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"
53 #include "stream-ssl.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto)
64 #include "sflow_api.h"
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_set_up_flow, uint16_t *nf_output_iface);
88 uint64_t flow_cookie; /* Controller-issued identifier.
89 (Kept in network-byte order.) */
90 uint16_t idle_timeout; /* In seconds from time of last use. */
91 uint16_t hard_timeout; /* In seconds from time of creation. */
92 bool send_flow_removed; /* Send a flow removed message? */
93 long long int used; /* Last-used time (0 if never used). */
94 long long int created; /* Creation time. */
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
97 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
98 tag_type tags; /* Tags (set only by hooks). */
99 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
101 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
102 * exact-match rule (having cr.wc.wildcards of 0) generated from the
103 * wildcard rule 'super'. In this case, 'list' is an element of the
106 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
107 * a list of subrules. A super-rule with no wildcards (where
108 * cr.wc.wildcards is 0) will never have any subrules. */
114 * 'n_actions' is the number of elements in the 'actions' array. A single
115 * action may take up more more than one element's worth of space.
117 * A subrule has no actions (it uses the super-rule's actions). */
119 union ofp_action *actions;
123 * A super-rule with wildcard fields never has ODP actions (since the
124 * datapath only supports exact-match flows). */
125 bool installed; /* Installed in datapath? */
126 bool may_install; /* True ordinarily; false if actions must
127 * be reassessed for every packet. */
129 union odp_action *odp_actions;
133 rule_is_hidden(const struct rule *rule)
135 /* Subrules are merely an implementation detail, so hide them from the
137 if (rule->super != NULL) {
141 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
142 * (e.g. by in-band control) and are intentionally hidden from the
144 if (rule->cr.priority > UINT16_MAX) {
151 static struct rule *rule_create(struct ofproto *, struct rule *super,
152 const union ofp_action *, size_t n_actions,
153 uint16_t idle_timeout, uint16_t hard_timeout,
154 uint64_t flow_cookie, bool send_flow_removed);
155 static void rule_free(struct rule *);
156 static void rule_destroy(struct ofproto *, struct rule *);
157 static struct rule *rule_from_cls_rule(const struct cls_rule *);
158 static void rule_insert(struct ofproto *, struct rule *,
159 struct ofpbuf *packet, uint16_t in_port);
160 static void rule_remove(struct ofproto *, struct rule *);
161 static bool rule_make_actions(struct ofproto *, struct rule *,
162 const struct ofpbuf *packet);
163 static void rule_install(struct ofproto *, struct rule *,
164 struct rule *displaced_rule);
165 static void rule_uninstall(struct ofproto *, struct rule *);
166 static void rule_post_uninstall(struct ofproto *, struct rule *);
167 static void send_flow_removed(struct ofproto *p, struct rule *rule,
168 long long int now, uint8_t reason);
170 /* ofproto supports two kinds of OpenFlow connections:
172 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
173 * maintains persistent connections to these controllers and by default
174 * sends them asynchronous messages such as packet-ins.
176 * - "Service" connections, e.g. from ovs-ofctl. When these connections
177 * drop, it is the other side's responsibility to reconnect them if
178 * necessary. ofproto does not send them asynchronous messages by default.
180 * Currently, active (tcp, ssl, unix) connections are always "primary"
181 * connections and passive (ptcp, pssl, punix) connections are always "service"
182 * connections. There is no inherent reason for this, but it reflects the
186 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
187 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
190 /* A listener for incoming OpenFlow "service" connections. */
192 struct hmap_node node; /* In struct ofproto's "services" hmap. */
193 struct pvconn *pvconn; /* OpenFlow connection listener. */
195 /* These are not used by ofservice directly. They are settings for
196 * accepted "struct ofconn"s from the pvconn. */
197 int probe_interval; /* Max idle time before probing, in seconds. */
198 int rate_limit; /* Max packet-in rate in packets per second. */
199 int burst_limit; /* Limit on accumulating packet credits. */
202 static struct ofservice *ofservice_lookup(struct ofproto *,
204 static int ofservice_create(struct ofproto *,
205 const struct ofproto_controller *);
206 static void ofservice_reconfigure(struct ofservice *,
207 const struct ofproto_controller *);
208 static void ofservice_destroy(struct ofproto *, struct ofservice *);
210 /* An OpenFlow connection. */
212 struct ofproto *ofproto; /* The ofproto that owns this connection. */
213 struct list node; /* In struct ofproto's "all_conns" list. */
214 struct rconn *rconn; /* OpenFlow connection. */
215 enum ofconn_type type; /* Type. */
217 /* OFPT_PACKET_IN related data. */
218 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
219 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
220 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
221 int miss_send_len; /* Bytes to send of buffered packets. */
223 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
224 * requests, and the maximum number before we stop reading OpenFlow
226 #define OFCONN_REPLY_MAX 100
227 struct rconn_packet_counter *reply_counter;
229 /* type == OFCONN_PRIMARY only. */
230 enum nx_role role; /* Role. */
231 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
232 struct discovery *discovery; /* Controller discovery object, if enabled. */
233 struct status_category *ss; /* Switch status category. */
234 enum ofproto_band band; /* In-band or out-of-band? */
237 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
238 * "schedulers" array. Their values are 0 and 1, and their meanings and values
239 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
240 * case anything ever changes, check their values here. */
241 #define N_SCHEDULERS 2
242 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
243 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
244 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
245 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
247 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
249 static void ofconn_destroy(struct ofconn *);
250 static void ofconn_run(struct ofconn *, struct ofproto *);
251 static void ofconn_wait(struct ofconn *);
252 static bool ofconn_receives_async_msgs(const struct ofconn *);
253 static char *ofconn_make_name(const struct ofproto *, const char *target);
254 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
256 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
257 struct rconn_packet_counter *counter);
259 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
260 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
264 uint64_t datapath_id; /* Datapath ID. */
265 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
266 char *mfr_desc; /* Manufacturer. */
267 char *hw_desc; /* Hardware. */
268 char *sw_desc; /* Software version. */
269 char *serial_desc; /* Serial number. */
270 char *dp_desc; /* Datapath description. */
274 struct netdev_monitor *netdev_monitor;
275 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
277 struct shash port_by_name;
281 struct switch_status *switch_status;
282 struct fail_open *fail_open;
283 struct netflow *netflow;
284 struct ofproto_sflow *sflow;
286 /* In-band control. */
287 struct in_band *in_band;
288 long long int next_in_band_update;
289 struct sockaddr_in *extra_in_band_remotes;
290 size_t n_extra_remotes;
293 struct classifier cls;
294 bool need_revalidate;
295 long long int next_expiration;
296 struct tag_set revalidate_set;
297 bool tun_id_from_cookie;
299 /* OpenFlow connections. */
300 struct hmap controllers; /* Controller "struct ofconn"s. */
301 struct list all_conns; /* Contains "struct ofconn"s. */
302 enum ofproto_fail_mode fail_mode;
304 /* OpenFlow listeners. */
305 struct hmap services; /* Contains "struct ofservice"s. */
306 struct pvconn **snoops;
309 /* Hooks for ovs-vswitchd. */
310 const struct ofhooks *ofhooks;
313 /* Used by default ofhooks. */
314 struct mac_learning *ml;
317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
319 static const struct ofhooks default_ofhooks;
321 static uint64_t pick_datapath_id(const struct ofproto *);
322 static uint64_t pick_fallback_dpid(void);
324 static int ofproto_expire(struct ofproto *);
326 static void update_stats(struct ofproto *, struct rule *,
327 const struct odp_flow_stats *);
328 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
329 static void revalidate_cb(struct cls_rule *rule_, void *p_);
331 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
333 static void handle_openflow(struct ofconn *, struct ofproto *,
336 static void refresh_port_groups(struct ofproto *);
338 static void update_port(struct ofproto *, const char *devname);
339 static int init_ports(struct ofproto *);
340 static void reinit_ports(struct ofproto *);
343 ofproto_create(const char *datapath, const char *datapath_type,
344 const struct ofhooks *ofhooks, void *aux,
345 struct ofproto **ofprotop)
347 struct odp_stats stats;
354 /* Connect to datapath and start listening for messages. */
355 error = dpif_open(datapath, datapath_type, &dpif);
357 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
360 error = dpif_get_dp_stats(dpif, &stats);
362 VLOG_ERR("failed to obtain stats for datapath %s: %s",
363 datapath, strerror(error));
367 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
369 VLOG_ERR("failed to listen on datapath %s: %s",
370 datapath, strerror(error));
374 dpif_flow_flush(dpif);
375 dpif_recv_purge(dpif);
377 /* Initialize settings. */
378 p = xzalloc(sizeof *p);
379 p->fallback_dpid = pick_fallback_dpid();
380 p->datapath_id = p->fallback_dpid;
381 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
382 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
383 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
384 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
385 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
387 /* Initialize datapath. */
389 p->netdev_monitor = netdev_monitor_create();
390 port_array_init(&p->ports);
391 shash_init(&p->port_by_name);
392 p->max_ports = stats.max_ports;
394 /* Initialize submodules. */
395 p->switch_status = switch_status_create(p);
401 /* Initialize flow table. */
402 classifier_init(&p->cls);
403 p->need_revalidate = false;
404 p->next_expiration = time_msec() + 1000;
405 tag_set_init(&p->revalidate_set);
407 /* Initialize OpenFlow connections. */
408 list_init(&p->all_conns);
409 hmap_init(&p->controllers);
410 hmap_init(&p->services);
414 /* Initialize hooks. */
416 p->ofhooks = ofhooks;
420 p->ofhooks = &default_ofhooks;
422 p->ml = mac_learning_create();
425 /* Pick final datapath ID. */
426 p->datapath_id = pick_datapath_id(p);
427 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
434 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
436 uint64_t old_dpid = p->datapath_id;
437 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
438 if (p->datapath_id != old_dpid) {
439 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
441 /* Force all active connections to reconnect, since there is no way to
442 * notify a controller that the datapath ID has changed. */
443 ofproto_reconnect_controllers(p);
448 is_discovery_controller(const struct ofproto_controller *c)
450 return !strcmp(c->target, "discover");
454 is_in_band_controller(const struct ofproto_controller *c)
456 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
459 /* Creates a new controller in 'ofproto'. Some of the settings are initially
460 * drawn from 'c', but update_controller() needs to be called later to finish
461 * the new ofconn's configuration. */
463 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
465 struct discovery *discovery;
466 struct ofconn *ofconn;
468 if (is_discovery_controller(c)) {
469 int error = discovery_create(c->accept_re, c->update_resolv_conf,
470 ofproto->dpif, ofproto->switch_status,
479 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
480 ofconn->pktbuf = pktbuf_create();
481 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
483 ofconn->discovery = discovery;
485 char *name = ofconn_make_name(ofproto, c->target);
486 rconn_connect(ofconn->rconn, c->target, name);
489 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
490 hash_string(c->target, 0));
493 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
494 * target or turn discovery on or off (these are done by creating new ofconns
495 * and deleting old ones), but it can update the rest of an ofconn's
498 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
502 ofconn->band = (is_in_band_controller(c)
503 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
505 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
507 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
508 rconn_set_probe_interval(ofconn->rconn, probe_interval);
510 if (ofconn->discovery) {
511 discovery_set_update_resolv_conf(ofconn->discovery,
512 c->update_resolv_conf);
513 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
516 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
520 ofconn_get_target(const struct ofconn *ofconn)
522 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
525 static struct ofconn *
526 find_controller_by_target(struct ofproto *ofproto, const char *target)
528 struct ofconn *ofconn;
530 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
531 hash_string(target, 0), &ofproto->controllers) {
532 if (!strcmp(ofconn_get_target(ofconn), target)) {
540 update_in_band_remotes(struct ofproto *ofproto)
542 const struct ofconn *ofconn;
543 struct sockaddr_in *addrs;
544 size_t max_addrs, n_addrs;
548 /* Allocate enough memory for as many remotes as we could possibly have. */
549 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
550 addrs = xmalloc(max_addrs * sizeof *addrs);
553 /* Add all the remotes. */
555 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
556 struct sockaddr_in *sin = &addrs[n_addrs];
558 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
562 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
563 if (sin->sin_addr.s_addr) {
564 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
567 if (ofconn->discovery) {
571 for (i = 0; i < ofproto->n_extra_remotes; i++) {
572 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
575 /* Create or update or destroy in-band.
577 * Ordinarily we only enable in-band if there's at least one remote
578 * address, but discovery needs the in-band rules for DHCP to be installed
579 * even before we know any remote addresses. */
580 if (n_addrs || discovery) {
581 if (!ofproto->in_band) {
582 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
585 if (ofproto->in_band) {
586 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
588 ofproto->next_in_band_update = time_msec() + 1000;
590 in_band_destroy(ofproto->in_band);
591 ofproto->in_band = NULL;
599 update_fail_open(struct ofproto *p)
601 struct ofconn *ofconn;
603 if (!hmap_is_empty(&p->controllers)
604 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
605 struct rconn **rconns;
609 p->fail_open = fail_open_create(p, p->switch_status);
613 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
614 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
615 rconns[n++] = ofconn->rconn;
618 fail_open_set_controllers(p->fail_open, rconns, n);
619 /* p->fail_open takes ownership of 'rconns'. */
621 fail_open_destroy(p->fail_open);
627 ofproto_set_controllers(struct ofproto *p,
628 const struct ofproto_controller *controllers,
629 size_t n_controllers)
631 struct shash new_controllers;
632 struct ofconn *ofconn, *next_ofconn;
633 struct ofservice *ofservice, *next_ofservice;
637 /* Create newly configured controllers and services.
638 * Create a name to ofproto_controller mapping in 'new_controllers'. */
639 shash_init(&new_controllers);
640 for (i = 0; i < n_controllers; i++) {
641 const struct ofproto_controller *c = &controllers[i];
643 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
644 if (!find_controller_by_target(p, c->target)) {
645 add_controller(p, c);
647 } else if (!pvconn_verify_name(c->target)) {
648 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
652 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
653 dpif_name(p->dpif), c->target);
657 shash_add_once(&new_controllers, c->target, &controllers[i]);
660 /* Delete controllers that are no longer configured.
661 * Update configuration of all now-existing controllers. */
663 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node,
665 struct ofproto_controller *c;
667 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
669 ofconn_destroy(ofconn);
671 update_controller(ofconn, c);
678 /* Delete services that are no longer configured.
679 * Update configuration of all now-existing services. */
680 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
682 struct ofproto_controller *c;
684 c = shash_find_data(&new_controllers,
685 pvconn_get_name(ofservice->pvconn));
687 ofservice_destroy(p, ofservice);
689 ofservice_reconfigure(ofservice, c);
693 shash_destroy(&new_controllers);
695 update_in_band_remotes(p);
698 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
699 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
700 struct ofconn, hmap_node);
701 ofconn->ss = switch_status_register(p->switch_status, "remote",
702 rconn_status_cb, ofconn->rconn);
707 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
709 p->fail_mode = fail_mode;
713 /* Drops the connections between 'ofproto' and all of its controllers, forcing
714 * them to reconnect. */
716 ofproto_reconnect_controllers(struct ofproto *ofproto)
718 struct ofconn *ofconn;
720 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
721 rconn_reconnect(ofconn->rconn);
726 any_extras_changed(const struct ofproto *ofproto,
727 const struct sockaddr_in *extras, size_t n)
731 if (n != ofproto->n_extra_remotes) {
735 for (i = 0; i < n; i++) {
736 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
737 const struct sockaddr_in *new = &extras[i];
739 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
740 old->sin_port != new->sin_port) {
748 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
749 * in-band control should guarantee access, in the same way that in-band
750 * control guarantees access to OpenFlow controllers. */
752 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
753 const struct sockaddr_in *extras, size_t n)
755 if (!any_extras_changed(ofproto, extras, n)) {
759 free(ofproto->extra_in_band_remotes);
760 ofproto->n_extra_remotes = n;
761 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
763 update_in_band_remotes(ofproto);
767 ofproto_set_desc(struct ofproto *p,
768 const char *mfr_desc, const char *hw_desc,
769 const char *sw_desc, const char *serial_desc,
772 struct ofp_desc_stats *ods;
775 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
776 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
777 sizeof ods->mfr_desc);
780 p->mfr_desc = xstrdup(mfr_desc);
783 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
784 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
785 sizeof ods->hw_desc);
788 p->hw_desc = xstrdup(hw_desc);
791 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
792 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
793 sizeof ods->sw_desc);
796 p->sw_desc = xstrdup(sw_desc);
799 if (strlen(serial_desc) >= sizeof ods->serial_num) {
800 VLOG_WARN("truncating serial_desc, must be less than %zu "
802 sizeof ods->serial_num);
804 free(p->serial_desc);
805 p->serial_desc = xstrdup(serial_desc);
808 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
809 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
810 sizeof ods->dp_desc);
813 p->dp_desc = xstrdup(dp_desc);
818 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
819 const struct svec *svec)
821 struct pvconn **pvconns = *pvconnsp;
822 size_t n_pvconns = *n_pvconnsp;
826 for (i = 0; i < n_pvconns; i++) {
827 pvconn_close(pvconns[i]);
831 pvconns = xmalloc(svec->n * sizeof *pvconns);
833 for (i = 0; i < svec->n; i++) {
834 const char *name = svec->names[i];
835 struct pvconn *pvconn;
838 error = pvconn_open(name, &pvconn);
840 pvconns[n_pvconns++] = pvconn;
842 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
850 *n_pvconnsp = n_pvconns;
856 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
858 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
862 ofproto_set_netflow(struct ofproto *ofproto,
863 const struct netflow_options *nf_options)
865 if (nf_options && nf_options->collectors.n) {
866 if (!ofproto->netflow) {
867 ofproto->netflow = netflow_create();
869 return netflow_set_options(ofproto->netflow, nf_options);
871 netflow_destroy(ofproto->netflow);
872 ofproto->netflow = NULL;
878 ofproto_set_sflow(struct ofproto *ofproto,
879 const struct ofproto_sflow_options *oso)
881 struct ofproto_sflow *os = ofproto->sflow;
884 struct ofport *ofport;
885 unsigned int odp_port;
887 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
888 refresh_port_groups(ofproto);
889 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
890 ofproto_sflow_add_port(os, odp_port,
891 netdev_get_name(ofport->netdev));
894 ofproto_sflow_set_options(os, oso);
896 ofproto_sflow_destroy(os);
897 ofproto->sflow = NULL;
902 ofproto_get_datapath_id(const struct ofproto *ofproto)
904 return ofproto->datapath_id;
908 ofproto_has_primary_controller(const struct ofproto *ofproto)
910 return !hmap_is_empty(&ofproto->controllers);
913 enum ofproto_fail_mode
914 ofproto_get_fail_mode(const struct ofproto *p)
920 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
924 for (i = 0; i < ofproto->n_snoops; i++) {
925 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
930 ofproto_destroy(struct ofproto *p)
932 struct ofservice *ofservice, *next_ofservice;
933 struct ofconn *ofconn, *next_ofconn;
934 struct ofport *ofport;
935 unsigned int port_no;
942 /* Destroy fail-open and in-band early, since they touch the classifier. */
943 fail_open_destroy(p->fail_open);
946 in_band_destroy(p->in_band);
948 free(p->extra_in_band_remotes);
950 ofproto_flush_flows(p);
951 classifier_destroy(&p->cls);
953 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
955 ofconn_destroy(ofconn);
957 hmap_destroy(&p->controllers);
960 netdev_monitor_destroy(p->netdev_monitor);
961 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
964 shash_destroy(&p->port_by_name);
966 switch_status_destroy(p->switch_status);
967 netflow_destroy(p->netflow);
968 ofproto_sflow_destroy(p->sflow);
970 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
972 ofservice_destroy(p, ofservice);
974 hmap_destroy(&p->services);
976 for (i = 0; i < p->n_snoops; i++) {
977 pvconn_close(p->snoops[i]);
981 mac_learning_destroy(p->ml);
986 free(p->serial_desc);
989 port_array_destroy(&p->ports);
995 ofproto_run(struct ofproto *p)
997 int error = ofproto_run1(p);
999 error = ofproto_run2(p, false);
1005 process_port_change(struct ofproto *ofproto, int error, char *devname)
1007 if (error == ENOBUFS) {
1008 reinit_ports(ofproto);
1009 } else if (!error) {
1010 update_port(ofproto, devname);
1015 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1016 * means that 'ofconn' is more interesting for monitoring than a lower return
1019 snoop_preference(const struct ofconn *ofconn)
1021 switch (ofconn->role) {
1022 case NX_ROLE_MASTER:
1029 /* Shouldn't happen. */
1034 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1035 * Connects this vconn to a controller. */
1037 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1039 struct ofconn *ofconn, *best;
1041 /* Pick a controller for monitoring. */
1043 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1044 if (ofconn->type == OFCONN_PRIMARY
1045 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1051 rconn_add_monitor(best->rconn, vconn);
1053 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1059 ofproto_run1(struct ofproto *p)
1061 struct ofconn *ofconn, *next_ofconn;
1062 struct ofservice *ofservice;
1067 if (shash_is_empty(&p->port_by_name)) {
1071 for (i = 0; i < 50; i++) {
1075 error = dpif_recv(p->dpif, &buf);
1077 if (error == ENODEV) {
1078 /* Someone destroyed the datapath behind our back. The caller
1079 * better destroy us and give up, because we're just going to
1080 * spin from here on out. */
1081 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1082 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1083 dpif_name(p->dpif));
1089 handle_odp_msg(p, buf);
1092 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1093 process_port_change(p, error, devname);
1095 while ((error = netdev_monitor_poll(p->netdev_monitor,
1096 &devname)) != EAGAIN) {
1097 process_port_change(p, error, devname);
1101 if (time_msec() >= p->next_in_band_update) {
1102 update_in_band_remotes(p);
1104 in_band_run(p->in_band);
1107 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1109 ofconn_run(ofconn, p);
1112 /* Fail-open maintenance. Do this after processing the ofconns since
1113 * fail-open checks the status of the controller rconn. */
1115 fail_open_run(p->fail_open);
1118 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1119 struct vconn *vconn;
1122 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1124 struct ofconn *ofconn;
1125 struct rconn *rconn;
1128 rconn = rconn_create(ofservice->probe_interval, 0);
1129 name = ofconn_make_name(p, vconn_get_name(vconn));
1130 rconn_connect_unreliably(rconn, vconn, name);
1133 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1134 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1135 ofservice->burst_limit);
1136 } else if (retval != EAGAIN) {
1137 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1141 for (i = 0; i < p->n_snoops; i++) {
1142 struct vconn *vconn;
1145 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1147 add_snooper(p, vconn);
1148 } else if (retval != EAGAIN) {
1149 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1153 if (time_msec() >= p->next_expiration) {
1154 int delay = ofproto_expire(p);
1155 p->next_expiration = time_msec() + delay;
1156 COVERAGE_INC(ofproto_expiration);
1160 netflow_run(p->netflow);
1163 ofproto_sflow_run(p->sflow);
1169 struct revalidate_cbdata {
1170 struct ofproto *ofproto;
1171 bool revalidate_all; /* Revalidate all exact-match rules? */
1172 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1173 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1177 ofproto_run2(struct ofproto *p, bool revalidate_all)
1179 if (p->need_revalidate || revalidate_all
1180 || !tag_set_is_empty(&p->revalidate_set)) {
1181 struct revalidate_cbdata cbdata;
1183 cbdata.revalidate_all = revalidate_all;
1184 cbdata.revalidate_subrules = p->need_revalidate;
1185 cbdata.revalidate_set = p->revalidate_set;
1186 tag_set_init(&p->revalidate_set);
1187 COVERAGE_INC(ofproto_revalidate);
1188 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1189 p->need_revalidate = false;
1196 ofproto_wait(struct ofproto *p)
1198 struct ofservice *ofservice;
1199 struct ofconn *ofconn;
1202 dpif_recv_wait(p->dpif);
1203 dpif_port_poll_wait(p->dpif);
1204 netdev_monitor_poll_wait(p->netdev_monitor);
1205 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1206 ofconn_wait(ofconn);
1209 poll_timer_wait_until(p->next_in_band_update);
1210 in_band_wait(p->in_band);
1213 fail_open_wait(p->fail_open);
1216 ofproto_sflow_wait(p->sflow);
1218 if (!tag_set_is_empty(&p->revalidate_set)) {
1219 poll_immediate_wake();
1221 if (p->need_revalidate) {
1222 /* Shouldn't happen, but if it does just go around again. */
1223 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1224 poll_immediate_wake();
1225 } else if (p->next_expiration != LLONG_MAX) {
1226 poll_timer_wait_until(p->next_expiration);
1228 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1229 pvconn_wait(ofservice->pvconn);
1231 for (i = 0; i < p->n_snoops; i++) {
1232 pvconn_wait(p->snoops[i]);
1237 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1239 tag_set_add(&ofproto->revalidate_set, tag);
1243 ofproto_get_revalidate_set(struct ofproto *ofproto)
1245 return &ofproto->revalidate_set;
1249 ofproto_is_alive(const struct ofproto *p)
1251 return !hmap_is_empty(&p->controllers);
1255 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1256 const union ofp_action *actions, size_t n_actions,
1257 const struct ofpbuf *packet)
1259 struct odp_actions odp_actions;
1262 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1268 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1270 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1271 odp_actions.n_actions, packet);
1276 ofproto_add_flow(struct ofproto *p,
1277 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1278 const union ofp_action *actions, size_t n_actions,
1282 rule = rule_create(p, NULL, actions, n_actions,
1283 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1285 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1286 rule_insert(p, rule, NULL, 0);
1290 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1291 uint32_t wildcards, unsigned int priority)
1295 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1299 rule_remove(ofproto, rule);
1304 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1306 struct rule *rule = rule_from_cls_rule(rule_);
1307 struct ofproto *ofproto = ofproto_;
1309 /* Mark the flow as not installed, even though it might really be
1310 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1311 * There is no point in uninstalling it individually since we are about to
1312 * blow away all the flows with dpif_flow_flush(). */
1313 rule->installed = false;
1315 rule_remove(ofproto, rule);
1319 ofproto_flush_flows(struct ofproto *ofproto)
1321 COVERAGE_INC(ofproto_flush);
1322 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1323 dpif_flow_flush(ofproto->dpif);
1324 if (ofproto->in_band) {
1325 in_band_flushed(ofproto->in_band);
1327 if (ofproto->fail_open) {
1328 fail_open_flushed(ofproto->fail_open);
1333 reinit_ports(struct ofproto *p)
1335 struct svec devnames;
1336 struct ofport *ofport;
1337 unsigned int port_no;
1338 struct odp_port *odp_ports;
1342 svec_init(&devnames);
1343 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1344 svec_add (&devnames, (char *) ofport->opp.name);
1346 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1347 for (i = 0; i < n_odp_ports; i++) {
1348 svec_add (&devnames, odp_ports[i].devname);
1352 svec_sort_unique(&devnames);
1353 for (i = 0; i < devnames.n; i++) {
1354 update_port(p, devnames.names[i]);
1356 svec_destroy(&devnames);
1360 refresh_port_group(struct ofproto *p, unsigned int group)
1364 struct ofport *port;
1365 unsigned int port_no;
1367 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1369 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1371 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1372 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1373 ports[n_ports++] = port_no;
1376 dpif_port_group_set(p->dpif, group, ports, n_ports);
1383 refresh_port_groups(struct ofproto *p)
1385 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1386 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1388 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1392 static struct ofport *
1393 make_ofport(const struct odp_port *odp_port)
1395 struct netdev_options netdev_options;
1396 enum netdev_flags flags;
1397 struct ofport *ofport;
1398 struct netdev *netdev;
1402 memset(&netdev_options, 0, sizeof netdev_options);
1403 netdev_options.name = odp_port->devname;
1404 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1406 error = netdev_open(&netdev_options, &netdev);
1408 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1409 "cannot be opened (%s)",
1410 odp_port->devname, odp_port->port,
1411 odp_port->devname, strerror(error));
1415 ofport = xmalloc(sizeof *ofport);
1416 ofport->netdev = netdev;
1417 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1418 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1419 memcpy(ofport->opp.name, odp_port->devname,
1420 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1421 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1423 netdev_get_flags(netdev, &flags);
1424 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1426 netdev_get_carrier(netdev, &carrier);
1427 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1429 netdev_get_features(netdev,
1430 &ofport->opp.curr, &ofport->opp.advertised,
1431 &ofport->opp.supported, &ofport->opp.peer);
1436 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1438 if (port_array_get(&p->ports, odp_port->port)) {
1439 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1442 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1443 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1452 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1454 const struct ofp_phy_port *a = &a_->opp;
1455 const struct ofp_phy_port *b = &b_->opp;
1457 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1458 return (a->port_no == b->port_no
1459 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1460 && !strcmp((char *) a->name, (char *) b->name)
1461 && a->state == b->state
1462 && a->config == b->config
1463 && a->curr == b->curr
1464 && a->advertised == b->advertised
1465 && a->supported == b->supported
1466 && a->peer == b->peer);
1470 send_port_status(struct ofproto *p, const struct ofport *ofport,
1473 /* XXX Should limit the number of queued port status change messages. */
1474 struct ofconn *ofconn;
1475 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1476 struct ofp_port_status *ops;
1479 if (!ofconn_receives_async_msgs(ofconn)) {
1483 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1484 ops->reason = reason;
1485 ops->desc = ofport->opp;
1486 hton_ofp_phy_port(&ops->desc);
1487 queue_tx(b, ofconn, NULL);
1489 if (p->ofhooks->port_changed_cb) {
1490 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1495 ofport_install(struct ofproto *p, struct ofport *ofport)
1497 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1498 const char *netdev_name = (const char *) ofport->opp.name;
1500 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1501 port_array_set(&p->ports, odp_port, ofport);
1502 shash_add(&p->port_by_name, netdev_name, ofport);
1504 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1509 ofport_remove(struct ofproto *p, struct ofport *ofport)
1511 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1513 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1514 port_array_delete(&p->ports, odp_port);
1515 shash_delete(&p->port_by_name,
1516 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1518 ofproto_sflow_del_port(p->sflow, odp_port);
1523 ofport_free(struct ofport *ofport)
1526 netdev_close(ofport->netdev);
1532 update_port(struct ofproto *p, const char *devname)
1534 struct odp_port odp_port;
1535 struct ofport *old_ofport;
1536 struct ofport *new_ofport;
1539 COVERAGE_INC(ofproto_update_port);
1541 /* Query the datapath for port information. */
1542 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1544 /* Find the old ofport. */
1545 old_ofport = shash_find_data(&p->port_by_name, devname);
1548 /* There's no port named 'devname' but there might be a port with
1549 * the same port number. This could happen if a port is deleted
1550 * and then a new one added in its place very quickly, or if a port
1551 * is renamed. In the former case we want to send an OFPPR_DELETE
1552 * and an OFPPR_ADD, and in the latter case we want to send a
1553 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1554 * the old port's ifindex against the new port, or perhaps less
1555 * reliably but more portably by comparing the old port's MAC
1556 * against the new port's MAC. However, this code isn't that smart
1557 * and always sends an OFPPR_MODIFY (XXX). */
1558 old_ofport = port_array_get(&p->ports, odp_port.port);
1560 } else if (error != ENOENT && error != ENODEV) {
1561 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1562 "%s", strerror(error));
1566 /* Create a new ofport. */
1567 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1569 /* Eliminate a few pathological cases. */
1570 if (!old_ofport && !new_ofport) {
1572 } else if (old_ofport && new_ofport) {
1573 /* Most of the 'config' bits are OpenFlow soft state, but
1574 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1575 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1576 * leaves the other bits 0.) */
1577 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1579 if (ofport_equal(old_ofport, new_ofport)) {
1580 /* False alarm--no change. */
1581 ofport_free(new_ofport);
1586 /* Now deal with the normal cases. */
1588 ofport_remove(p, old_ofport);
1591 ofport_install(p, new_ofport);
1593 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1594 (!old_ofport ? OFPPR_ADD
1595 : !new_ofport ? OFPPR_DELETE
1597 ofport_free(old_ofport);
1599 /* Update port groups. */
1600 refresh_port_groups(p);
1604 init_ports(struct ofproto *p)
1606 struct odp_port *ports;
1611 error = dpif_port_list(p->dpif, &ports, &n_ports);
1616 for (i = 0; i < n_ports; i++) {
1617 const struct odp_port *odp_port = &ports[i];
1618 if (!ofport_conflicts(p, odp_port)) {
1619 struct ofport *ofport = make_ofport(odp_port);
1621 ofport_install(p, ofport);
1626 refresh_port_groups(p);
1630 static struct ofconn *
1631 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1633 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1634 ofconn->ofproto = p;
1635 list_push_back(&p->all_conns, &ofconn->node);
1636 ofconn->rconn = rconn;
1637 ofconn->type = type;
1638 ofconn->role = NX_ROLE_OTHER;
1639 ofconn->packet_in_counter = rconn_packet_counter_create ();
1640 ofconn->pktbuf = NULL;
1641 ofconn->miss_send_len = 0;
1642 ofconn->reply_counter = rconn_packet_counter_create ();
1647 ofconn_destroy(struct ofconn *ofconn)
1649 if (ofconn->type == OFCONN_PRIMARY) {
1650 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1652 discovery_destroy(ofconn->discovery);
1654 list_remove(&ofconn->node);
1655 switch_status_unregister(ofconn->ss);
1656 rconn_destroy(ofconn->rconn);
1657 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1658 rconn_packet_counter_destroy(ofconn->reply_counter);
1659 pktbuf_destroy(ofconn->pktbuf);
1664 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1669 if (ofconn->discovery) {
1670 char *controller_name;
1671 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1672 discovery_question_connectivity(ofconn->discovery);
1674 if (discovery_run(ofconn->discovery, &controller_name)) {
1675 if (controller_name) {
1676 char *ofconn_name = ofconn_make_name(p, controller_name);
1677 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1680 rconn_disconnect(ofconn->rconn);
1685 for (i = 0; i < N_SCHEDULERS; i++) {
1686 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1689 rconn_run(ofconn->rconn);
1691 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1692 /* Limit the number of iterations to prevent other tasks from
1694 for (iteration = 0; iteration < 50; iteration++) {
1695 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1700 fail_open_maybe_recover(p->fail_open);
1702 handle_openflow(ofconn, p, of_msg);
1703 ofpbuf_delete(of_msg);
1707 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1708 ofconn_destroy(ofconn);
1713 ofconn_wait(struct ofconn *ofconn)
1717 if (ofconn->discovery) {
1718 discovery_wait(ofconn->discovery);
1720 for (i = 0; i < N_SCHEDULERS; i++) {
1721 pinsched_wait(ofconn->schedulers[i]);
1723 rconn_run_wait(ofconn->rconn);
1724 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1725 rconn_recv_wait(ofconn->rconn);
1727 COVERAGE_INC(ofproto_ofconn_stuck);
1731 /* Returns true if 'ofconn' should receive asynchronous messages. */
1733 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1735 if (ofconn->type == OFCONN_PRIMARY) {
1736 /* Primary controllers always get asynchronous messages unless they
1737 * have configured themselves as "slaves". */
1738 return ofconn->role != NX_ROLE_SLAVE;
1740 /* Service connections don't get asynchronous messages unless they have
1741 * explicitly asked for them by setting a nonzero miss send length. */
1742 return ofconn->miss_send_len > 0;
1746 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1747 * and 'target', suitable for use in log messages for identifying the
1750 * The name is dynamically allocated. The caller should free it (with free())
1751 * when it is no longer needed. */
1753 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1755 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1759 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1763 for (i = 0; i < N_SCHEDULERS; i++) {
1764 struct pinsched **s = &ofconn->schedulers[i];
1768 *s = pinsched_create(rate, burst,
1769 ofconn->ofproto->switch_status);
1771 pinsched_set_limits(*s, rate, burst);
1774 pinsched_destroy(*s);
1781 ofservice_reconfigure(struct ofservice *ofservice,
1782 const struct ofproto_controller *c)
1784 ofservice->probe_interval = c->probe_interval;
1785 ofservice->rate_limit = c->rate_limit;
1786 ofservice->burst_limit = c->burst_limit;
1789 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1790 * positive errno value. */
1792 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1794 struct ofservice *ofservice;
1795 struct pvconn *pvconn;
1798 error = pvconn_open(c->target, &pvconn);
1803 ofservice = xzalloc(sizeof *ofservice);
1804 hmap_insert(&ofproto->services, &ofservice->node,
1805 hash_string(c->target, 0));
1806 ofservice->pvconn = pvconn;
1808 ofservice_reconfigure(ofservice, c);
1814 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1816 hmap_remove(&ofproto->services, &ofservice->node);
1817 pvconn_close(ofservice->pvconn);
1821 /* Finds and returns the ofservice within 'ofproto' that has the given
1822 * 'target', or a null pointer if none exists. */
1823 static struct ofservice *
1824 ofservice_lookup(struct ofproto *ofproto, const char *target)
1826 struct ofservice *ofservice;
1828 HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
1829 hash_string(target, 0), &ofproto->services) {
1830 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1837 /* Caller is responsible for initializing the 'cr' member of the returned
1839 static struct rule *
1840 rule_create(struct ofproto *ofproto, struct rule *super,
1841 const union ofp_action *actions, size_t n_actions,
1842 uint16_t idle_timeout, uint16_t hard_timeout,
1843 uint64_t flow_cookie, bool send_flow_removed)
1845 struct rule *rule = xzalloc(sizeof *rule);
1846 rule->idle_timeout = idle_timeout;
1847 rule->hard_timeout = hard_timeout;
1848 rule->flow_cookie = flow_cookie;
1849 rule->used = rule->created = time_msec();
1850 rule->send_flow_removed = send_flow_removed;
1851 rule->super = super;
1853 list_push_back(&super->list, &rule->list);
1855 list_init(&rule->list);
1857 rule->n_actions = n_actions;
1858 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1859 netflow_flow_clear(&rule->nf_flow);
1860 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1865 static struct rule *
1866 rule_from_cls_rule(const struct cls_rule *cls_rule)
1868 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1872 rule_free(struct rule *rule)
1874 free(rule->actions);
1875 free(rule->odp_actions);
1879 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1880 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1881 * through all of its subrules and revalidates them, destroying any that no
1882 * longer has a super-rule (which is probably all of them).
1884 * Before calling this function, the caller must make have removed 'rule' from
1885 * the classifier. If 'rule' is an exact-match rule, the caller is also
1886 * responsible for ensuring that it has been uninstalled from the datapath. */
1888 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1891 struct rule *subrule, *next;
1892 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1893 revalidate_rule(ofproto, subrule);
1896 list_remove(&rule->list);
1902 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1904 const union ofp_action *oa;
1905 struct actions_iterator i;
1907 if (out_port == htons(OFPP_NONE)) {
1910 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1911 oa = actions_next(&i)) {
1912 if (action_outputs_to_port(oa, out_port)) {
1919 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1920 * 'packet', which arrived on 'in_port'.
1922 * Takes ownership of 'packet'. */
1924 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1925 const union odp_action *actions, size_t n_actions,
1926 struct ofpbuf *packet)
1928 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1929 /* As an optimization, avoid a round-trip from userspace to kernel to
1930 * userspace. This also avoids possibly filling up kernel packet
1931 * buffers along the way. */
1932 struct odp_msg *msg;
1934 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1935 msg->type = _ODPL_ACTION_NR;
1936 msg->length = sizeof(struct odp_msg) + packet->size;
1937 msg->port = in_port;
1939 msg->arg = actions[0].controller.arg;
1941 send_packet_in(ofproto, packet);
1947 error = dpif_execute(ofproto->dpif, in_port,
1948 actions, n_actions, packet);
1949 ofpbuf_delete(packet);
1954 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1955 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1956 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1958 * The flow that 'packet' actually contains does not need to actually match
1959 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1960 * the packet and byte counters for 'rule' will be credited for the packet sent
1961 * out whether or not the packet actually matches 'rule'.
1963 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1964 * the caller must already have accurately composed ODP actions for it given
1965 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1966 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1967 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1968 * actions and apply them to 'packet'.
1970 * Takes ownership of 'packet'. */
1972 rule_execute(struct ofproto *ofproto, struct rule *rule,
1973 struct ofpbuf *packet, const flow_t *flow)
1975 const union odp_action *actions;
1976 struct odp_flow_stats stats;
1978 struct odp_actions a;
1980 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1982 /* Grab or compose the ODP actions.
1984 * The special case for an exact-match 'rule' where 'flow' is not the
1985 * rule's flow is important to avoid, e.g., sending a packet out its input
1986 * port simply because the ODP actions were composed for the wrong
1988 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1989 struct rule *super = rule->super ? rule->super : rule;
1990 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1991 packet, &a, NULL, 0, NULL)) {
1992 ofpbuf_delete(packet);
1995 actions = a.actions;
1996 n_actions = a.n_actions;
1998 actions = rule->odp_actions;
1999 n_actions = rule->n_odp_actions;
2002 /* Execute the ODP actions. */
2003 flow_extract_stats(flow, packet, &stats);
2004 if (execute_odp_actions(ofproto, flow->in_port,
2005 actions, n_actions, packet)) {
2006 update_stats(ofproto, rule, &stats);
2007 rule->used = time_msec();
2008 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2012 /* Inserts 'rule' into 'p''s flow table.
2014 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2015 * actions on it and credits the statistics for sending the packet to 'rule'.
2016 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2019 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2022 struct rule *displaced_rule;
2024 /* Insert the rule in the classifier. */
2025 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2026 if (!rule->cr.wc.wildcards) {
2027 rule_make_actions(p, rule, packet);
2030 /* Send the packet and credit it to the rule. */
2033 flow_extract(packet, 0, in_port, &flow);
2034 rule_execute(p, rule, packet, &flow);
2037 /* Install the rule in the datapath only after sending the packet, to
2038 * avoid packet reordering. */
2039 if (rule->cr.wc.wildcards) {
2040 COVERAGE_INC(ofproto_add_wc_flow);
2041 p->need_revalidate = true;
2043 rule_install(p, rule, displaced_rule);
2046 /* Free the rule that was displaced, if any. */
2047 if (displaced_rule) {
2048 rule_destroy(p, displaced_rule);
2052 static struct rule *
2053 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2056 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2057 rule->idle_timeout, rule->hard_timeout,
2059 COVERAGE_INC(ofproto_subrule_create);
2060 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2061 : rule->cr.priority), &subrule->cr);
2062 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2067 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2069 * - If 'rule' was installed in the datapath, uninstalls it and updates
2070 * 'rule''s statistics (or its super-rule's statistics, if it is a
2071 * subrule), via rule_uninstall().
2073 * - Removes 'rule' from the classifier.
2075 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2076 * uninstalls and destroys) its subrules, via rule_destroy().
2079 rule_remove(struct ofproto *ofproto, struct rule *rule)
2081 if (rule->cr.wc.wildcards) {
2082 COVERAGE_INC(ofproto_del_wc_flow);
2083 ofproto->need_revalidate = true;
2085 rule_uninstall(ofproto, rule);
2087 classifier_remove(&ofproto->cls, &rule->cr);
2088 rule_destroy(ofproto, rule);
2091 /* Returns true if the actions changed, false otherwise. */
2093 rule_make_actions(struct ofproto *p, struct rule *rule,
2094 const struct ofpbuf *packet)
2096 const struct rule *super;
2097 struct odp_actions a;
2100 assert(!rule->cr.wc.wildcards);
2102 super = rule->super ? rule->super : rule;
2104 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2105 packet, &a, &rule->tags, &rule->may_install,
2106 &rule->nf_flow.output_iface);
2108 actions_len = a.n_actions * sizeof *a.actions;
2109 if (rule->n_odp_actions != a.n_actions
2110 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2111 COVERAGE_INC(ofproto_odp_unchanged);
2112 free(rule->odp_actions);
2113 rule->n_odp_actions = a.n_actions;
2114 rule->odp_actions = xmemdup(a.actions, actions_len);
2122 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2123 struct odp_flow_put *put)
2125 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2126 put->flow.key = rule->cr.flow;
2127 put->flow.actions = rule->odp_actions;
2128 put->flow.n_actions = rule->n_odp_actions;
2129 put->flow.flags = 0;
2131 return dpif_flow_put(ofproto->dpif, put);
2135 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2137 assert(!rule->cr.wc.wildcards);
2139 if (rule->may_install) {
2140 struct odp_flow_put put;
2141 if (!do_put_flow(p, rule,
2142 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2144 rule->installed = true;
2145 if (displaced_rule) {
2146 update_stats(p, displaced_rule, &put.flow.stats);
2147 rule_post_uninstall(p, displaced_rule);
2150 } else if (displaced_rule) {
2151 rule_uninstall(p, displaced_rule);
2156 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2158 if (rule->installed) {
2159 struct odp_flow_put put;
2160 COVERAGE_INC(ofproto_dp_missed);
2161 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2163 rule_install(ofproto, rule, NULL);
2168 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2170 bool actions_changed;
2171 uint16_t new_out_iface, old_out_iface;
2173 old_out_iface = rule->nf_flow.output_iface;
2174 actions_changed = rule_make_actions(ofproto, rule, NULL);
2176 if (rule->may_install) {
2177 if (rule->installed) {
2178 if (actions_changed) {
2179 struct odp_flow_put put;
2180 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2181 | ODPPF_ZERO_STATS, &put);
2182 update_stats(ofproto, rule, &put.flow.stats);
2184 /* Temporarily set the old output iface so that NetFlow
2185 * messages have the correct output interface for the old
2187 new_out_iface = rule->nf_flow.output_iface;
2188 rule->nf_flow.output_iface = old_out_iface;
2189 rule_post_uninstall(ofproto, rule);
2190 rule->nf_flow.output_iface = new_out_iface;
2193 rule_install(ofproto, rule, NULL);
2196 rule_uninstall(ofproto, rule);
2201 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2203 uint64_t total_bytes = rule->byte_count + extra_bytes;
2205 if (ofproto->ofhooks->account_flow_cb
2206 && total_bytes > rule->accounted_bytes)
2208 ofproto->ofhooks->account_flow_cb(
2209 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2210 total_bytes - rule->accounted_bytes, ofproto->aux);
2211 rule->accounted_bytes = total_bytes;
2215 /* 'rule' must be an exact-match rule in 'p'.
2217 * If 'rule' is installed in the datapath, uninstalls it and updates's
2218 * statistics. If 'rule' is a subrule, the statistics that are updated are
2219 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2222 * If 'rule' is not installed, this function has no effect. */
2224 rule_uninstall(struct ofproto *p, struct rule *rule)
2226 assert(!rule->cr.wc.wildcards);
2227 if (rule->installed) {
2228 struct odp_flow odp_flow;
2230 odp_flow.key = rule->cr.flow;
2231 odp_flow.actions = NULL;
2232 odp_flow.n_actions = 0;
2234 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2235 update_stats(p, rule, &odp_flow.stats);
2237 rule->installed = false;
2239 rule_post_uninstall(p, rule);
2244 is_controller_rule(struct rule *rule)
2246 /* If the only action is send to the controller then don't report
2247 * NetFlow expiration messages since it is just part of the control
2248 * logic for the network and not real traffic. */
2252 && rule->super->n_actions == 1
2253 && action_outputs_to_port(&rule->super->actions[0],
2254 htons(OFPP_CONTROLLER)));
2258 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2260 struct rule *super = rule->super;
2262 rule_account(ofproto, rule, 0);
2264 if (ofproto->netflow && !is_controller_rule(rule)) {
2265 struct ofexpired expired;
2266 expired.flow = rule->cr.flow;
2267 expired.packet_count = rule->packet_count;
2268 expired.byte_count = rule->byte_count;
2269 expired.used = rule->used;
2270 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2273 super->packet_count += rule->packet_count;
2274 super->byte_count += rule->byte_count;
2276 /* Reset counters to prevent double counting if the rule ever gets
2278 rule->packet_count = 0;
2279 rule->byte_count = 0;
2280 rule->accounted_bytes = 0;
2282 netflow_flow_clear(&rule->nf_flow);
2287 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2288 struct rconn_packet_counter *counter)
2290 update_openflow_length(msg);
2291 if (rconn_send(ofconn->rconn, msg, counter)) {
2297 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2298 int error, const void *data, size_t len)
2301 struct ofp_error_msg *oem;
2303 if (!(error >> 16)) {
2304 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2309 COVERAGE_INC(ofproto_error);
2310 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2311 oh ? oh->xid : 0, &buf);
2312 oem->type = htons((unsigned int) error >> 16);
2313 oem->code = htons(error & 0xffff);
2314 memcpy(oem->data, data, len);
2315 queue_tx(buf, ofconn, ofconn->reply_counter);
2319 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2322 size_t oh_length = ntohs(oh->length);
2323 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2327 hton_ofp_phy_port(struct ofp_phy_port *opp)
2329 opp->port_no = htons(opp->port_no);
2330 opp->config = htonl(opp->config);
2331 opp->state = htonl(opp->state);
2332 opp->curr = htonl(opp->curr);
2333 opp->advertised = htonl(opp->advertised);
2334 opp->supported = htonl(opp->supported);
2335 opp->peer = htonl(opp->peer);
2339 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2341 struct ofp_header *rq = oh;
2342 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2347 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2348 struct ofp_header *oh)
2350 struct ofp_switch_features *osf;
2352 unsigned int port_no;
2353 struct ofport *port;
2355 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2356 osf->datapath_id = htonll(p->datapath_id);
2357 osf->n_buffers = htonl(pktbuf_capacity());
2359 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2360 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2361 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2362 (1u << OFPAT_SET_VLAN_VID) |
2363 (1u << OFPAT_SET_VLAN_PCP) |
2364 (1u << OFPAT_STRIP_VLAN) |
2365 (1u << OFPAT_SET_DL_SRC) |
2366 (1u << OFPAT_SET_DL_DST) |
2367 (1u << OFPAT_SET_NW_SRC) |
2368 (1u << OFPAT_SET_NW_DST) |
2369 (1u << OFPAT_SET_NW_TOS) |
2370 (1u << OFPAT_SET_TP_SRC) |
2371 (1u << OFPAT_SET_TP_DST) |
2372 (1u << OFPAT_ENQUEUE));
2374 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2375 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2378 queue_tx(buf, ofconn, ofconn->reply_counter);
2383 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2384 struct ofp_header *oh)
2387 struct ofp_switch_config *osc;
2391 /* Figure out flags. */
2392 dpif_get_drop_frags(p->dpif, &drop_frags);
2393 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2396 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2397 osc->flags = htons(flags);
2398 osc->miss_send_len = htons(ofconn->miss_send_len);
2399 queue_tx(buf, ofconn, ofconn->reply_counter);
2405 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2406 struct ofp_switch_config *osc)
2411 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2415 flags = ntohs(osc->flags);
2417 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2418 switch (flags & OFPC_FRAG_MASK) {
2419 case OFPC_FRAG_NORMAL:
2420 dpif_set_drop_frags(p->dpif, false);
2422 case OFPC_FRAG_DROP:
2423 dpif_set_drop_frags(p->dpif, true);
2426 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2432 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2438 add_output_group_action(struct odp_actions *actions, uint16_t group,
2439 uint16_t *nf_output_iface)
2441 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2443 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2444 *nf_output_iface = NF_OUT_FLOOD;
2449 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2451 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2452 a->controller.arg = max_len;
2455 struct action_xlate_ctx {
2457 flow_t flow; /* Flow to which these actions correspond. */
2458 int recurse; /* Recursion level, via xlate_table_action. */
2459 struct ofproto *ofproto;
2460 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2461 * null pointer if we are revalidating
2462 * without a packet to refer to. */
2465 struct odp_actions *out; /* Datapath actions. */
2466 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2467 bool may_set_up_flow; /* True ordinarily; false if the actions must
2468 * be reassessed for every packet. */
2469 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2472 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2473 * flow translation. */
2474 #define MAX_RESUBMIT_RECURSION 8
2476 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2477 struct action_xlate_ctx *ctx);
2480 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2482 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2485 if (ofport->opp.config & OFPPC_NO_FWD) {
2486 /* Forwarding disabled on port. */
2491 * We don't have an ofport record for this port, but it doesn't hurt to
2492 * allow forwarding to it anyhow. Maybe such a port will appear later
2493 * and we're pre-populating the flow table.
2497 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2498 ctx->nf_output_iface = port;
2501 static struct rule *
2502 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2505 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2507 /* The rule we found might not be valid, since we could be in need of
2508 * revalidation. If it is not valid, don't return it. */
2511 && ofproto->need_revalidate
2512 && !revalidate_rule(ofproto, rule)) {
2513 COVERAGE_INC(ofproto_invalidated);
2521 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2523 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2524 uint16_t old_in_port;
2527 /* Look up a flow with 'in_port' as the input port. Then restore the
2528 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2529 * have surprising behavior). */
2530 old_in_port = ctx->flow.in_port;
2531 ctx->flow.in_port = in_port;
2532 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2533 ctx->flow.in_port = old_in_port;
2541 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2545 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2547 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2548 MAX_RESUBMIT_RECURSION);
2553 xlate_output_action__(struct action_xlate_ctx *ctx,
2554 uint16_t port, uint16_t max_len)
2557 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2559 ctx->nf_output_iface = NF_OUT_DROP;
2563 add_output_action(ctx, ctx->flow.in_port);
2566 xlate_table_action(ctx, ctx->flow.in_port);
2569 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2570 ctx->out, ctx->tags,
2571 &ctx->nf_output_iface,
2572 ctx->ofproto->aux)) {
2573 COVERAGE_INC(ofproto_uninstallable);
2574 ctx->may_set_up_flow = false;
2578 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2579 &ctx->nf_output_iface);
2582 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2584 case OFPP_CONTROLLER:
2585 add_controller_action(ctx->out, max_len);
2588 add_output_action(ctx, ODPP_LOCAL);
2591 odp_port = ofp_port_to_odp_port(port);
2592 if (odp_port != ctx->flow.in_port) {
2593 add_output_action(ctx, odp_port);
2598 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2599 ctx->nf_output_iface = NF_OUT_FLOOD;
2600 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2601 ctx->nf_output_iface = prev_nf_output_iface;
2602 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2603 ctx->nf_output_iface != NF_OUT_FLOOD) {
2604 ctx->nf_output_iface = NF_OUT_MULTI;
2609 xlate_output_action(struct action_xlate_ctx *ctx,
2610 const struct ofp_action_output *oao)
2612 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2615 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2616 * optimization, because we're going to add another action that sets the
2617 * priority immediately after, or because there are no actions following the
2620 remove_pop_action(struct action_xlate_ctx *ctx)
2622 size_t n = ctx->out->n_actions;
2623 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2624 ctx->out->n_actions--;
2629 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2630 const struct ofp_action_enqueue *oae)
2632 uint16_t ofp_port, odp_port;
2636 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2639 /* Fall back to ordinary output action. */
2640 xlate_output_action__(ctx, ntohs(oae->port), 0);
2644 /* Figure out ODP output port. */
2645 ofp_port = ntohs(oae->port);
2646 if (ofp_port != OFPP_IN_PORT) {
2647 odp_port = ofp_port_to_odp_port(ofp_port);
2649 odp_port = ctx->flow.in_port;
2652 /* Add ODP actions. */
2653 remove_pop_action(ctx);
2654 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2656 add_output_action(ctx, odp_port);
2657 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2659 /* Update NetFlow output port. */
2660 if (ctx->nf_output_iface == NF_OUT_DROP) {
2661 ctx->nf_output_iface = odp_port;
2662 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2663 ctx->nf_output_iface = NF_OUT_MULTI;
2668 xlate_nicira_action(struct action_xlate_ctx *ctx,
2669 const struct nx_action_header *nah)
2671 const struct nx_action_resubmit *nar;
2672 const struct nx_action_set_tunnel *nast;
2673 union odp_action *oa;
2674 int subtype = ntohs(nah->subtype);
2676 assert(nah->vendor == htonl(NX_VENDOR_ID));
2678 case NXAST_RESUBMIT:
2679 nar = (const struct nx_action_resubmit *) nah;
2680 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2683 case NXAST_SET_TUNNEL:
2684 nast = (const struct nx_action_set_tunnel *) nah;
2685 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2686 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2689 case NXAST_DROP_SPOOFED_ARP:
2690 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2691 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2695 /* If you add a new action here that modifies flow data, don't forget to
2696 * update the flow key in ctx->flow at the same time. */
2699 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2705 do_xlate_actions(const union ofp_action *in, size_t n_in,
2706 struct action_xlate_ctx *ctx)
2708 struct actions_iterator iter;
2709 const union ofp_action *ia;
2710 const struct ofport *port;
2712 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2713 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2714 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2715 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2716 /* Drop this flow. */
2720 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2721 uint16_t type = ntohs(ia->type);
2722 union odp_action *oa;
2726 xlate_output_action(ctx, &ia->output);
2729 case OFPAT_SET_VLAN_VID:
2730 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2731 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2734 case OFPAT_SET_VLAN_PCP:
2735 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2736 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2739 case OFPAT_STRIP_VLAN:
2740 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2741 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2742 ctx->flow.dl_vlan_pcp = 0;
2745 case OFPAT_SET_DL_SRC:
2746 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2747 memcpy(oa->dl_addr.dl_addr,
2748 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2749 memcpy(ctx->flow.dl_src,
2750 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2753 case OFPAT_SET_DL_DST:
2754 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2755 memcpy(oa->dl_addr.dl_addr,
2756 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2757 memcpy(ctx->flow.dl_dst,
2758 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2761 case OFPAT_SET_NW_SRC:
2762 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2763 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2766 case OFPAT_SET_NW_DST:
2767 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2768 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2771 case OFPAT_SET_NW_TOS:
2772 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2773 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2776 case OFPAT_SET_TP_SRC:
2777 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2778 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2781 case OFPAT_SET_TP_DST:
2782 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2783 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2787 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2791 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2795 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2802 xlate_actions(const union ofp_action *in, size_t n_in,
2803 const flow_t *flow, struct ofproto *ofproto,
2804 const struct ofpbuf *packet,
2805 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2806 uint16_t *nf_output_iface)
2808 tag_type no_tags = 0;
2809 struct action_xlate_ctx ctx;
2810 COVERAGE_INC(ofproto_ofp2odp);
2811 odp_actions_init(out);
2814 ctx.ofproto = ofproto;
2815 ctx.packet = packet;
2817 ctx.tags = tags ? tags : &no_tags;
2818 ctx.may_set_up_flow = true;
2819 ctx.nf_output_iface = NF_OUT_DROP;
2820 do_xlate_actions(in, n_in, &ctx);
2821 remove_pop_action(&ctx);
2823 /* Check with in-band control to see if we're allowed to set up this
2825 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2826 ctx.may_set_up_flow = false;
2829 if (may_set_up_flow) {
2830 *may_set_up_flow = ctx.may_set_up_flow;
2832 if (nf_output_iface) {
2833 *nf_output_iface = ctx.nf_output_iface;
2835 if (odp_actions_overflow(out)) {
2836 COVERAGE_INC(odp_overflow);
2837 odp_actions_init(out);
2838 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2843 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2844 * error message code (composed with ofp_mkerr()) for the caller to propagate
2845 * upward. Otherwise, returns 0.
2847 * 'oh' is used to make log messages more informative. */
2849 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2851 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2852 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2855 type_name = ofp_message_type_to_string(oh->type);
2856 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2860 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2867 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2868 struct ofp_header *oh)
2870 struct ofp_packet_out *opo;
2871 struct ofpbuf payload, *buffer;
2872 struct odp_actions actions;
2878 error = reject_slave_controller(ofconn, oh);
2883 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2887 opo = (struct ofp_packet_out *) oh;
2889 COVERAGE_INC(ofproto_packet_out);
2890 if (opo->buffer_id != htonl(UINT32_MAX)) {
2891 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2893 if (error || !buffer) {
2901 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2902 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2903 &flow, p, &payload, &actions, NULL, NULL, NULL);
2908 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2910 ofpbuf_delete(buffer);
2916 update_port_config(struct ofproto *p, struct ofport *port,
2917 uint32_t config, uint32_t mask)
2919 mask &= config ^ port->opp.config;
2920 if (mask & OFPPC_PORT_DOWN) {
2921 if (config & OFPPC_PORT_DOWN) {
2922 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2924 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2927 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2928 if (mask & REVALIDATE_BITS) {
2929 COVERAGE_INC(ofproto_costly_flags);
2930 port->opp.config ^= mask & REVALIDATE_BITS;
2931 p->need_revalidate = true;
2933 #undef REVALIDATE_BITS
2934 if (mask & OFPPC_NO_FLOOD) {
2935 port->opp.config ^= OFPPC_NO_FLOOD;
2936 refresh_port_groups(p);
2938 if (mask & OFPPC_NO_PACKET_IN) {
2939 port->opp.config ^= OFPPC_NO_PACKET_IN;
2944 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2945 struct ofp_header *oh)
2947 const struct ofp_port_mod *opm;
2948 struct ofport *port;
2951 error = reject_slave_controller(ofconn, oh);
2955 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2959 opm = (struct ofp_port_mod *) oh;
2961 port = port_array_get(&p->ports,
2962 ofp_port_to_odp_port(ntohs(opm->port_no)));
2964 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2965 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2966 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2968 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2969 if (opm->advertise) {
2970 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2976 static struct ofpbuf *
2977 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2979 struct ofp_stats_reply *osr;
2982 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2983 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2985 osr->flags = htons(0);
2989 static struct ofpbuf *
2990 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2992 return make_stats_reply(request->header.xid, request->type, body_len);
2996 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2998 struct ofpbuf *msg = *msgp;
2999 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3000 if (nbytes + msg->size > UINT16_MAX) {
3001 struct ofp_stats_reply *reply = msg->data;
3002 reply->flags = htons(OFPSF_REPLY_MORE);
3003 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3004 queue_tx(msg, ofconn, ofconn->reply_counter);
3006 return ofpbuf_put_uninit(*msgp, nbytes);
3010 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3011 struct ofp_stats_request *request)
3013 struct ofp_desc_stats *ods;
3016 msg = start_stats_reply(request, sizeof *ods);
3017 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3018 memset(ods, 0, sizeof *ods);
3019 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3020 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3021 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3022 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3023 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3024 queue_tx(msg, ofconn, ofconn->reply_counter);
3030 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3031 struct ofp_stats_request *request)
3033 struct ofp_table_stats *ots;
3035 struct odp_stats dpstats;
3036 int n_exact, n_subrules, n_wild;
3039 msg = start_stats_reply(request, sizeof *ots * 2);
3041 /* Count rules of various kinds. */
3043 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, struct rule, cr, &p->cls) {
3048 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3049 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3052 dpif_get_dp_stats(p->dpif, &dpstats);
3053 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3054 memset(ots, 0, sizeof *ots);
3055 ots->table_id = TABLEID_HASH;
3056 strcpy(ots->name, "hash");
3057 ots->wildcards = htonl(0);
3058 ots->max_entries = htonl(dpstats.max_capacity);
3059 ots->active_count = htonl(n_exact);
3060 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3062 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3064 /* Classifier table. */
3065 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3066 memset(ots, 0, sizeof *ots);
3067 ots->table_id = TABLEID_CLASSIFIER;
3068 strcpy(ots->name, "classifier");
3069 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3071 ots->max_entries = htonl(65536);
3072 ots->active_count = htonl(n_wild);
3073 ots->lookup_count = htonll(0); /* XXX */
3074 ots->matched_count = htonll(0); /* XXX */
3076 queue_tx(msg, ofconn, ofconn->reply_counter);
3081 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
3082 struct ofpbuf **msgp)
3084 struct netdev_stats stats;
3085 struct ofp_port_stats *ops;
3087 /* Intentionally ignore return value, since errors will set
3088 * 'stats' to all-1s, which is correct for OpenFlow, and
3089 * netdev_get_stats() will log errors. */
3090 netdev_get_stats(port->netdev, &stats);
3092 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3093 ops->port_no = htons(odp_port_to_ofp_port(port_no));
3094 memset(ops->pad, 0, sizeof ops->pad);
3095 ops->rx_packets = htonll(stats.rx_packets);
3096 ops->tx_packets = htonll(stats.tx_packets);
3097 ops->rx_bytes = htonll(stats.rx_bytes);
3098 ops->tx_bytes = htonll(stats.tx_bytes);
3099 ops->rx_dropped = htonll(stats.rx_dropped);
3100 ops->tx_dropped = htonll(stats.tx_dropped);
3101 ops->rx_errors = htonll(stats.rx_errors);
3102 ops->tx_errors = htonll(stats.tx_errors);
3103 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3104 ops->rx_over_err = htonll(stats.rx_over_errors);
3105 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3106 ops->collisions = htonll(stats.collisions);
3110 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3111 struct ofp_stats_request *osr,
3114 struct ofp_port_stats_request *psr;
3115 struct ofp_port_stats *ops;
3117 struct ofport *port;
3118 unsigned int port_no;
3120 if (arg_size != sizeof *psr) {
3121 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3123 psr = (struct ofp_port_stats_request *) osr->body;
3125 msg = start_stats_reply(osr, sizeof *ops * 16);
3126 if (psr->port_no != htons(OFPP_NONE)) {
3127 port = port_array_get(&p->ports,
3128 ofp_port_to_odp_port(ntohs(psr->port_no)));
3130 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
3133 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
3134 append_port_stat(port, port_no, ofconn, &msg);
3138 queue_tx(msg, ofconn, ofconn->reply_counter);
3142 struct flow_stats_cbdata {
3143 struct ofproto *ofproto;
3144 struct ofconn *ofconn;
3149 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3150 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3151 * returned statistic include statistics for all of 'rule''s subrules. */
3153 query_stats(struct ofproto *p, struct rule *rule,
3154 uint64_t *packet_countp, uint64_t *byte_countp)
3156 uint64_t packet_count, byte_count;
3157 struct rule *subrule;
3158 struct odp_flow *odp_flows;
3161 /* Start from historical data for 'rule' itself that are no longer tracked
3162 * by the datapath. This counts, for example, subrules that have
3164 packet_count = rule->packet_count;
3165 byte_count = rule->byte_count;
3167 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3168 * wildcarded then on all of its subrules.
3170 * Also, add any statistics that are not tracked by the datapath for each
3171 * subrule. This includes, for example, statistics for packets that were
3172 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3174 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3175 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3176 if (rule->cr.wc.wildcards) {
3178 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3179 odp_flows[i++].key = subrule->cr.flow;
3180 packet_count += subrule->packet_count;
3181 byte_count += subrule->byte_count;
3184 odp_flows[0].key = rule->cr.flow;
3187 /* Fetch up-to-date statistics from the datapath and add them in. */
3188 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3190 for (i = 0; i < n_odp_flows; i++) {
3191 struct odp_flow *odp_flow = &odp_flows[i];
3192 packet_count += odp_flow->stats.n_packets;
3193 byte_count += odp_flow->stats.n_bytes;
3198 /* Return the stats to the caller. */
3199 *packet_countp = packet_count;
3200 *byte_countp = byte_count;
3204 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3206 struct rule *rule = rule_from_cls_rule(rule_);
3207 struct flow_stats_cbdata *cbdata = cbdata_;
3208 struct ofp_flow_stats *ofs;
3209 uint64_t packet_count, byte_count;
3210 size_t act_len, len;
3211 long long int tdiff = time_msec() - rule->created;
3212 uint32_t sec = tdiff / 1000;
3213 uint32_t msec = tdiff - (sec * 1000);
3215 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3219 act_len = sizeof *rule->actions * rule->n_actions;
3220 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3222 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3224 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3225 ofs->length = htons(len);
3226 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3228 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3229 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3230 ofs->duration_sec = htonl(sec);
3231 ofs->duration_nsec = htonl(msec * 1000000);
3232 ofs->cookie = rule->flow_cookie;
3233 ofs->priority = htons(rule->cr.priority);
3234 ofs->idle_timeout = htons(rule->idle_timeout);
3235 ofs->hard_timeout = htons(rule->hard_timeout);
3236 memset(ofs->pad2, 0, sizeof ofs->pad2);
3237 ofs->packet_count = htonll(packet_count);
3238 ofs->byte_count = htonll(byte_count);
3239 memcpy(ofs->actions, rule->actions, act_len);
3243 table_id_to_include(uint8_t table_id)
3245 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3246 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3247 : table_id == 0xff ? CLS_INC_ALL
3252 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3253 const struct ofp_stats_request *osr,
3256 struct ofp_flow_stats_request *fsr;
3257 struct flow_stats_cbdata cbdata;
3258 struct cls_rule target;
3260 if (arg_size != sizeof *fsr) {
3261 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3263 fsr = (struct ofp_flow_stats_request *) osr->body;
3265 COVERAGE_INC(ofproto_flows_req);
3267 cbdata.ofconn = ofconn;
3268 cbdata.out_port = fsr->out_port;
3269 cbdata.msg = start_stats_reply(osr, 1024);
3270 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3271 classifier_for_each_match(&p->cls, &target,
3272 table_id_to_include(fsr->table_id),
3273 flow_stats_cb, &cbdata);
3274 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3278 struct flow_stats_ds_cbdata {
3279 struct ofproto *ofproto;
3284 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3286 struct rule *rule = rule_from_cls_rule(rule_);
3287 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3288 struct ds *results = cbdata->results;
3289 struct ofp_match match;
3290 uint64_t packet_count, byte_count;
3291 size_t act_len = sizeof *rule->actions * rule->n_actions;
3293 /* Don't report on subrules. */
3294 if (rule->super != NULL) {
3298 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3299 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3300 cbdata->ofproto->tun_id_from_cookie, &match);
3302 ds_put_format(results, "duration=%llds, ",
3303 (time_msec() - rule->created) / 1000);
3304 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3305 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3306 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3307 ofp_print_match(results, &match, true);
3308 ofp_print_actions(results, &rule->actions->header, act_len);
3309 ds_put_cstr(results, "\n");
3312 /* Adds a pretty-printed description of all flows to 'results', including
3313 * those marked hidden by secchan (e.g., by in-band control). */
3315 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3317 struct ofp_match match;
3318 struct cls_rule target;
3319 struct flow_stats_ds_cbdata cbdata;
3321 memset(&match, 0, sizeof match);
3322 match.wildcards = htonl(OVSFW_ALL);
3325 cbdata.results = results;
3327 cls_rule_from_match(&match, 0, false, 0, &target);
3328 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3329 flow_stats_ds_cb, &cbdata);
3332 struct aggregate_stats_cbdata {
3333 struct ofproto *ofproto;
3335 uint64_t packet_count;
3336 uint64_t byte_count;
3341 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3343 struct rule *rule = rule_from_cls_rule(rule_);
3344 struct aggregate_stats_cbdata *cbdata = cbdata_;
3345 uint64_t packet_count, byte_count;
3347 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3351 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3353 cbdata->packet_count += packet_count;
3354 cbdata->byte_count += byte_count;
3359 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3360 const struct ofp_stats_request *osr,
3363 struct ofp_aggregate_stats_request *asr;
3364 struct ofp_aggregate_stats_reply *reply;
3365 struct aggregate_stats_cbdata cbdata;
3366 struct cls_rule target;
3369 if (arg_size != sizeof *asr) {
3370 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3372 asr = (struct ofp_aggregate_stats_request *) osr->body;
3374 COVERAGE_INC(ofproto_agg_request);
3376 cbdata.out_port = asr->out_port;
3377 cbdata.packet_count = 0;
3378 cbdata.byte_count = 0;
3380 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3381 classifier_for_each_match(&p->cls, &target,
3382 table_id_to_include(asr->table_id),
3383 aggregate_stats_cb, &cbdata);
3385 msg = start_stats_reply(osr, sizeof *reply);
3386 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3387 reply->flow_count = htonl(cbdata.n_flows);
3388 reply->packet_count = htonll(cbdata.packet_count);
3389 reply->byte_count = htonll(cbdata.byte_count);
3390 queue_tx(msg, ofconn, ofconn->reply_counter);
3394 struct queue_stats_cbdata {
3395 struct ofconn *ofconn;
3401 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3402 const struct netdev_queue_stats *stats)
3404 struct ofp_queue_stats *reply;
3406 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3407 reply->port_no = htons(cbdata->port_no);
3408 memset(reply->pad, 0, sizeof reply->pad);
3409 reply->queue_id = htonl(queue_id);
3410 reply->tx_bytes = htonll(stats->tx_bytes);
3411 reply->tx_packets = htonll(stats->tx_packets);
3412 reply->tx_errors = htonll(stats->tx_errors);
3416 handle_queue_stats_dump_cb(uint32_t queue_id,
3417 struct netdev_queue_stats *stats,
3420 struct queue_stats_cbdata *cbdata = cbdata_;
3422 put_queue_stats(cbdata, queue_id, stats);
3426 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3428 struct queue_stats_cbdata *cbdata)
3430 cbdata->port_no = port_no;
3431 if (queue_id == OFPQ_ALL) {
3432 netdev_dump_queue_stats(port->netdev,
3433 handle_queue_stats_dump_cb, cbdata);
3435 struct netdev_queue_stats stats;
3437 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3438 put_queue_stats(cbdata, queue_id, &stats);
3443 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3444 const struct ofp_stats_request *osr,
3447 struct ofp_queue_stats_request *qsr;
3448 struct queue_stats_cbdata cbdata;
3449 struct ofport *port;
3450 unsigned int port_no;
3453 if (arg_size != sizeof *qsr) {
3454 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3456 qsr = (struct ofp_queue_stats_request *) osr->body;
3458 COVERAGE_INC(ofproto_queue_req);
3460 cbdata.ofconn = ofconn;
3461 cbdata.msg = start_stats_reply(osr, 128);
3463 port_no = ntohs(qsr->port_no);
3464 queue_id = ntohl(qsr->queue_id);
3465 if (port_no == OFPP_ALL) {
3466 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3467 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3469 } else if (port_no < ofproto->max_ports) {
3470 port = port_array_get(&ofproto->ports, port_no);
3472 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3475 ofpbuf_delete(cbdata.msg);
3476 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3478 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3484 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3485 struct ofp_header *oh)
3487 struct ofp_stats_request *osr;
3491 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3496 osr = (struct ofp_stats_request *) oh;
3498 switch (ntohs(osr->type)) {
3500 return handle_desc_stats_request(p, ofconn, osr);
3503 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3505 case OFPST_AGGREGATE:
3506 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3509 return handle_table_stats_request(p, ofconn, osr);
3512 return handle_port_stats_request(p, ofconn, osr, arg_size);
3515 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3518 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3521 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3525 static long long int
3526 msec_from_nsec(uint64_t sec, uint32_t nsec)
3528 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3532 update_time(struct ofproto *ofproto, struct rule *rule,
3533 const struct odp_flow_stats *stats)
3535 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3536 if (used > rule->used) {
3538 if (rule->super && used > rule->super->used) {
3539 rule->super->used = used;
3541 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3546 update_stats(struct ofproto *ofproto, struct rule *rule,
3547 const struct odp_flow_stats *stats)
3549 if (stats->n_packets) {
3550 update_time(ofproto, rule, stats);
3551 rule->packet_count += stats->n_packets;
3552 rule->byte_count += stats->n_bytes;
3553 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3557 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3558 * in which no matching flow already exists in the flow table.
3560 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3561 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3562 * code as encoded by ofp_mkerr() on failure.
3564 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3567 add_flow(struct ofproto *p, struct ofconn *ofconn,
3568 const struct ofp_flow_mod *ofm, size_t n_actions)
3570 struct ofpbuf *packet;
3575 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3579 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3581 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3582 ntohs(ofm->priority))) {
3583 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3587 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3588 n_actions, ntohs(ofm->idle_timeout),
3589 ntohs(ofm->hard_timeout), ofm->cookie,
3590 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3591 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3592 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3595 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3596 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3600 in_port = UINT16_MAX;
3603 rule_insert(p, rule, packet, in_port);
3607 static struct rule *
3608 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3613 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3615 return rule_from_cls_rule(classifier_find_rule_exactly(
3616 &p->cls, &flow, wildcards,
3617 ntohs(ofm->priority)));
3621 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3622 struct rule *rule, const struct ofp_flow_mod *ofm)
3624 struct ofpbuf *packet;
3629 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3633 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3639 flow_extract(packet, 0, in_port, &flow);
3640 rule_execute(ofproto, rule, packet, &flow);
3645 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3647 struct modify_flows_cbdata {
3648 struct ofproto *ofproto;
3649 const struct ofp_flow_mod *ofm;
3654 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3655 size_t n_actions, struct rule *);
3656 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3658 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3659 * encoded by ofp_mkerr() on failure.
3661 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3664 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3665 const struct ofp_flow_mod *ofm, size_t n_actions)
3667 struct modify_flows_cbdata cbdata;
3668 struct cls_rule target;
3672 cbdata.n_actions = n_actions;
3673 cbdata.match = NULL;
3675 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3678 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3679 modify_flows_cb, &cbdata);
3681 /* This credits the packet to whichever flow happened to happened to
3682 * match last. That's weird. Maybe we should do a lookup for the
3683 * flow that actually matches the packet? Who knows. */
3684 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3687 return add_flow(p, ofconn, ofm, n_actions);
3691 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3692 * code as encoded by ofp_mkerr() on failure.
3694 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3697 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3698 struct ofp_flow_mod *ofm, size_t n_actions)
3700 struct rule *rule = find_flow_strict(p, ofm);
3701 if (rule && !rule_is_hidden(rule)) {
3702 modify_flow(p, ofm, n_actions, rule);
3703 return send_buffered_packet(p, ofconn, rule, ofm);
3705 return add_flow(p, ofconn, ofm, n_actions);
3709 /* Callback for modify_flows_loose(). */
3711 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3713 struct rule *rule = rule_from_cls_rule(rule_);
3714 struct modify_flows_cbdata *cbdata = cbdata_;
3716 if (!rule_is_hidden(rule)) {
3717 cbdata->match = rule;
3718 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3722 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3723 * been identified as a flow in 'p''s flow table to be modified, by changing
3724 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3725 * ofp_action[] structures). */
3727 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3728 size_t n_actions, struct rule *rule)
3730 size_t actions_len = n_actions * sizeof *rule->actions;
3732 rule->flow_cookie = ofm->cookie;
3734 /* If the actions are the same, do nothing. */
3735 if (n_actions == rule->n_actions
3736 && !memcmp(ofm->actions, rule->actions, actions_len))
3741 /* Replace actions. */
3742 free(rule->actions);
3743 rule->actions = xmemdup(ofm->actions, actions_len);
3744 rule->n_actions = n_actions;
3746 /* Make sure that the datapath gets updated properly. */
3747 if (rule->cr.wc.wildcards) {
3748 COVERAGE_INC(ofproto_mod_wc_flow);
3749 p->need_revalidate = true;
3751 rule_update_actions(p, rule);
3757 /* OFPFC_DELETE implementation. */
3759 struct delete_flows_cbdata {
3760 struct ofproto *ofproto;
3764 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3765 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3767 /* Implements OFPFC_DELETE. */
3769 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3771 struct delete_flows_cbdata cbdata;
3772 struct cls_rule target;
3775 cbdata.out_port = ofm->out_port;
3777 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3780 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3781 delete_flows_cb, &cbdata);
3784 /* Implements OFPFC_DELETE_STRICT. */
3786 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3788 struct rule *rule = find_flow_strict(p, ofm);
3790 delete_flow(p, rule, ofm->out_port);
3794 /* Callback for delete_flows_loose(). */
3796 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3798 struct rule *rule = rule_from_cls_rule(rule_);
3799 struct delete_flows_cbdata *cbdata = cbdata_;
3801 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3804 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3805 * been identified as a flow to delete from 'p''s flow table, by deleting the
3806 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3809 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3810 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3811 * specified 'out_port'. */
3813 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3815 if (rule_is_hidden(rule)) {
3819 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3823 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3824 rule_remove(p, rule);
3828 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3829 struct ofp_flow_mod *ofm)
3831 struct ofp_match orig_match;
3835 error = reject_slave_controller(ofconn, &ofm->header);
3839 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3840 sizeof *ofm->actions, &n_actions);
3845 /* We do not support the emergency flow cache. It will hopefully
3846 * get dropped from OpenFlow in the near future. */
3847 if (ofm->flags & htons(OFPFF_EMERG)) {
3848 /* There isn't a good fit for an error code, so just state that the
3849 * flow table is full. */
3850 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3853 /* Normalize ofp->match. If normalization actually changes anything, then
3854 * log the differences. */
3855 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3856 orig_match = ofm->match;
3857 normalize_match(&ofm->match);
3858 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3859 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3860 if (!VLOG_DROP_INFO(&normal_rl)) {
3861 char *old = ofp_match_to_literal_string(&orig_match);
3862 char *new = ofp_match_to_literal_string(&ofm->match);
3863 VLOG_INFO("%s: normalization changed ofp_match, details:",
3864 rconn_get_name(ofconn->rconn));
3865 VLOG_INFO(" pre: %s", old);
3866 VLOG_INFO("post: %s", new);
3872 if (!ofm->match.wildcards) {
3873 ofm->priority = htons(UINT16_MAX);
3876 error = validate_actions((const union ofp_action *) ofm->actions,
3877 n_actions, p->max_ports);
3882 switch (ntohs(ofm->command)) {
3884 return add_flow(p, ofconn, ofm, n_actions);
3887 return modify_flows_loose(p, ofconn, ofm, n_actions);
3889 case OFPFC_MODIFY_STRICT:
3890 return modify_flow_strict(p, ofconn, ofm, n_actions);
3893 delete_flows_loose(p, ofm);
3896 case OFPFC_DELETE_STRICT:
3897 delete_flow_strict(p, ofm);
3901 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3906 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3910 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3915 p->tun_id_from_cookie = !!msg->set;
3920 handle_role_request(struct ofproto *ofproto,
3921 struct ofconn *ofconn, struct nicira_header *msg)
3923 struct nx_role_request *nrr;
3924 struct nx_role_request *reply;
3928 if (ntohs(msg->header.length) != sizeof *nrr) {
3929 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3930 ntohs(msg->header.length), sizeof *nrr);
3931 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3933 nrr = (struct nx_role_request *) msg;
3935 if (ofconn->type != OFCONN_PRIMARY) {
3936 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3938 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3941 role = ntohl(nrr->role);
3942 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3943 && role != NX_ROLE_SLAVE) {
3944 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3946 /* There's no good error code for this. */
3947 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3950 if (role == NX_ROLE_MASTER) {
3951 struct ofconn *other;
3953 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3954 &ofproto->controllers) {
3955 if (other->role == NX_ROLE_MASTER) {
3956 other->role = NX_ROLE_SLAVE;
3960 ofconn->role = role;
3962 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3964 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3965 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3966 reply->role = htonl(role);
3967 queue_tx(buf, ofconn, ofconn->reply_counter);
3973 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3975 struct ofp_vendor_header *ovh = msg;
3976 struct nicira_header *nh;
3978 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3979 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3980 "(expected at least %zu)",
3981 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3982 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3984 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3985 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3987 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3988 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3989 "(expected at least %zu)",
3990 ntohs(ovh->header.length), sizeof(struct nicira_header));
3991 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3995 switch (ntohl(nh->subtype)) {
3996 case NXT_STATUS_REQUEST:
3997 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4000 case NXT_TUN_ID_FROM_COOKIE:
4001 return handle_tun_id_from_cookie(p, msg);
4003 case NXT_ROLE_REQUEST:
4004 return handle_role_request(p, ofconn, msg);
4007 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4011 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4013 struct ofp_header *ob;
4016 /* Currently, everything executes synchronously, so we can just
4017 * immediately send the barrier reply. */
4018 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4019 queue_tx(buf, ofconn, ofconn->reply_counter);
4024 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4025 struct ofpbuf *ofp_msg)
4027 struct ofp_header *oh = ofp_msg->data;
4030 COVERAGE_INC(ofproto_recv_openflow);
4032 case OFPT_ECHO_REQUEST:
4033 error = handle_echo_request(ofconn, oh);
4036 case OFPT_ECHO_REPLY:
4040 case OFPT_FEATURES_REQUEST:
4041 error = handle_features_request(p, ofconn, oh);
4044 case OFPT_GET_CONFIG_REQUEST:
4045 error = handle_get_config_request(p, ofconn, oh);
4048 case OFPT_SET_CONFIG:
4049 error = handle_set_config(p, ofconn, ofp_msg->data);
4052 case OFPT_PACKET_OUT:
4053 error = handle_packet_out(p, ofconn, ofp_msg->data);
4057 error = handle_port_mod(p, ofconn, oh);
4061 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4064 case OFPT_STATS_REQUEST:
4065 error = handle_stats_request(p, ofconn, oh);
4069 error = handle_vendor(p, ofconn, ofp_msg->data);
4072 case OFPT_BARRIER_REQUEST:
4073 error = handle_barrier_request(ofconn, oh);
4077 if (VLOG_IS_WARN_ENABLED()) {
4078 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4079 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4082 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4087 send_error_oh(ofconn, ofp_msg->data, error);
4092 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4094 struct odp_msg *msg = packet->data;
4096 struct ofpbuf payload;
4099 payload.data = msg + 1;
4100 payload.size = msg->length - sizeof *msg;
4101 flow_extract(&payload, msg->arg, msg->port, &flow);
4103 /* Check with in-band control to see if this packet should be sent
4104 * to the local port regardless of the flow table. */
4105 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4106 union odp_action action;
4108 memset(&action, 0, sizeof(action));
4109 action.output.type = ODPAT_OUTPUT;
4110 action.output.port = ODPP_LOCAL;
4111 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
4114 rule = lookup_valid_rule(p, &flow);
4116 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4117 struct ofport *port = port_array_get(&p->ports, msg->port);
4119 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4120 COVERAGE_INC(ofproto_no_packet_in);
4121 /* XXX install 'drop' flow entry */
4122 ofpbuf_delete(packet);
4126 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4129 COVERAGE_INC(ofproto_packet_in);
4130 send_packet_in(p, packet);
4134 if (rule->cr.wc.wildcards) {
4135 rule = rule_create_subrule(p, rule, &flow);
4136 rule_make_actions(p, rule, packet);
4138 if (!rule->may_install) {
4139 /* The rule is not installable, that is, we need to process every
4140 * packet, so process the current packet and set its actions into
4142 rule_make_actions(p, rule, packet);
4144 /* XXX revalidate rule if it needs it */
4148 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4150 * Extra-special case for fail-open mode.
4152 * We are in fail-open mode and the packet matched the fail-open rule,
4153 * but we are connected to a controller too. We should send the packet
4154 * up to the controller in the hope that it will try to set up a flow
4155 * and thereby allow us to exit fail-open.
4157 * See the top-level comment in fail-open.c for more information.
4159 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4160 DPIF_RECV_MSG_PADDING));
4163 ofpbuf_pull(packet, sizeof *msg);
4164 rule_execute(p, rule, packet, &flow);
4165 rule_reinstall(p, rule);
4169 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4171 struct odp_msg *msg = packet->data;
4173 switch (msg->type) {
4174 case _ODPL_ACTION_NR:
4175 COVERAGE_INC(ofproto_ctlr_action);
4176 send_packet_in(p, packet);
4179 case _ODPL_SFLOW_NR:
4181 ofproto_sflow_received(p->sflow, msg);
4183 ofpbuf_delete(packet);
4187 handle_odp_miss_msg(p, packet);
4191 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4197 /* Flow expiration. */
4199 struct expire_cbdata {
4200 struct ofproto *ofproto;
4204 static int ofproto_dp_max_idle(const struct ofproto *);
4205 static void ofproto_update_used(struct ofproto *);
4206 static void rule_expire(struct cls_rule *, void *cbdata);
4208 /* This function is called periodically by ofproto_run(). Its job is to
4209 * collect updates for the flows that have been installed into the datapath,
4210 * most importantly when they last were used, and then use that information to
4211 * expire flows that have not been used recently.
4213 * Returns the number of milliseconds after which it should be called again. */
4215 ofproto_expire(struct ofproto *ofproto)
4217 struct expire_cbdata cbdata;
4219 /* Update 'used' for each flow in the datapath. */
4220 ofproto_update_used(ofproto);
4222 /* Expire idle flows.
4224 * A wildcarded flow is idle only when all of its subrules have expired due
4225 * to becoming idle, so iterate through the exact-match flows first. */
4226 cbdata.ofproto = ofproto;
4227 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4228 classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
4229 classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
4231 /* Let the hook know that we're at a stable point: all outstanding data
4232 * in existing flows has been accounted to the account_cb. Thus, the
4233 * hook can now reasonably do operations that depend on having accurate
4234 * flow volume accounting (currently, that's just bond rebalancing). */
4235 if (ofproto->ofhooks->account_checkpoint_cb) {
4236 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4239 return MIN(cbdata.dp_max_idle, 1000);
4242 /* Update 'used' member of each flow currently installed into the datapath. */
4244 ofproto_update_used(struct ofproto *p)
4246 struct odp_flow *flows;
4251 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4256 for (i = 0; i < n_flows; i++) {
4257 struct odp_flow *f = &flows[i];
4260 rule = rule_from_cls_rule(
4261 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4263 if (rule && rule->installed) {
4264 update_time(p, rule, &f->stats);
4265 rule_account(p, rule, f->stats.n_bytes);
4267 /* There's a flow in the datapath that we know nothing about.
4269 COVERAGE_INC(ofproto_unexpected_rule);
4270 dpif_flow_del(p->dpif, f);
4277 /* Calculates and returns the number of milliseconds of idle time after which
4278 * flows should expire from the datapath and we should fold their statistics
4279 * into their parent rules in userspace. */
4281 ofproto_dp_max_idle(const struct ofproto *ofproto)
4284 * Idle time histogram.
4286 * Most of the time a switch has a relatively small number of flows. When
4287 * this is the case we might as well keep statistics for all of them in
4288 * userspace and to cache them in the kernel datapath for performance as
4291 * As the number of flows increases, the memory required to maintain
4292 * statistics about them in userspace and in the kernel becomes
4293 * significant. However, with a large number of flows it is likely that
4294 * only a few of them are "heavy hitters" that consume a large amount of
4295 * bandwidth. At this point, only heavy hitters are worth caching in the
4296 * kernel and maintaining in userspaces; other flows we can discard.
4298 * The technique used to compute the idle time is to build a histogram with
4299 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4300 * is installed in the kernel gets dropped in the appropriate bucket.
4301 * After the histogram has been built, we compute the cutoff so that only
4302 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4303 * cached. At least the most-recently-used bucket of flows is kept, so
4304 * actually an arbitrary number of flows can be kept in any given
4305 * expiration run (though the next run will delete most of those unless
4306 * they receive additional data).
4308 * This requires a second pass through the exact-match flows, in addition
4309 * to the pass made by ofproto_update_used(), because the former function
4310 * never looks at uninstallable flows.
4312 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4313 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4314 int buckets[N_BUCKETS] = { 0 };
4320 total = classifier_count_exact(&ofproto->cls);
4321 if (total <= 1000) {
4322 return N_BUCKETS * BUCKET_WIDTH;
4325 /* Build histogram. */
4327 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, struct rule, cr, &ofproto->cls) {
4328 long long int idle = now - rule->used;
4329 int bucket = (idle <= 0 ? 0
4330 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4331 : (unsigned int) idle / BUCKET_WIDTH);
4335 /* Find the first bucket whose flows should be expired. */
4336 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4337 if (buckets[bucket]) {
4340 subtotal += buckets[bucket++];
4341 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4346 if (VLOG_IS_DBG_ENABLED()) {
4350 ds_put_cstr(&s, "keep");
4351 for (i = 0; i < N_BUCKETS; i++) {
4353 ds_put_cstr(&s, ", drop");
4356 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4359 VLOG_INFO("%s: %s (msec:count)",
4360 dpif_name(ofproto->dpif), ds_cstr(&s));
4364 return bucket * BUCKET_WIDTH;
4368 rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
4370 if (ofproto->netflow && !is_controller_rule(rule) &&
4371 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4372 struct ofexpired expired;
4373 struct odp_flow odp_flow;
4375 /* Get updated flow stats.
4377 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4378 * updated TCP flags and (2) the dpif_flow_list_all() in
4379 * ofproto_update_used() zeroed TCP flags. */
4380 memset(&odp_flow, 0, sizeof odp_flow);
4381 if (rule->installed) {
4382 odp_flow.key = rule->cr.flow;
4383 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4384 dpif_flow_get(ofproto->dpif, &odp_flow);
4386 if (odp_flow.stats.n_packets) {
4387 update_time(ofproto, rule, &odp_flow.stats);
4388 netflow_flow_update_flags(&rule->nf_flow,
4389 odp_flow.stats.tcp_flags);
4393 expired.flow = rule->cr.flow;
4394 expired.packet_count = rule->packet_count +
4395 odp_flow.stats.n_packets;
4396 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4397 expired.used = rule->used;
4399 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4403 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4404 * rules, then delete it entirely.
4406 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4407 * the datapath and fold its statistics back into its super-rule.
4409 * (This is a callback function for classifier_for_each().) */
4411 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4413 struct expire_cbdata *cbdata = cbdata_;
4414 struct ofproto *ofproto = cbdata->ofproto;
4415 struct rule *rule = rule_from_cls_rule(cls_rule);
4416 long long int hard_expire, idle_expire, expire, now;
4418 /* Calculate OpenFlow expiration times for 'rule'. */
4419 hard_expire = (rule->hard_timeout
4420 ? rule->created + rule->hard_timeout * 1000
4422 idle_expire = (rule->idle_timeout
4423 && (rule->super || list_is_empty(&rule->list))
4424 ? rule->used + rule->idle_timeout * 1000
4426 expire = MIN(hard_expire, idle_expire);
4430 /* 'rule' has not expired according to OpenFlow rules. */
4431 if (!rule->cr.wc.wildcards) {
4432 if (now >= rule->used + cbdata->dp_max_idle) {
4433 /* This rule is idle, so drop it to free up resources. */
4435 /* It's not part of the OpenFlow flow table, so we can
4436 * delete it entirely and fold its statistics into its
4438 rule_remove(ofproto, rule);
4440 /* It is part of the OpenFlow flow table, so we have to
4441 * keep the rule but we can at least uninstall it from the
4443 rule_uninstall(ofproto, rule);
4446 /* Send NetFlow active timeout if appropriate. */
4447 rule_active_timeout(cbdata->ofproto, rule);
4451 /* 'rule' has expired according to OpenFlow rules. */
4452 COVERAGE_INC(ofproto_expired);
4454 /* Update stats. (This is a no-op if the rule expired due to an idle
4455 * timeout, because that only happens when the rule has no subrules
4457 if (rule->cr.wc.wildcards) {
4458 struct rule *subrule, *next;
4459 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list,
4461 rule_remove(cbdata->ofproto, subrule);
4464 rule_uninstall(cbdata->ofproto, rule);
4467 /* Get rid of the rule. */
4468 if (!rule_is_hidden(rule)) {
4469 send_flow_removed(cbdata->ofproto, rule, now,
4471 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4473 rule_remove(cbdata->ofproto, rule);
4478 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4480 struct rule *sub = rule_from_cls_rule(sub_);
4481 struct revalidate_cbdata *cbdata = cbdata_;
4483 if (cbdata->revalidate_all
4484 || (cbdata->revalidate_subrules && sub->super)
4485 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4486 revalidate_rule(cbdata->ofproto, sub);
4491 revalidate_rule(struct ofproto *p, struct rule *rule)
4493 const flow_t *flow = &rule->cr.flow;
4495 COVERAGE_INC(ofproto_revalidate_rule);
4498 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4500 rule_remove(p, rule);
4502 } else if (super != rule->super) {
4503 COVERAGE_INC(ofproto_revalidate_moved);
4504 list_remove(&rule->list);
4505 list_push_back(&super->list, &rule->list);
4506 rule->super = super;
4507 rule->hard_timeout = super->hard_timeout;
4508 rule->idle_timeout = super->idle_timeout;
4509 rule->created = super->created;
4514 rule_update_actions(p, rule);
4518 static struct ofpbuf *
4519 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4520 long long int now, uint8_t reason)
4522 struct ofp_flow_removed *ofr;
4524 long long int tdiff = now - rule->created;
4525 uint32_t sec = tdiff / 1000;
4526 uint32_t msec = tdiff - (sec * 1000);
4528 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4529 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4531 ofr->cookie = rule->flow_cookie;
4532 ofr->priority = htons(rule->cr.priority);
4533 ofr->reason = reason;
4534 ofr->duration_sec = htonl(sec);
4535 ofr->duration_nsec = htonl(msec * 1000000);
4536 ofr->idle_timeout = htons(rule->idle_timeout);
4537 ofr->packet_count = htonll(rule->packet_count);
4538 ofr->byte_count = htonll(rule->byte_count);
4544 send_flow_removed(struct ofproto *p, struct rule *rule,
4545 long long int now, uint8_t reason)
4547 struct ofconn *ofconn;
4548 struct ofconn *prev;
4549 struct ofpbuf *buf = NULL;
4551 if (!rule->send_flow_removed) {
4555 /* We limit the maximum number of queued flow expirations it by accounting
4556 * them under the counter for replies. That works because preventing
4557 * OpenFlow requests from being processed also prevents new flows from
4558 * being added (and expiring). (It also prevents processing OpenFlow
4559 * requests that would not add new flows, so it is imperfect.) */
4562 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4563 if (rconn_is_connected(ofconn->rconn)
4564 && ofconn_receives_async_msgs(ofconn)) {
4566 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4568 buf = compose_flow_removed(p, rule, now, reason);
4574 queue_tx(buf, prev, prev->reply_counter);
4578 /* pinsched callback for sending 'packet' on 'ofconn'. */
4580 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4582 struct ofconn *ofconn = ofconn_;
4584 rconn_send_with_limit(ofconn->rconn, packet,
4585 ofconn->packet_in_counter, 100);
4588 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4589 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4590 * packet scheduler for sending.
4592 * 'max_len' specifies the maximum number of bytes of the packet to send on
4593 * 'ofconn' (INT_MAX specifies no limit).
4595 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4596 * ownership is transferred to this function. */
4598 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4601 struct ofproto *ofproto = ofconn->ofproto;
4602 struct ofp_packet_in *opi = packet->data;
4603 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4604 int send_len, trim_size;
4608 if (opi->reason == OFPR_ACTION) {
4609 buffer_id = UINT32_MAX;
4610 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4611 buffer_id = pktbuf_get_null();
4612 } else if (!ofconn->pktbuf) {
4613 buffer_id = UINT32_MAX;
4615 struct ofpbuf payload;
4616 payload.data = opi->data;
4617 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4618 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4621 /* Figure out how much of the packet to send. */
4622 send_len = ntohs(opi->total_len);
4623 if (buffer_id != UINT32_MAX) {
4624 send_len = MIN(send_len, ofconn->miss_send_len);
4626 send_len = MIN(send_len, max_len);
4628 /* Adjust packet length and clone if necessary. */
4629 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4631 packet = ofpbuf_clone_data(packet->data, trim_size);
4634 packet->size = trim_size;
4637 /* Update packet headers. */
4638 opi->buffer_id = htonl(buffer_id);
4639 update_openflow_length(packet);
4641 /* Hand over to packet scheduler. It might immediately call into
4642 * do_send_packet_in() or it might buffer it for a while (until a later
4643 * call to pinsched_run()). */
4644 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4645 packet, do_send_packet_in, ofconn);
4648 /* Replace struct odp_msg header in 'packet' by equivalent struct
4649 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4650 * returned by dpif_recv()).
4652 * The conversion is not complete: the caller still needs to trim any unneeded
4653 * payload off the end of the buffer, set the length in the OpenFlow header,
4654 * and set buffer_id. Those require us to know the controller settings and so
4655 * must be done on a per-controller basis.
4657 * Returns the maximum number of bytes of the packet that should be sent to
4658 * the controller (INT_MAX if no limit). */
4660 do_convert_to_packet_in(struct ofpbuf *packet)
4662 struct odp_msg *msg = packet->data;
4663 struct ofp_packet_in *opi;
4669 /* Extract relevant header fields */
4670 if (msg->type == _ODPL_ACTION_NR) {
4671 reason = OFPR_ACTION;
4674 reason = OFPR_NO_MATCH;
4677 total_len = msg->length - sizeof *msg;
4678 in_port = odp_port_to_ofp_port(msg->port);
4680 /* Repurpose packet buffer by overwriting header. */
4681 ofpbuf_pull(packet, sizeof(struct odp_msg));
4682 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4683 opi->header.version = OFP_VERSION;
4684 opi->header.type = OFPT_PACKET_IN;
4685 opi->total_len = htons(total_len);
4686 opi->in_port = htons(in_port);
4687 opi->reason = reason;
4692 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4693 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4694 * as necessary according to their individual configurations.
4696 * 'packet' must have sufficient headroom to convert it into a struct
4697 * ofp_packet_in (e.g. as returned by dpif_recv()).
4699 * Takes ownership of 'packet'. */
4701 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4703 struct ofconn *ofconn, *prev;
4706 max_len = do_convert_to_packet_in(packet);
4709 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4710 if (ofconn_receives_async_msgs(ofconn)) {
4712 schedule_packet_in(prev, packet, max_len, true);
4718 schedule_packet_in(prev, packet, max_len, false);
4720 ofpbuf_delete(packet);
4725 pick_datapath_id(const struct ofproto *ofproto)
4727 const struct ofport *port;
4729 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4731 uint8_t ea[ETH_ADDR_LEN];
4734 error = netdev_get_etheraddr(port->netdev, ea);
4736 return eth_addr_to_uint64(ea);
4738 VLOG_WARN("could not get MAC address for %s (%s)",
4739 netdev_get_name(port->netdev), strerror(error));
4741 return ofproto->fallback_dpid;
4745 pick_fallback_dpid(void)
4747 uint8_t ea[ETH_ADDR_LEN];
4748 eth_addr_nicira_random(ea);
4749 return eth_addr_to_uint64(ea);
4753 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4754 struct odp_actions *actions, tag_type *tags,
4755 uint16_t *nf_output_iface, void *ofproto_)
4757 struct ofproto *ofproto = ofproto_;
4760 /* Drop frames for reserved multicast addresses. */
4761 if (eth_addr_is_reserved(flow->dl_dst)) {
4765 /* Learn source MAC (but don't try to learn from revalidation). */
4766 if (packet != NULL) {
4767 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4769 GRAT_ARP_LOCK_NONE);
4771 /* The log messages here could actually be useful in debugging,
4772 * so keep the rate limit relatively high. */
4773 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4774 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4775 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4776 ofproto_revalidate(ofproto, rev_tag);
4780 /* Determine output port. */
4781 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4784 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4785 } else if (out_port != flow->in_port) {
4786 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4787 *nf_output_iface = out_port;
4795 static const struct ofhooks default_ofhooks = {
4797 default_normal_ofhook_cb,