sizeof(action));
3944 action.output.type = ODPAT_OUTPUT;
3945 action.output.port = ODPP_LOCAL;
3946 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3949 rule = lookup_valid_rule(p, &flow);
3951 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3952 struct ofport *port = port_array_get(&p->ports, msg->port);
3954 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3955 COVERAGE_INC(ofproto_no_packet_in);
3956 /* XXX install 'drop' flow entry */
3957 ofpbuf_delete(packet);
3961 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3964 COVERAGE_INC(ofproto_packet_in);
3965 send_packet_in(p, packet);
3969 if (rule->cr.wc.wildcards) {
3970 rule = rule_create_subrule(p, rule, &flow);
3971 rule_make_actions(p, rule, packet);
3973 if (!rule->may_install) {
3974 /* The rule is not installable, that is, we need to process every
3975 * packet, so process the current packet and set its actions into
3977 rule_make_actions(p, rule, packet);
3979 /* XXX revalidate rule if it needs it */
3983 rule_execute(p, rule, &payload, &flow);
3984 rule_reinstall(p, rule);
3986 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3988 * Extra-special case for fail-open mode.
3990 * We are in fail-open mode and the packet matched the fail-open rule,
3991 * but we are connected to a controller too. We should send the packet
3992 * up to the controller in the hope that it will try to set up a flow
3993 * and thereby allow us to exit fail-open.
3995 * See the top-level comment in fail-open.c for more information.
3997 send_packet_in(p, packet);
3999 ofpbuf_delete(packet);
4004 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4006 struct odp_msg *msg = packet->data;
4008 switch (msg->type) {
4009 case _ODPL_ACTION_NR:
4010 COVERAGE_INC(ofproto_ctlr_action);
4011 send_packet_in(p, packet);
4014 case _ODPL_SFLOW_NR:
4016 ofproto_sflow_received(p->sflow, msg);
4018 ofpbuf_delete(packet);
4022 handle_odp_miss_msg(p, packet);
4026 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4033 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4035 struct rule *sub = rule_from_cls_rule(sub_);
4036 struct revalidate_cbdata *cbdata = cbdata_;
4038 if (cbdata->revalidate_all
4039 || (cbdata->revalidate_subrules && sub->super)
4040 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4041 revalidate_rule(cbdata->ofproto, sub);
4046 revalidate_rule(struct ofproto *p, struct rule *rule)
4048 const flow_t *flow = &rule->cr.flow;
4050 COVERAGE_INC(ofproto_revalidate_rule);
4053 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4055 rule_remove(p, rule);
4057 } else if (super != rule->super) {
4058 COVERAGE_INC(ofproto_revalidate_moved);
4059 list_remove(&rule->list);
4060 list_push_back(&super->list, &rule->list);
4061 rule->super = super;
4062 rule->hard_timeout = super->hard_timeout;
4063 rule->idle_timeout = super->idle_timeout;
4064 rule->created = super->created;
4069 rule_update_actions(p, rule);
4073 static struct ofpbuf *
4074 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4075 long long int now, uint8_t reason)
4077 struct ofp_flow_removed *ofr;
4079 long long int tdiff = now - rule->created;
4080 uint32_t sec = tdiff / 1000;
4081 uint32_t msec = tdiff - (sec * 1000);
4083 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4084 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4086 ofr->cookie = rule->flow_cookie;
4087 ofr->priority = htons(rule->cr.priority);
4088 ofr->reason = reason;
4089 ofr->duration_sec = htonl(sec);
4090 ofr->duration_nsec = htonl(msec * 1000000);
4091 ofr->idle_timeout = htons(rule->idle_timeout);
4092 ofr->packet_count = htonll(rule->packet_count);
4093 ofr->byte_count = htonll(rule->byte_count);
4099 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4101 assert(rule->installed);
4102 assert(!rule->cr.wc.wildcards);
4105 rule_remove(ofproto, rule);
4107 rule_uninstall(ofproto, rule);
4112 send_flow_removed(struct ofproto *p, struct rule *rule,
4113 long long int now, uint8_t reason)
4115 struct ofconn *ofconn;
4116 struct ofconn *prev;
4117 struct ofpbuf *buf = NULL;
4119 /* We limit the maximum number of queued flow expirations it by accounting
4120 * them under the counter for replies. That works because preventing
4121 * OpenFlow requests from being processed also prevents new flows from
4122 * being added (and expiring). (It also prevents processing OpenFlow
4123 * requests that would not add new flows, so it is imperfect.) */
4126 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4127 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4128 && ofconn_receives_async_msgs(ofconn)) {
4130 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4132 buf = compose_flow_removed(p, rule, now, reason);
4138 queue_tx(buf, prev, prev->reply_counter);
4144 expire_rule(struct cls_rule *cls_rule, void *p_)
4146 struct ofproto *p = p_;
4147 struct rule *rule = rule_from_cls_rule(cls_rule);
4148 long long int hard_expire, idle_expire, expire, now;
4150 hard_expire = (rule->hard_timeout
4151 ? rule->created + rule->hard_timeout * 1000
4153 idle_expire = (rule->idle_timeout
4154 && (rule->super || list_is_empty(&rule->list))
4155 ? rule->used + rule->idle_timeout * 1000
4157 expire = MIN(hard_expire, idle_expire);
4161 if (rule->installed && now >= rule->used + 5000) {
4162 uninstall_idle_flow(p, rule);
4163 } else if (!rule->cr.wc.wildcards) {
4164 active_timeout(p, rule);
4170 COVERAGE_INC(ofproto_expired);
4172 /* Update stats. This code will be a no-op if the rule expired
4173 * due to an idle timeout. */
4174 if (rule->cr.wc.wildcards) {
4175 struct rule *subrule, *next;
4176 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4177 rule_remove(p, subrule);
4180 rule_uninstall(p, rule);
4183 if (!rule_is_hidden(rule)) {
4184 send_flow_removed(p, rule, now,
4186 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4188 rule_remove(p, rule);
4192 active_timeout(struct ofproto *ofproto, struct rule *rule)
4194 if (ofproto->netflow && !is_controller_rule(rule) &&
4195 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4196 struct ofexpired expired;
4197 struct odp_flow odp_flow;
4199 /* Get updated flow stats. */
4200 memset(&odp_flow, 0, sizeof odp_flow);
4201 if (rule->installed) {
4202 odp_flow.key = rule->cr.flow;
4203 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4204 dpif_flow_get(ofproto->dpif, &odp_flow);
4206 if (odp_flow.stats.n_packets) {
4207 update_time(ofproto, rule, &odp_flow.stats);
4208 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
4209 odp_flow.stats.tcp_flags);
4213 expired.flow = rule->cr.flow;
4214 expired.packet_count = rule->packet_count +
4215 odp_flow.stats.n_packets;
4216 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4217 expired.used = rule->used;
4219 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4221 /* Schedule us to send the accumulated records once we have
4222 * collected all of them. */
4223 poll_immediate_wake();
4228 update_used(struct ofproto *p)
4230 struct odp_flow *flows;
4235 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4240 for (i = 0; i < n_flows; i++) {
4241 struct odp_flow *f = &flows[i];
4244 rule = rule_from_cls_rule(
4245 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4246 if (!rule || !rule->installed) {
4247 COVERAGE_INC(ofproto_unexpected_rule);
4248 dpif_flow_del(p->dpif, f);
4252 update_time(p, rule, &f->stats);
4253 rule_account(p, rule, f->stats.n_bytes);
4258 /* pinsched callback for sending 'packet' on 'ofconn'. */
4260 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4262 struct ofconn *ofconn = ofconn_;
4264 rconn_send_with_limit(ofconn->rconn, packet,
4265 ofconn->packet_in_counter, 100);
4268 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4269 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4270 * packet scheduler for sending.
4272 * 'max_len' specifies the maximum number of bytes of the packet to send on
4273 * 'ofconn' (INT_MAX specifies no limit).
4275 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4276 * ownership is transferred to this function. */
4278 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4281 struct ofproto *ofproto = ofconn->ofproto;
4282 struct ofp_packet_in *opi = packet->data;
4283 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4284 int send_len, trim_size;
4288 if (opi->reason == OFPR_ACTION) {
4289 buffer_id = UINT32_MAX;
4290 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4291 buffer_id = pktbuf_get_null();
4292 } else if (!ofconn->pktbuf) {
4293 buffer_id = UINT32_MAX;
4295 struct ofpbuf payload;
4296 payload.data = opi->data;
4297 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4298 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4301 /* Figure out how much of the packet to send. */
4302 send_len = ntohs(opi->total_len);
4303 if (buffer_id != UINT32_MAX) {
4304 send_len = MIN(send_len, ofconn->miss_send_len);
4306 send_len = MIN(send_len, max_len);
4308 /* Adjust packet length and clone if necessary. */
4309 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4311 packet = ofpbuf_clone_data(packet->data, trim_size);
4314 packet->size = trim_size;
4317 /* Update packet headers. */
4318 opi->buffer_id = htonl(buffer_id);
4319 update_openflow_length(packet);
4321 /* Hand over to packet scheduler. It might immediately call into
4322 * do_send_packet_in() or it might buffer it for a while (until a later
4323 * call to pinsched_run()). */
4324 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4325 packet, do_send_packet_in, ofconn);
4328 /* Replace struct odp_msg header in 'packet' by equivalent struct
4329 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4330 * returned by dpif_recv()).
4332 * The conversion is not complete: the caller still needs to trim any unneeded
4333 * payload off the end of the buffer, set the length in the OpenFlow header,
4334 * and set buffer_id. Those require us to know the controller settings and so
4335 * must be done on a per-controller basis.
4337 * Returns the maximum number of bytes of the packet that should be sent to
4338 * the controller (INT_MAX if no limit). */
4340 do_convert_to_packet_in(struct ofpbuf *packet)
4342 struct odp_msg *msg = packet->data;
4343 struct ofp_packet_in *opi;
4349 /* Extract relevant header fields */
4350 if (msg->type == _ODPL_ACTION_NR) {
4351 reason = OFPR_ACTION;
4354 reason = OFPR_NO_MATCH;
4357 total_len = msg->length - sizeof *msg;
4358 in_port = odp_port_to_ofp_port(msg->port);
4360 /* Repurpose packet buffer by overwriting header. */
4361 ofpbuf_pull(packet, sizeof(struct odp_msg));
4362 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4363 opi->header.version = OFP_VERSION;
4364 opi->header.type = OFPT_PACKET_IN;
4365 opi->total_len = htons(total_len);
4366 opi->in_port = htons(in_port);
4367 opi->reason = reason;
4372 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4373 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4374 * as necessary according to their individual configurations.
4376 * 'packet' must have sufficient headroom to convert it into a struct
4377 * ofp_packet_in (e.g. as returned by dpif_recv()).
4379 * Takes ownership of 'packet'. */
4381 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4383 struct ofconn *ofconn, *prev;
4386 max_len = do_convert_to_packet_in(packet);
4389 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4390 if (ofconn_receives_async_msgs(ofconn)) {
4392 schedule_packet_in(prev, packet, max_len, true);
4398 schedule_packet_in(prev, packet, max_len, false);
4400 ofpbuf_delete(packet);
4405 pick_datapath_id(const struct ofproto *ofproto)
4407 const struct ofport *port;
4409 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4411 uint8_t ea[ETH_ADDR_LEN];
4414 error = netdev_get_etheraddr(port->netdev, ea);
4416 return eth_addr_to_uint64(ea);
4418 VLOG_WARN("could not get MAC address for %s (%s)",
4419 netdev_get_name(port->netdev), strerror(error));
4421 return ofproto->fallback_dpid;
4425 pick_fallback_dpid(void)
4427 uint8_t ea[ETH_ADDR_LEN];
4428 eth_addr_nicira_random(ea);
4429 return eth_addr_to_uint64(ea);
4433 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4434 struct odp_actions *actions, tag_type *tags,
4435 uint16_t *nf_output_iface, void *ofproto_)
4437 struct ofproto *ofproto = ofproto_;
4440 /* Drop frames for reserved multicast addresses. */
4441 if (eth_addr_is_reserved(flow->dl_dst)) {
4445 /* Learn source MAC (but don't try to learn from revalidation). */
4446 if (packet != NULL) {
4447 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4449 GRAT_ARP_LOCK_NONE);
4451 /* The log messages here could actually be useful in debugging,
4452 * so keep the rate limit relatively high. */
4453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4454 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4455 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4456 ofproto_revalidate(ofproto, rev_tag);
4460 /* Determine output port. */
4461 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4464 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4465 } else if (out_port != flow->in_port) {
4466 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4467 *nf_output_iface = out_port;
4475 static const struct ofhooks default_ofhooks = {
4477 default_normal_ofhook_cb,