2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
28 #include "classifier.h"
30 #include "discovery.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
42 #include "ofp-print.h"
44 #include "ofproto-sflow.h"
46 #include "openflow/nicira-ext.h"
47 #include "openflow/openflow.h"
48 #include "openvswitch/datapath-protocol.h"
52 #include "poll-loop.h"
56 #include "stream-ssl.h"
64 VLOG_DEFINE_THIS_MODULE(ofproto);
66 #include "sflow_api.h"
69 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
70 struct netdev *netdev;
71 struct ofp_phy_port opp; /* In host byte order. */
75 static void ofport_free(struct ofport *);
76 static void hton_ofp_phy_port(struct ofp_phy_port *);
78 static int xlate_actions(const union ofp_action *in, size_t n_in,
79 const struct flow *, struct ofproto *,
80 const struct ofpbuf *packet,
81 struct odp_actions *out, tag_type *tags,
82 bool *may_set_up_flow, uint16_t *nf_output_iface);
87 ovs_be64 flow_cookie; /* Controller-issued identifier. */
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 bool send_flow_removed; /* Send a flow removed message? */
91 long long int used; /* Last-used time (0 if never used). */
92 long long int created; /* Creation time. */
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
95 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
96 tag_type tags; /* Tags (set only by hooks). */
97 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * 'n_actions' is the number of elements in the 'actions' array. A single
113 * action may take up more more than one element's worth of space.
115 * A subrule has no actions (it uses the super-rule's actions). */
117 union ofp_action *actions;
121 * A super-rule with wildcard fields never has ODP actions (since the
122 * datapath only supports exact-match flows). */
123 bool installed; /* Installed in datapath? */
124 bool may_install; /* True ordinarily; false if actions must
125 * be reassessed for every packet. */
127 union odp_action *odp_actions;
131 rule_is_hidden(const struct rule *rule)
133 /* Subrules are merely an implementation detail, so hide them from the
135 if (rule->super != NULL) {
139 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
140 * (e.g. by in-band control) and are intentionally hidden from the
142 if (rule->cr.priority > UINT16_MAX) {
149 static struct rule *rule_create(struct ofproto *, struct rule *super,
150 const union ofp_action *, size_t n_actions,
151 uint16_t idle_timeout, uint16_t hard_timeout,
152 ovs_be64 flow_cookie, bool send_flow_removed);
153 static void rule_free(struct rule *);
154 static void rule_destroy(struct ofproto *, struct rule *);
155 static struct rule *rule_from_cls_rule(const struct cls_rule *);
156 static void rule_insert(struct ofproto *, struct rule *,
157 struct ofpbuf *packet, uint16_t in_port);
158 static void rule_remove(struct ofproto *, struct rule *);
159 static bool rule_make_actions(struct ofproto *, struct rule *,
160 const struct ofpbuf *packet);
161 static void rule_install(struct ofproto *, struct rule *,
162 struct rule *displaced_rule);
163 static void rule_uninstall(struct ofproto *, struct rule *);
164 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 static void send_flow_removed(struct ofproto *, struct rule *, uint8_t reason);
167 /* ofproto supports two kinds of OpenFlow connections:
169 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
170 * maintains persistent connections to these controllers and by default
171 * sends them asynchronous messages such as packet-ins.
173 * - "Service" connections, e.g. from ovs-ofctl. When these connections
174 * drop, it is the other side's responsibility to reconnect them if
175 * necessary. ofproto does not send them asynchronous messages by default.
177 * Currently, active (tcp, ssl, unix) connections are always "primary"
178 * connections and passive (ptcp, pssl, punix) connections are always "service"
179 * connections. There is no inherent reason for this, but it reflects the
183 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
184 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
187 /* A listener for incoming OpenFlow "service" connections. */
189 struct hmap_node node; /* In struct ofproto's "services" hmap. */
190 struct pvconn *pvconn; /* OpenFlow connection listener. */
192 /* These are not used by ofservice directly. They are settings for
193 * accepted "struct ofconn"s from the pvconn. */
194 int probe_interval; /* Max idle time before probing, in seconds. */
195 int rate_limit; /* Max packet-in rate in packets per second. */
196 int burst_limit; /* Limit on accumulating packet credits. */
199 static struct ofservice *ofservice_lookup(struct ofproto *,
201 static int ofservice_create(struct ofproto *,
202 const struct ofproto_controller *);
203 static void ofservice_reconfigure(struct ofservice *,
204 const struct ofproto_controller *);
205 static void ofservice_destroy(struct ofproto *, struct ofservice *);
207 /* An OpenFlow connection. */
209 struct ofproto *ofproto; /* The ofproto that owns this connection. */
210 struct list node; /* In struct ofproto's "all_conns" list. */
211 struct rconn *rconn; /* OpenFlow connection. */
212 enum ofconn_type type; /* Type. */
213 int flow_format; /* One of NXFF_*. */
215 /* OFPT_PACKET_IN related data. */
216 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
217 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
218 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
219 int miss_send_len; /* Bytes to send of buffered packets. */
221 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
222 * requests, and the maximum number before we stop reading OpenFlow
224 #define OFCONN_REPLY_MAX 100
225 struct rconn_packet_counter *reply_counter;
227 /* type == OFCONN_PRIMARY only. */
228 enum nx_role role; /* Role. */
229 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
230 struct discovery *discovery; /* Controller discovery object, if enabled. */
231 struct status_category *ss; /* Switch status category. */
232 enum ofproto_band band; /* In-band or out-of-band? */
235 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
236 * "schedulers" array. Their values are 0 and 1, and their meanings and values
237 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
238 * case anything ever changes, check their values here. */
239 #define N_SCHEDULERS 2
240 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
241 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
242 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
243 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
245 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
247 static void ofconn_destroy(struct ofconn *);
248 static void ofconn_run(struct ofconn *);
249 static void ofconn_wait(struct ofconn *);
250 static bool ofconn_receives_async_msgs(const struct ofconn *);
251 static char *ofconn_make_name(const struct ofproto *, const char *target);
252 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
254 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
255 struct rconn_packet_counter *counter);
257 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
258 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
262 uint64_t datapath_id; /* Datapath ID. */
263 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
264 char *mfr_desc; /* Manufacturer. */
265 char *hw_desc; /* Hardware. */
266 char *sw_desc; /* Software version. */
267 char *serial_desc; /* Serial number. */
268 char *dp_desc; /* Datapath description. */
272 struct netdev_monitor *netdev_monitor;
273 struct hmap ports; /* Contains "struct ofport"s. */
274 struct shash port_by_name;
278 struct switch_status *switch_status;
279 struct fail_open *fail_open;
280 struct netflow *netflow;
281 struct ofproto_sflow *sflow;
283 /* In-band control. */
284 struct in_band *in_band;
285 long long int next_in_band_update;
286 struct sockaddr_in *extra_in_band_remotes;
287 size_t n_extra_remotes;
290 struct classifier cls;
291 bool need_revalidate;
292 long long int next_expiration;
293 struct tag_set revalidate_set;
295 /* OpenFlow connections. */
296 struct hmap controllers; /* Controller "struct ofconn"s. */
297 struct list all_conns; /* Contains "struct ofconn"s. */
298 enum ofproto_fail_mode fail_mode;
300 /* OpenFlow listeners. */
301 struct hmap services; /* Contains "struct ofservice"s. */
302 struct pvconn **snoops;
305 /* Hooks for ovs-vswitchd. */
306 const struct ofhooks *ofhooks;
309 /* Used by default ofhooks. */
310 struct mac_learning *ml;
313 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
315 static const struct ofhooks default_ofhooks;
317 static uint64_t pick_datapath_id(const struct ofproto *);
318 static uint64_t pick_fallback_dpid(void);
320 static int ofproto_expire(struct ofproto *);
322 static void update_stats(struct ofproto *, struct rule *,
323 const struct odp_flow_stats *);
324 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
325 static void revalidate_cb(struct cls_rule *rule_, void *p_);
327 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
329 static void handle_openflow(struct ofconn *, struct ofpbuf *);
331 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
332 static void update_port(struct ofproto *, const char *devname);
333 static int init_ports(struct ofproto *);
334 static void reinit_ports(struct ofproto *);
337 ofproto_create(const char *datapath, const char *datapath_type,
338 const struct ofhooks *ofhooks, void *aux,
339 struct ofproto **ofprotop)
341 struct odp_stats stats;
348 /* Connect to datapath and start listening for messages. */
349 error = dpif_open(datapath, datapath_type, &dpif);
351 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
354 error = dpif_get_dp_stats(dpif, &stats);
356 VLOG_ERR("failed to obtain stats for datapath %s: %s",
357 datapath, strerror(error));
361 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
363 VLOG_ERR("failed to listen on datapath %s: %s",
364 datapath, strerror(error));
368 dpif_flow_flush(dpif);
369 dpif_recv_purge(dpif);
371 /* Initialize settings. */
372 p = xzalloc(sizeof *p);
373 p->fallback_dpid = pick_fallback_dpid();
374 p->datapath_id = p->fallback_dpid;
375 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
376 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
377 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
378 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
379 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
381 /* Initialize datapath. */
383 p->netdev_monitor = netdev_monitor_create();
384 hmap_init(&p->ports);
385 shash_init(&p->port_by_name);
386 p->max_ports = stats.max_ports;
388 /* Initialize submodules. */
389 p->switch_status = switch_status_create(p);
395 /* Initialize flow table. */
396 classifier_init(&p->cls);
397 p->need_revalidate = false;
398 p->next_expiration = time_msec() + 1000;
399 tag_set_init(&p->revalidate_set);
401 /* Initialize OpenFlow connections. */
402 list_init(&p->all_conns);
403 hmap_init(&p->controllers);
404 hmap_init(&p->services);
408 /* Initialize hooks. */
410 p->ofhooks = ofhooks;
414 p->ofhooks = &default_ofhooks;
416 p->ml = mac_learning_create();
419 /* Pick final datapath ID. */
420 p->datapath_id = pick_datapath_id(p);
421 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
428 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
430 uint64_t old_dpid = p->datapath_id;
431 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
432 if (p->datapath_id != old_dpid) {
433 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
435 /* Force all active connections to reconnect, since there is no way to
436 * notify a controller that the datapath ID has changed. */
437 ofproto_reconnect_controllers(p);
442 is_discovery_controller(const struct ofproto_controller *c)
444 return !strcmp(c->target, "discover");
448 is_in_band_controller(const struct ofproto_controller *c)
450 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
453 /* Creates a new controller in 'ofproto'. Some of the settings are initially
454 * drawn from 'c', but update_controller() needs to be called later to finish
455 * the new ofconn's configuration. */
457 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
459 struct discovery *discovery;
460 struct ofconn *ofconn;
462 if (is_discovery_controller(c)) {
463 int error = discovery_create(c->accept_re, c->update_resolv_conf,
464 ofproto->dpif, ofproto->switch_status,
473 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
474 ofconn->pktbuf = pktbuf_create();
475 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
477 ofconn->discovery = discovery;
479 char *name = ofconn_make_name(ofproto, c->target);
480 rconn_connect(ofconn->rconn, c->target, name);
483 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
484 hash_string(c->target, 0));
487 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
488 * target or turn discovery on or off (these are done by creating new ofconns
489 * and deleting old ones), but it can update the rest of an ofconn's
492 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
496 ofconn->band = (is_in_band_controller(c)
497 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
499 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
501 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
502 rconn_set_probe_interval(ofconn->rconn, probe_interval);
504 if (ofconn->discovery) {
505 discovery_set_update_resolv_conf(ofconn->discovery,
506 c->update_resolv_conf);
507 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
510 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
514 ofconn_get_target(const struct ofconn *ofconn)
516 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
519 static struct ofconn *
520 find_controller_by_target(struct ofproto *ofproto, const char *target)
522 struct ofconn *ofconn;
524 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
525 hash_string(target, 0), &ofproto->controllers) {
526 if (!strcmp(ofconn_get_target(ofconn), target)) {
534 update_in_band_remotes(struct ofproto *ofproto)
536 const struct ofconn *ofconn;
537 struct sockaddr_in *addrs;
538 size_t max_addrs, n_addrs;
542 /* Allocate enough memory for as many remotes as we could possibly have. */
543 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
544 addrs = xmalloc(max_addrs * sizeof *addrs);
547 /* Add all the remotes. */
549 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
550 struct sockaddr_in *sin = &addrs[n_addrs];
552 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
556 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
557 if (sin->sin_addr.s_addr) {
558 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
561 if (ofconn->discovery) {
565 for (i = 0; i < ofproto->n_extra_remotes; i++) {
566 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
569 /* Create or update or destroy in-band.
571 * Ordinarily we only enable in-band if there's at least one remote
572 * address, but discovery needs the in-band rules for DHCP to be installed
573 * even before we know any remote addresses. */
574 if (n_addrs || discovery) {
575 if (!ofproto->in_band) {
576 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
579 if (ofproto->in_band) {
580 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
582 ofproto->next_in_band_update = time_msec() + 1000;
584 in_band_destroy(ofproto->in_band);
585 ofproto->in_band = NULL;
593 update_fail_open(struct ofproto *p)
595 struct ofconn *ofconn;
597 if (!hmap_is_empty(&p->controllers)
598 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
599 struct rconn **rconns;
603 p->fail_open = fail_open_create(p, p->switch_status);
607 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
608 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
609 rconns[n++] = ofconn->rconn;
612 fail_open_set_controllers(p->fail_open, rconns, n);
613 /* p->fail_open takes ownership of 'rconns'. */
615 fail_open_destroy(p->fail_open);
621 ofproto_set_controllers(struct ofproto *p,
622 const struct ofproto_controller *controllers,
623 size_t n_controllers)
625 struct shash new_controllers;
626 struct ofconn *ofconn, *next_ofconn;
627 struct ofservice *ofservice, *next_ofservice;
631 /* Create newly configured controllers and services.
632 * Create a name to ofproto_controller mapping in 'new_controllers'. */
633 shash_init(&new_controllers);
634 for (i = 0; i < n_controllers; i++) {
635 const struct ofproto_controller *c = &controllers[i];
637 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
638 if (!find_controller_by_target(p, c->target)) {
639 add_controller(p, c);
641 } else if (!pvconn_verify_name(c->target)) {
642 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
646 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
647 dpif_name(p->dpif), c->target);
651 shash_add_once(&new_controllers, c->target, &controllers[i]);
654 /* Delete controllers that are no longer configured.
655 * Update configuration of all now-existing controllers. */
657 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
658 struct ofproto_controller *c;
660 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
662 ofconn_destroy(ofconn);
664 update_controller(ofconn, c);
671 /* Delete services that are no longer configured.
672 * Update configuration of all now-existing services. */
673 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
674 struct ofproto_controller *c;
676 c = shash_find_data(&new_controllers,
677 pvconn_get_name(ofservice->pvconn));
679 ofservice_destroy(p, ofservice);
681 ofservice_reconfigure(ofservice, c);
685 shash_destroy(&new_controllers);
687 update_in_band_remotes(p);
690 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
691 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
692 struct ofconn, hmap_node);
693 ofconn->ss = switch_status_register(p->switch_status, "remote",
694 rconn_status_cb, ofconn->rconn);
699 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
701 p->fail_mode = fail_mode;
705 /* Drops the connections between 'ofproto' and all of its controllers, forcing
706 * them to reconnect. */
708 ofproto_reconnect_controllers(struct ofproto *ofproto)
710 struct ofconn *ofconn;
712 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
713 rconn_reconnect(ofconn->rconn);
718 any_extras_changed(const struct ofproto *ofproto,
719 const struct sockaddr_in *extras, size_t n)
723 if (n != ofproto->n_extra_remotes) {
727 for (i = 0; i < n; i++) {
728 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
729 const struct sockaddr_in *new = &extras[i];
731 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
732 old->sin_port != new->sin_port) {
740 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
741 * in-band control should guarantee access, in the same way that in-band
742 * control guarantees access to OpenFlow controllers. */
744 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
745 const struct sockaddr_in *extras, size_t n)
747 if (!any_extras_changed(ofproto, extras, n)) {
751 free(ofproto->extra_in_band_remotes);
752 ofproto->n_extra_remotes = n;
753 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
755 update_in_band_remotes(ofproto);
759 ofproto_set_desc(struct ofproto *p,
760 const char *mfr_desc, const char *hw_desc,
761 const char *sw_desc, const char *serial_desc,
764 struct ofp_desc_stats *ods;
767 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
768 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
769 sizeof ods->mfr_desc);
772 p->mfr_desc = xstrdup(mfr_desc);
775 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
776 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
777 sizeof ods->hw_desc);
780 p->hw_desc = xstrdup(hw_desc);
783 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
784 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
785 sizeof ods->sw_desc);
788 p->sw_desc = xstrdup(sw_desc);
791 if (strlen(serial_desc) >= sizeof ods->serial_num) {
792 VLOG_WARN("truncating serial_desc, must be less than %zu "
794 sizeof ods->serial_num);
796 free(p->serial_desc);
797 p->serial_desc = xstrdup(serial_desc);
800 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
801 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
802 sizeof ods->dp_desc);
805 p->dp_desc = xstrdup(dp_desc);
810 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
811 const struct svec *svec)
813 struct pvconn **pvconns = *pvconnsp;
814 size_t n_pvconns = *n_pvconnsp;
818 for (i = 0; i < n_pvconns; i++) {
819 pvconn_close(pvconns[i]);
823 pvconns = xmalloc(svec->n * sizeof *pvconns);
825 for (i = 0; i < svec->n; i++) {
826 const char *name = svec->names[i];
827 struct pvconn *pvconn;
830 error = pvconn_open(name, &pvconn);
832 pvconns[n_pvconns++] = pvconn;
834 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
842 *n_pvconnsp = n_pvconns;
848 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
850 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
854 ofproto_set_netflow(struct ofproto *ofproto,
855 const struct netflow_options *nf_options)
857 if (nf_options && nf_options->collectors.n) {
858 if (!ofproto->netflow) {
859 ofproto->netflow = netflow_create();
861 return netflow_set_options(ofproto->netflow, nf_options);
863 netflow_destroy(ofproto->netflow);
864 ofproto->netflow = NULL;
870 ofproto_set_sflow(struct ofproto *ofproto,
871 const struct ofproto_sflow_options *oso)
873 struct ofproto_sflow *os = ofproto->sflow;
876 struct ofport *ofport;
878 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
879 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
880 ofproto_sflow_add_port(os, ofport->odp_port,
881 netdev_get_name(ofport->netdev));
884 ofproto_sflow_set_options(os, oso);
886 ofproto_sflow_destroy(os);
887 ofproto->sflow = NULL;
892 ofproto_get_datapath_id(const struct ofproto *ofproto)
894 return ofproto->datapath_id;
898 ofproto_has_primary_controller(const struct ofproto *ofproto)
900 return !hmap_is_empty(&ofproto->controllers);
903 enum ofproto_fail_mode
904 ofproto_get_fail_mode(const struct ofproto *p)
910 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
914 for (i = 0; i < ofproto->n_snoops; i++) {
915 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
920 ofproto_destroy(struct ofproto *p)
922 struct ofservice *ofservice, *next_ofservice;
923 struct ofconn *ofconn, *next_ofconn;
924 struct ofport *ofport, *next_ofport;
931 /* Destroy fail-open and in-band early, since they touch the classifier. */
932 fail_open_destroy(p->fail_open);
935 in_band_destroy(p->in_band);
937 free(p->extra_in_band_remotes);
939 ofproto_flush_flows(p);
940 classifier_destroy(&p->cls);
942 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
943 ofconn_destroy(ofconn);
945 hmap_destroy(&p->controllers);
948 netdev_monitor_destroy(p->netdev_monitor);
949 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
950 hmap_remove(&p->ports, &ofport->hmap_node);
953 shash_destroy(&p->port_by_name);
955 switch_status_destroy(p->switch_status);
956 netflow_destroy(p->netflow);
957 ofproto_sflow_destroy(p->sflow);
959 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
960 ofservice_destroy(p, ofservice);
962 hmap_destroy(&p->services);
964 for (i = 0; i < p->n_snoops; i++) {
965 pvconn_close(p->snoops[i]);
969 mac_learning_destroy(p->ml);
974 free(p->serial_desc);
977 hmap_destroy(&p->ports);
983 ofproto_run(struct ofproto *p)
985 int error = ofproto_run1(p);
987 error = ofproto_run2(p, false);
993 process_port_change(struct ofproto *ofproto, int error, char *devname)
995 if (error == ENOBUFS) {
996 reinit_ports(ofproto);
998 update_port(ofproto, devname);
1003 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1004 * means that 'ofconn' is more interesting for monitoring than a lower return
1007 snoop_preference(const struct ofconn *ofconn)
1009 switch (ofconn->role) {
1010 case NX_ROLE_MASTER:
1017 /* Shouldn't happen. */
1022 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1023 * Connects this vconn to a controller. */
1025 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1027 struct ofconn *ofconn, *best;
1029 /* Pick a controller for monitoring. */
1031 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1032 if (ofconn->type == OFCONN_PRIMARY
1033 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1039 rconn_add_monitor(best->rconn, vconn);
1041 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1047 ofproto_run1(struct ofproto *p)
1049 struct ofconn *ofconn, *next_ofconn;
1050 struct ofservice *ofservice;
1055 if (shash_is_empty(&p->port_by_name)) {
1059 for (i = 0; i < 50; i++) {
1062 error = dpif_recv(p->dpif, &buf);
1064 if (error == ENODEV) {
1065 /* Someone destroyed the datapath behind our back. The caller
1066 * better destroy us and give up, because we're just going to
1067 * spin from here on out. */
1068 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1069 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1070 dpif_name(p->dpif));
1076 handle_odp_msg(p, buf);
1079 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1080 process_port_change(p, error, devname);
1082 while ((error = netdev_monitor_poll(p->netdev_monitor,
1083 &devname)) != EAGAIN) {
1084 process_port_change(p, error, devname);
1088 if (time_msec() >= p->next_in_band_update) {
1089 update_in_band_remotes(p);
1091 in_band_run(p->in_band);
1094 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1098 /* Fail-open maintenance. Do this after processing the ofconns since
1099 * fail-open checks the status of the controller rconn. */
1101 fail_open_run(p->fail_open);
1104 HMAP_FOR_EACH (ofservice, node, &p->services) {
1105 struct vconn *vconn;
1108 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1110 struct rconn *rconn;
1113 rconn = rconn_create(ofservice->probe_interval, 0);
1114 name = ofconn_make_name(p, vconn_get_name(vconn));
1115 rconn_connect_unreliably(rconn, vconn, name);
1118 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1119 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1120 ofservice->burst_limit);
1121 } else if (retval != EAGAIN) {
1122 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1126 for (i = 0; i < p->n_snoops; i++) {
1127 struct vconn *vconn;
1130 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1132 add_snooper(p, vconn);
1133 } else if (retval != EAGAIN) {
1134 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1138 if (time_msec() >= p->next_expiration) {
1139 int delay = ofproto_expire(p);
1140 p->next_expiration = time_msec() + delay;
1141 COVERAGE_INC(ofproto_expiration);
1145 netflow_run(p->netflow);
1148 ofproto_sflow_run(p->sflow);
1154 struct revalidate_cbdata {
1155 struct ofproto *ofproto;
1156 bool revalidate_all; /* Revalidate all exact-match rules? */
1157 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1158 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1162 ofproto_run2(struct ofproto *p, bool revalidate_all)
1164 if (p->need_revalidate || revalidate_all
1165 || !tag_set_is_empty(&p->revalidate_set)) {
1166 struct revalidate_cbdata cbdata;
1168 cbdata.revalidate_all = revalidate_all;
1169 cbdata.revalidate_subrules = p->need_revalidate;
1170 cbdata.revalidate_set = p->revalidate_set;
1171 tag_set_init(&p->revalidate_set);
1172 COVERAGE_INC(ofproto_revalidate);
1173 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1174 p->need_revalidate = false;
1181 ofproto_wait(struct ofproto *p)
1183 struct ofservice *ofservice;
1184 struct ofconn *ofconn;
1187 dpif_recv_wait(p->dpif);
1188 dpif_port_poll_wait(p->dpif);
1189 netdev_monitor_poll_wait(p->netdev_monitor);
1190 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1191 ofconn_wait(ofconn);
1194 poll_timer_wait_until(p->next_in_band_update);
1195 in_band_wait(p->in_band);
1198 fail_open_wait(p->fail_open);
1201 ofproto_sflow_wait(p->sflow);
1203 if (!tag_set_is_empty(&p->revalidate_set)) {
1204 poll_immediate_wake();
1206 if (p->need_revalidate) {
1207 /* Shouldn't happen, but if it does just go around again. */
1208 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1209 poll_immediate_wake();
1210 } else if (p->next_expiration != LLONG_MAX) {
1211 poll_timer_wait_until(p->next_expiration);
1213 HMAP_FOR_EACH (ofservice, node, &p->services) {
1214 pvconn_wait(ofservice->pvconn);
1216 for (i = 0; i < p->n_snoops; i++) {
1217 pvconn_wait(p->snoops[i]);
1222 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1224 tag_set_add(&ofproto->revalidate_set, tag);
1228 ofproto_get_revalidate_set(struct ofproto *ofproto)
1230 return &ofproto->revalidate_set;
1234 ofproto_is_alive(const struct ofproto *p)
1236 return !hmap_is_empty(&p->controllers);
1239 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1241 * This is almost the same as calling dpif_port_del() directly on the
1242 * datapath, but it also makes 'ofproto' close its open netdev for the port
1243 * (if any). This makes it possible to create a new netdev of a different
1244 * type under the same name, which otherwise the netdev library would refuse
1245 * to do because of the conflict. (The netdev would eventually get closed on
1246 * the next trip through ofproto_run(), but this interface is more direct.)
1248 * Returns 0 if successful, otherwise a positive errno. */
1250 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1252 struct ofport *ofport = get_port(ofproto, odp_port);
1253 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1256 error = dpif_port_del(ofproto->dpif, odp_port);
1258 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1259 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1260 } else if (ofport) {
1261 /* 'name' is ofport->opp.name and update_port() is going to destroy
1262 * 'ofport'. Just in case update_port() refers to 'name' after it
1263 * destroys 'ofport', make a copy of it around the update_port()
1265 char *devname = xstrdup(name);
1266 update_port(ofproto, devname);
1272 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1273 * true if 'odp_port' exists and should be included, false otherwise. */
1275 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1277 struct ofport *ofport = get_port(ofproto, odp_port);
1278 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1282 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1283 const union ofp_action *actions, size_t n_actions,
1284 const struct ofpbuf *packet)
1286 struct odp_actions odp_actions;
1289 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1295 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1297 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
1302 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1303 const union ofp_action *actions, size_t n_actions,
1307 rule = rule_create(p, NULL, actions, n_actions,
1308 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1310 rule->cr = *cls_rule;
1311 rule_insert(p, rule, NULL, 0);
1315 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1319 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1322 rule_remove(ofproto, rule);
1327 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1329 struct rule *rule = rule_from_cls_rule(rule_);
1330 struct ofproto *ofproto = ofproto_;
1332 /* Mark the flow as not installed, even though it might really be
1333 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1334 * There is no point in uninstalling it individually since we are about to
1335 * blow away all the flows with dpif_flow_flush(). */
1336 rule->installed = false;
1338 rule_remove(ofproto, rule);
1342 ofproto_flush_flows(struct ofproto *ofproto)
1344 COVERAGE_INC(ofproto_flush);
1345 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1346 dpif_flow_flush(ofproto->dpif);
1347 if (ofproto->in_band) {
1348 in_band_flushed(ofproto->in_band);
1350 if (ofproto->fail_open) {
1351 fail_open_flushed(ofproto->fail_open);
1356 reinit_ports(struct ofproto *p)
1358 struct svec devnames;
1359 struct ofport *ofport;
1360 struct odp_port *odp_ports;
1364 COVERAGE_INC(ofproto_reinit_ports);
1366 svec_init(&devnames);
1367 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1368 svec_add (&devnames, (char *) ofport->opp.name);
1370 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1371 for (i = 0; i < n_odp_ports; i++) {
1372 svec_add (&devnames, odp_ports[i].devname);
1376 svec_sort_unique(&devnames);
1377 for (i = 0; i < devnames.n; i++) {
1378 update_port(p, devnames.names[i]);
1380 svec_destroy(&devnames);
1383 static struct ofport *
1384 make_ofport(const struct odp_port *odp_port)
1386 struct netdev_options netdev_options;
1387 enum netdev_flags flags;
1388 struct ofport *ofport;
1389 struct netdev *netdev;
1392 memset(&netdev_options, 0, sizeof netdev_options);
1393 netdev_options.name = odp_port->devname;
1394 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1396 error = netdev_open(&netdev_options, &netdev);
1398 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1399 "cannot be opened (%s)",
1400 odp_port->devname, odp_port->port,
1401 odp_port->devname, strerror(error));
1405 ofport = xmalloc(sizeof *ofport);
1406 ofport->netdev = netdev;
1407 ofport->odp_port = odp_port->port;
1408 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1409 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1410 memcpy(ofport->opp.name, odp_port->devname,
1411 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1412 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1414 netdev_get_flags(netdev, &flags);
1415 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1417 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1419 netdev_get_features(netdev,
1420 &ofport->opp.curr, &ofport->opp.advertised,
1421 &ofport->opp.supported, &ofport->opp.peer);
1426 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1428 if (get_port(p, odp_port->port)) {
1429 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1432 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1433 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1442 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1444 const struct ofp_phy_port *a = &a_->opp;
1445 const struct ofp_phy_port *b = &b_->opp;
1447 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1448 return (a->port_no == b->port_no
1449 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1450 && !strcmp((char *) a->name, (char *) b->name)
1451 && a->state == b->state
1452 && a->config == b->config
1453 && a->curr == b->curr
1454 && a->advertised == b->advertised
1455 && a->supported == b->supported
1456 && a->peer == b->peer);
1460 send_port_status(struct ofproto *p, const struct ofport *ofport,
1463 /* XXX Should limit the number of queued port status change messages. */
1464 struct ofconn *ofconn;
1465 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1466 struct ofp_port_status *ops;
1469 if (!ofconn_receives_async_msgs(ofconn)) {
1473 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1474 ops->reason = reason;
1475 ops->desc = ofport->opp;
1476 hton_ofp_phy_port(&ops->desc);
1477 queue_tx(b, ofconn, NULL);
1482 ofport_install(struct ofproto *p, struct ofport *ofport)
1484 const char *netdev_name = (const char *) ofport->opp.name;
1486 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1487 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1488 shash_add(&p->port_by_name, netdev_name, ofport);
1490 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1495 ofport_remove(struct ofproto *p, struct ofport *ofport)
1497 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1498 hmap_remove(&p->ports, &ofport->hmap_node);
1499 shash_delete(&p->port_by_name,
1500 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1502 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1507 ofport_free(struct ofport *ofport)
1510 netdev_close(ofport->netdev);
1515 static struct ofport *
1516 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1518 struct ofport *port;
1520 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1521 hash_int(odp_port, 0), &ofproto->ports) {
1522 if (port->odp_port == odp_port) {
1530 update_port(struct ofproto *p, const char *devname)
1532 struct odp_port odp_port;
1533 struct ofport *old_ofport;
1534 struct ofport *new_ofport;
1537 COVERAGE_INC(ofproto_update_port);
1539 /* Query the datapath for port information. */
1540 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1542 /* Find the old ofport. */
1543 old_ofport = shash_find_data(&p->port_by_name, devname);
1546 /* There's no port named 'devname' but there might be a port with
1547 * the same port number. This could happen if a port is deleted
1548 * and then a new one added in its place very quickly, or if a port
1549 * is renamed. In the former case we want to send an OFPPR_DELETE
1550 * and an OFPPR_ADD, and in the latter case we want to send a
1551 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1552 * the old port's ifindex against the new port, or perhaps less
1553 * reliably but more portably by comparing the old port's MAC
1554 * against the new port's MAC. However, this code isn't that smart
1555 * and always sends an OFPPR_MODIFY (XXX). */
1556 old_ofport = get_port(p, odp_port.port);
1558 } else if (error != ENOENT && error != ENODEV) {
1559 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1560 "%s", strerror(error));
1564 /* Create a new ofport. */
1565 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1567 /* Eliminate a few pathological cases. */
1568 if (!old_ofport && !new_ofport) {
1570 } else if (old_ofport && new_ofport) {
1571 /* Most of the 'config' bits are OpenFlow soft state, but
1572 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1573 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1574 * leaves the other bits 0.) */
1575 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1577 if (ofport_equal(old_ofport, new_ofport)) {
1578 /* False alarm--no change. */
1579 ofport_free(new_ofport);
1584 /* Now deal with the normal cases. */
1586 ofport_remove(p, old_ofport);
1589 ofport_install(p, new_ofport);
1591 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1592 (!old_ofport ? OFPPR_ADD
1593 : !new_ofport ? OFPPR_DELETE
1595 ofport_free(old_ofport);
1599 init_ports(struct ofproto *p)
1601 struct odp_port *ports;
1606 error = dpif_port_list(p->dpif, &ports, &n_ports);
1611 for (i = 0; i < n_ports; i++) {
1612 const struct odp_port *odp_port = &ports[i];
1613 if (!ofport_conflicts(p, odp_port)) {
1614 struct ofport *ofport = make_ofport(odp_port);
1616 ofport_install(p, ofport);
1624 static struct ofconn *
1625 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1627 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1628 ofconn->ofproto = p;
1629 list_push_back(&p->all_conns, &ofconn->node);
1630 ofconn->rconn = rconn;
1631 ofconn->type = type;
1632 ofconn->flow_format = NXFF_OPENFLOW10;
1633 ofconn->role = NX_ROLE_OTHER;
1634 ofconn->packet_in_counter = rconn_packet_counter_create ();
1635 ofconn->pktbuf = NULL;
1636 ofconn->miss_send_len = 0;
1637 ofconn->reply_counter = rconn_packet_counter_create ();
1642 ofconn_destroy(struct ofconn *ofconn)
1644 if (ofconn->type == OFCONN_PRIMARY) {
1645 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1647 discovery_destroy(ofconn->discovery);
1649 list_remove(&ofconn->node);
1650 switch_status_unregister(ofconn->ss);
1651 rconn_destroy(ofconn->rconn);
1652 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1653 rconn_packet_counter_destroy(ofconn->reply_counter);
1654 pktbuf_destroy(ofconn->pktbuf);
1659 ofconn_run(struct ofconn *ofconn)
1661 struct ofproto *p = ofconn->ofproto;
1665 if (ofconn->discovery) {
1666 char *controller_name;
1667 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1668 discovery_question_connectivity(ofconn->discovery);
1670 if (discovery_run(ofconn->discovery, &controller_name)) {
1671 if (controller_name) {
1672 char *ofconn_name = ofconn_make_name(p, controller_name);
1673 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1676 rconn_disconnect(ofconn->rconn);
1681 for (i = 0; i < N_SCHEDULERS; i++) {
1682 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1685 rconn_run(ofconn->rconn);
1687 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1688 /* Limit the number of iterations to prevent other tasks from
1690 for (iteration = 0; iteration < 50; iteration++) {
1691 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1696 fail_open_maybe_recover(p->fail_open);
1698 handle_openflow(ofconn, of_msg);
1699 ofpbuf_delete(of_msg);
1703 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1704 ofconn_destroy(ofconn);
1709 ofconn_wait(struct ofconn *ofconn)
1713 if (ofconn->discovery) {
1714 discovery_wait(ofconn->discovery);
1716 for (i = 0; i < N_SCHEDULERS; i++) {
1717 pinsched_wait(ofconn->schedulers[i]);
1719 rconn_run_wait(ofconn->rconn);
1720 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1721 rconn_recv_wait(ofconn->rconn);
1723 COVERAGE_INC(ofproto_ofconn_stuck);
1727 /* Returns true if 'ofconn' should receive asynchronous messages. */
1729 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1731 if (ofconn->type == OFCONN_PRIMARY) {
1732 /* Primary controllers always get asynchronous messages unless they
1733 * have configured themselves as "slaves". */
1734 return ofconn->role != NX_ROLE_SLAVE;
1736 /* Service connections don't get asynchronous messages unless they have
1737 * explicitly asked for them by setting a nonzero miss send length. */
1738 return ofconn->miss_send_len > 0;
1742 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1743 * and 'target', suitable for use in log messages for identifying the
1746 * The name is dynamically allocated. The caller should free it (with free())
1747 * when it is no longer needed. */
1749 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1751 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1755 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1759 for (i = 0; i < N_SCHEDULERS; i++) {
1760 struct pinsched **s = &ofconn->schedulers[i];
1764 *s = pinsched_create(rate, burst,
1765 ofconn->ofproto->switch_status);
1767 pinsched_set_limits(*s, rate, burst);
1770 pinsched_destroy(*s);
1777 ofservice_reconfigure(struct ofservice *ofservice,
1778 const struct ofproto_controller *c)
1780 ofservice->probe_interval = c->probe_interval;
1781 ofservice->rate_limit = c->rate_limit;
1782 ofservice->burst_limit = c->burst_limit;
1785 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1786 * positive errno value. */
1788 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1790 struct ofservice *ofservice;
1791 struct pvconn *pvconn;
1794 error = pvconn_open(c->target, &pvconn);
1799 ofservice = xzalloc(sizeof *ofservice);
1800 hmap_insert(&ofproto->services, &ofservice->node,
1801 hash_string(c->target, 0));
1802 ofservice->pvconn = pvconn;
1804 ofservice_reconfigure(ofservice, c);
1810 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1812 hmap_remove(&ofproto->services, &ofservice->node);
1813 pvconn_close(ofservice->pvconn);
1817 /* Finds and returns the ofservice within 'ofproto' that has the given
1818 * 'target', or a null pointer if none exists. */
1819 static struct ofservice *
1820 ofservice_lookup(struct ofproto *ofproto, const char *target)
1822 struct ofservice *ofservice;
1824 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1825 &ofproto->services) {
1826 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1833 /* Caller is responsible for initializing the 'cr' member of the returned
1835 static struct rule *
1836 rule_create(struct ofproto *ofproto, struct rule *super,
1837 const union ofp_action *actions, size_t n_actions,
1838 uint16_t idle_timeout, uint16_t hard_timeout,
1839 ovs_be64 flow_cookie, bool send_flow_removed)
1841 struct rule *rule = xzalloc(sizeof *rule);
1842 rule->idle_timeout = idle_timeout;
1843 rule->hard_timeout = hard_timeout;
1844 rule->flow_cookie = flow_cookie;
1845 rule->used = rule->created = time_msec();
1846 rule->send_flow_removed = send_flow_removed;
1847 rule->super = super;
1849 list_push_back(&super->list, &rule->list);
1851 list_init(&rule->list);
1853 if (n_actions > 0) {
1854 rule->n_actions = n_actions;
1855 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1857 netflow_flow_clear(&rule->nf_flow);
1858 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1863 static struct rule *
1864 rule_from_cls_rule(const struct cls_rule *cls_rule)
1866 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1870 rule_free(struct rule *rule)
1872 free(rule->actions);
1873 free(rule->odp_actions);
1877 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1878 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1879 * through all of its subrules and revalidates them, destroying any that no
1880 * longer has a super-rule (which is probably all of them).
1882 * Before calling this function, the caller must make have removed 'rule' from
1883 * the classifier. If 'rule' is an exact-match rule, the caller is also
1884 * responsible for ensuring that it has been uninstalled from the datapath. */
1886 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1889 struct rule *subrule, *next;
1890 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
1891 revalidate_rule(ofproto, subrule);
1894 list_remove(&rule->list);
1900 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1902 const union ofp_action *oa;
1903 struct actions_iterator i;
1905 if (out_port == htons(OFPP_NONE)) {
1908 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1909 oa = actions_next(&i)) {
1910 if (action_outputs_to_port(oa, out_port)) {
1917 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1918 * 'packet', which arrived on 'in_port'.
1920 * Takes ownership of 'packet'. */
1922 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1923 const union odp_action *actions, size_t n_actions,
1924 struct ofpbuf *packet)
1926 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1927 /* As an optimization, avoid a round-trip from userspace to kernel to
1928 * userspace. This also avoids possibly filling up kernel packet
1929 * buffers along the way. */
1930 struct odp_msg *msg;
1932 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1933 msg->type = _ODPL_ACTION_NR;
1934 msg->length = sizeof(struct odp_msg) + packet->size;
1935 msg->port = in_port;
1937 msg->arg = actions[0].controller.arg;
1939 send_packet_in(ofproto, packet);
1945 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
1946 ofpbuf_delete(packet);
1951 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1952 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1953 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1955 * The flow that 'packet' actually contains does not need to actually match
1956 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1957 * the packet and byte counters for 'rule' will be credited for the packet sent
1958 * out whether or not the packet actually matches 'rule'.
1960 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1961 * the caller must already have accurately composed ODP actions for it given
1962 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1963 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1964 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1965 * actions and apply them to 'packet'.
1967 * Takes ownership of 'packet'. */
1969 rule_execute(struct ofproto *ofproto, struct rule *rule,
1970 struct ofpbuf *packet, const struct flow *flow)
1972 const union odp_action *actions;
1973 struct odp_flow_stats stats;
1975 struct odp_actions a;
1977 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1979 /* Grab or compose the ODP actions.
1981 * The special case for an exact-match 'rule' where 'flow' is not the
1982 * rule's flow is important to avoid, e.g., sending a packet out its input
1983 * port simply because the ODP actions were composed for the wrong
1985 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1986 struct rule *super = rule->super ? rule->super : rule;
1987 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1988 packet, &a, NULL, 0, NULL)) {
1989 ofpbuf_delete(packet);
1992 actions = a.actions;
1993 n_actions = a.n_actions;
1995 actions = rule->odp_actions;
1996 n_actions = rule->n_odp_actions;
1999 /* Execute the ODP actions. */
2000 flow_extract_stats(flow, packet, &stats);
2001 if (execute_odp_actions(ofproto, flow->in_port,
2002 actions, n_actions, packet)) {
2003 update_stats(ofproto, rule, &stats);
2004 rule->used = time_msec();
2005 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2009 /* Inserts 'rule' into 'p''s flow table.
2011 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2012 * actions on it and credits the statistics for sending the packet to 'rule'.
2013 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2016 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2019 struct rule *displaced_rule;
2021 /* Insert the rule in the classifier. */
2022 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2023 if (!rule->cr.wc.wildcards) {
2024 rule_make_actions(p, rule, packet);
2027 /* Send the packet and credit it to the rule. */
2030 flow_extract(packet, 0, in_port, &flow);
2031 rule_execute(p, rule, packet, &flow);
2034 /* Install the rule in the datapath only after sending the packet, to
2035 * avoid packet reordering. */
2036 if (rule->cr.wc.wildcards) {
2037 COVERAGE_INC(ofproto_add_wc_flow);
2038 p->need_revalidate = true;
2040 rule_install(p, rule, displaced_rule);
2043 /* Free the rule that was displaced, if any. */
2044 if (displaced_rule) {
2045 rule_destroy(p, displaced_rule);
2049 static struct rule *
2050 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2051 const struct flow *flow)
2053 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2054 rule->idle_timeout, rule->hard_timeout,
2056 COVERAGE_INC(ofproto_subrule_create);
2057 cls_rule_init_exact(flow, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2058 : rule->cr.priority),
2061 if (classifier_insert(&ofproto->cls, &subrule->cr)) {
2069 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2071 * - If 'rule' was installed in the datapath, uninstalls it and updates
2072 * 'rule''s statistics (or its super-rule's statistics, if it is a
2073 * subrule), via rule_uninstall().
2075 * - Removes 'rule' from the classifier.
2077 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2078 * uninstalls and destroys) its subrules, via rule_destroy().
2081 rule_remove(struct ofproto *ofproto, struct rule *rule)
2083 if (rule->cr.wc.wildcards) {
2084 COVERAGE_INC(ofproto_del_wc_flow);
2085 ofproto->need_revalidate = true;
2087 rule_uninstall(ofproto, rule);
2089 classifier_remove(&ofproto->cls, &rule->cr);
2090 rule_destroy(ofproto, rule);
2093 /* Returns true if the actions changed, false otherwise. */
2095 rule_make_actions(struct ofproto *p, struct rule *rule,
2096 const struct ofpbuf *packet)
2098 const struct rule *super;
2099 struct odp_actions a;
2102 assert(!rule->cr.wc.wildcards);
2104 super = rule->super ? rule->super : rule;
2106 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2107 packet, &a, &rule->tags, &rule->may_install,
2108 &rule->nf_flow.output_iface);
2110 actions_len = a.n_actions * sizeof *a.actions;
2111 if (rule->n_odp_actions != a.n_actions
2112 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2113 COVERAGE_INC(ofproto_odp_unchanged);
2114 free(rule->odp_actions);
2115 rule->n_odp_actions = a.n_actions;
2116 rule->odp_actions = xmemdup(a.actions, actions_len);
2124 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2125 struct odp_flow_put *put)
2127 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2128 odp_flow_key_from_flow(&put->flow.key, &rule->cr.flow);
2129 put->flow.actions = rule->odp_actions;
2130 put->flow.n_actions = rule->n_odp_actions;
2131 put->flow.flags = 0;
2133 return dpif_flow_put(ofproto->dpif, put);
2137 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2139 assert(!rule->cr.wc.wildcards);
2141 if (rule->may_install) {
2142 struct odp_flow_put put;
2143 if (!do_put_flow(p, rule,
2144 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2146 rule->installed = true;
2147 if (displaced_rule) {
2148 update_stats(p, displaced_rule, &put.flow.stats);
2149 rule_post_uninstall(p, displaced_rule);
2152 } else if (displaced_rule) {
2153 rule_uninstall(p, displaced_rule);
2158 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2160 if (rule->installed) {
2161 struct odp_flow_put put;
2162 COVERAGE_INC(ofproto_dp_missed);
2163 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2165 rule_install(ofproto, rule, NULL);
2170 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2172 bool actions_changed;
2173 uint16_t new_out_iface, old_out_iface;
2175 old_out_iface = rule->nf_flow.output_iface;
2176 actions_changed = rule_make_actions(ofproto, rule, NULL);
2178 if (rule->may_install) {
2179 if (rule->installed) {
2180 if (actions_changed) {
2181 struct odp_flow_put put;
2182 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2183 | ODPPF_ZERO_STATS, &put);
2184 update_stats(ofproto, rule, &put.flow.stats);
2186 /* Temporarily set the old output iface so that NetFlow
2187 * messages have the correct output interface for the old
2189 new_out_iface = rule->nf_flow.output_iface;
2190 rule->nf_flow.output_iface = old_out_iface;
2191 rule_post_uninstall(ofproto, rule);
2192 rule->nf_flow.output_iface = new_out_iface;
2195 rule_install(ofproto, rule, NULL);
2198 rule_uninstall(ofproto, rule);
2203 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2205 uint64_t total_bytes = rule->byte_count + extra_bytes;
2207 if (ofproto->ofhooks->account_flow_cb
2208 && total_bytes > rule->accounted_bytes)
2210 ofproto->ofhooks->account_flow_cb(
2211 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2212 total_bytes - rule->accounted_bytes, ofproto->aux);
2213 rule->accounted_bytes = total_bytes;
2217 /* 'rule' must be an exact-match rule in 'p'.
2219 * If 'rule' is installed in the datapath, uninstalls it and updates's
2220 * statistics. If 'rule' is a subrule, the statistics that are updated are
2221 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2224 * If 'rule' is not installed, this function has no effect. */
2226 rule_uninstall(struct ofproto *p, struct rule *rule)
2228 assert(!rule->cr.wc.wildcards);
2229 if (rule->installed) {
2230 struct odp_flow odp_flow;
2232 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
2233 odp_flow.actions = NULL;
2234 odp_flow.n_actions = 0;
2236 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2237 update_stats(p, rule, &odp_flow.stats);
2239 rule->installed = false;
2241 rule_post_uninstall(p, rule);
2246 is_controller_rule(struct rule *rule)
2248 /* If the only action is send to the controller then don't report
2249 * NetFlow expiration messages since it is just part of the control
2250 * logic for the network and not real traffic. */
2254 && rule->super->n_actions == 1
2255 && action_outputs_to_port(&rule->super->actions[0],
2256 htons(OFPP_CONTROLLER)));
2260 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2262 struct rule *super = rule->super;
2264 rule_account(ofproto, rule, 0);
2266 if (ofproto->netflow && !is_controller_rule(rule)) {
2267 struct ofexpired expired;
2268 expired.flow = rule->cr.flow;
2269 expired.packet_count = rule->packet_count;
2270 expired.byte_count = rule->byte_count;
2271 expired.used = rule->used;
2272 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2275 super->packet_count += rule->packet_count;
2276 super->byte_count += rule->byte_count;
2278 /* Reset counters to prevent double counting if the rule ever gets
2280 rule->packet_count = 0;
2281 rule->byte_count = 0;
2282 rule->accounted_bytes = 0;
2284 netflow_flow_clear(&rule->nf_flow);
2289 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2290 struct rconn_packet_counter *counter)
2292 update_openflow_length(msg);
2293 if (rconn_send(ofconn->rconn, msg, counter)) {
2299 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2302 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2304 COVERAGE_INC(ofproto_error);
2305 queue_tx(buf, ofconn, ofconn->reply_counter);
2310 hton_ofp_phy_port(struct ofp_phy_port *opp)
2312 opp->port_no = htons(opp->port_no);
2313 opp->config = htonl(opp->config);
2314 opp->state = htonl(opp->state);
2315 opp->curr = htonl(opp->curr);
2316 opp->advertised = htonl(opp->advertised);
2317 opp->supported = htonl(opp->supported);
2318 opp->peer = htonl(opp->peer);
2322 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2324 struct ofp_header *rq = oh;
2325 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2330 handle_features_request(struct ofconn *ofconn, struct ofp_header *oh)
2332 struct ofp_switch_features *osf;
2334 struct ofport *port;
2336 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2337 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2338 osf->n_buffers = htonl(pktbuf_capacity());
2340 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2341 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2342 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2343 (1u << OFPAT_SET_VLAN_VID) |
2344 (1u << OFPAT_SET_VLAN_PCP) |
2345 (1u << OFPAT_STRIP_VLAN) |
2346 (1u << OFPAT_SET_DL_SRC) |
2347 (1u << OFPAT_SET_DL_DST) |
2348 (1u << OFPAT_SET_NW_SRC) |
2349 (1u << OFPAT_SET_NW_DST) |
2350 (1u << OFPAT_SET_NW_TOS) |
2351 (1u << OFPAT_SET_TP_SRC) |
2352 (1u << OFPAT_SET_TP_DST) |
2353 (1u << OFPAT_ENQUEUE));
2355 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2356 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2359 queue_tx(buf, ofconn, ofconn->reply_counter);
2364 handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh)
2367 struct ofp_switch_config *osc;
2371 /* Figure out flags. */
2372 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2373 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2376 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2377 osc->flags = htons(flags);
2378 osc->miss_send_len = htons(ofconn->miss_send_len);
2379 queue_tx(buf, ofconn, ofconn->reply_counter);
2385 handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc)
2390 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2394 flags = ntohs(osc->flags);
2396 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2397 switch (flags & OFPC_FRAG_MASK) {
2398 case OFPC_FRAG_NORMAL:
2399 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2401 case OFPC_FRAG_DROP:
2402 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2405 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2411 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2417 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2419 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2420 a->controller.arg = max_len;
2423 struct action_xlate_ctx {
2425 struct flow flow; /* Flow to which these actions correspond. */
2426 int recurse; /* Recursion level, via xlate_table_action. */
2427 struct ofproto *ofproto;
2428 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2429 * null pointer if we are revalidating
2430 * without a packet to refer to. */
2433 struct odp_actions *out; /* Datapath actions. */
2434 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2435 bool may_set_up_flow; /* True ordinarily; false if the actions must
2436 * be reassessed for every packet. */
2437 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2440 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2441 * flow translation. */
2442 #define MAX_RESUBMIT_RECURSION 8
2444 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2445 struct action_xlate_ctx *ctx);
2448 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2450 const struct ofport *ofport = get_port(ctx->ofproto, port);
2453 if (ofport->opp.config & OFPPC_NO_FWD) {
2454 /* Forwarding disabled on port. */
2459 * We don't have an ofport record for this port, but it doesn't hurt to
2460 * allow forwarding to it anyhow. Maybe such a port will appear later
2461 * and we're pre-populating the flow table.
2465 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2466 ctx->nf_output_iface = port;
2469 static struct rule *
2470 lookup_valid_rule(struct ofproto *ofproto, const struct flow *flow)
2473 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow,
2476 /* The rule we found might not be valid, since we could be in need of
2477 * revalidation. If it is not valid, don't return it. */
2480 && ofproto->need_revalidate
2481 && !revalidate_rule(ofproto, rule)) {
2482 COVERAGE_INC(ofproto_invalidated);
2490 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2492 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2493 uint16_t old_in_port;
2496 /* Look up a flow with 'in_port' as the input port. Then restore the
2497 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2498 * have surprising behavior). */
2499 old_in_port = ctx->flow.in_port;
2500 ctx->flow.in_port = in_port;
2501 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2502 ctx->flow.in_port = old_in_port;
2510 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2514 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2516 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2517 MAX_RESUBMIT_RECURSION);
2522 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2523 uint16_t *nf_output_iface, struct odp_actions *actions)
2525 struct ofport *ofport;
2527 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2528 uint16_t odp_port = ofport->odp_port;
2529 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2530 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2533 *nf_output_iface = NF_OUT_FLOOD;
2537 xlate_output_action__(struct action_xlate_ctx *ctx,
2538 uint16_t port, uint16_t max_len)
2541 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2543 ctx->nf_output_iface = NF_OUT_DROP;
2547 add_output_action(ctx, ctx->flow.in_port);
2550 xlate_table_action(ctx, ctx->flow.in_port);
2553 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2554 ctx->out, ctx->tags,
2555 &ctx->nf_output_iface,
2556 ctx->ofproto->aux)) {
2557 COVERAGE_INC(ofproto_uninstallable);
2558 ctx->may_set_up_flow = false;
2562 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2563 &ctx->nf_output_iface, ctx->out);
2566 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2567 &ctx->nf_output_iface, ctx->out);
2569 case OFPP_CONTROLLER:
2570 add_controller_action(ctx->out, max_len);
2573 add_output_action(ctx, ODPP_LOCAL);
2576 odp_port = ofp_port_to_odp_port(port);
2577 if (odp_port != ctx->flow.in_port) {
2578 add_output_action(ctx, odp_port);
2583 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2584 ctx->nf_output_iface = NF_OUT_FLOOD;
2585 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2586 ctx->nf_output_iface = prev_nf_output_iface;
2587 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2588 ctx->nf_output_iface != NF_OUT_FLOOD) {
2589 ctx->nf_output_iface = NF_OUT_MULTI;
2594 xlate_output_action(struct action_xlate_ctx *ctx,
2595 const struct ofp_action_output *oao)
2597 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2600 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2601 * optimization, because we're going to add another action that sets the
2602 * priority immediately after, or because there are no actions following the
2605 remove_pop_action(struct action_xlate_ctx *ctx)
2607 size_t n = ctx->out->n_actions;
2608 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2609 ctx->out->n_actions--;
2614 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2615 const struct ofp_action_enqueue *oae)
2617 uint16_t ofp_port, odp_port;
2621 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2624 /* Fall back to ordinary output action. */
2625 xlate_output_action__(ctx, ntohs(oae->port), 0);
2629 /* Figure out ODP output port. */
2630 ofp_port = ntohs(oae->port);
2631 if (ofp_port != OFPP_IN_PORT) {
2632 odp_port = ofp_port_to_odp_port(ofp_port);
2634 odp_port = ctx->flow.in_port;
2637 /* Add ODP actions. */
2638 remove_pop_action(ctx);
2639 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2641 add_output_action(ctx, odp_port);
2642 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2644 /* Update NetFlow output port. */
2645 if (ctx->nf_output_iface == NF_OUT_DROP) {
2646 ctx->nf_output_iface = odp_port;
2647 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2648 ctx->nf_output_iface = NF_OUT_MULTI;
2653 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2654 const struct nx_action_set_queue *nasq)
2659 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2662 /* Couldn't translate queue to a priority, so ignore. A warning
2663 * has already been logged. */
2667 remove_pop_action(ctx);
2668 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2673 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2675 ovs_be16 dl_vlan = ctx->flow.dl_vlan;
2676 uint8_t dl_vlan_pcp = ctx->flow.dl_vlan_pcp;
2678 if (dl_vlan == htons(OFP_VLAN_NONE)) {
2679 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2681 union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2682 oa->dl_tci.tci = htons(ntohs(dl_vlan & htons(VLAN_VID_MASK))
2683 | (dl_vlan_pcp << VLAN_PCP_SHIFT)
2689 xlate_nicira_action(struct action_xlate_ctx *ctx,
2690 const struct nx_action_header *nah)
2692 const struct nx_action_resubmit *nar;
2693 const struct nx_action_set_tunnel *nast;
2694 const struct nx_action_set_queue *nasq;
2695 union odp_action *oa;
2696 int subtype = ntohs(nah->subtype);
2698 assert(nah->vendor == htonl(NX_VENDOR_ID));
2700 case NXAST_RESUBMIT:
2701 nar = (const struct nx_action_resubmit *) nah;
2702 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2705 case NXAST_SET_TUNNEL:
2706 nast = (const struct nx_action_set_tunnel *) nah;
2707 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2708 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2711 case NXAST_DROP_SPOOFED_ARP:
2712 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2713 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2717 case NXAST_SET_QUEUE:
2718 nasq = (const struct nx_action_set_queue *) nah;
2719 xlate_set_queue_action(ctx, nasq);
2722 case NXAST_POP_QUEUE:
2723 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2726 /* If you add a new action here that modifies flow data, don't forget to
2727 * update the flow key in ctx->flow at the same time. */
2730 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2736 do_xlate_actions(const union ofp_action *in, size_t n_in,
2737 struct action_xlate_ctx *ctx)
2739 struct actions_iterator iter;
2740 const union ofp_action *ia;
2741 const struct ofport *port;
2743 port = get_port(ctx->ofproto, ctx->flow.in_port);
2744 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2745 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2746 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2747 /* Drop this flow. */
2751 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2752 uint16_t type = ntohs(ia->type);
2753 union odp_action *oa;
2757 xlate_output_action(ctx, &ia->output);
2760 case OFPAT_SET_VLAN_VID:
2761 ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
2762 xlate_set_dl_tci(ctx);
2765 case OFPAT_SET_VLAN_PCP:
2766 ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
2767 xlate_set_dl_tci(ctx);
2770 case OFPAT_STRIP_VLAN:
2771 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2772 ctx->flow.dl_vlan_pcp = 0;
2773 xlate_set_dl_tci(ctx);
2776 case OFPAT_SET_DL_SRC:
2777 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2778 memcpy(oa->dl_addr.dl_addr,
2779 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2780 memcpy(ctx->flow.dl_src,
2781 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2784 case OFPAT_SET_DL_DST:
2785 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2786 memcpy(oa->dl_addr.dl_addr,
2787 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2788 memcpy(ctx->flow.dl_dst,
2789 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2792 case OFPAT_SET_NW_SRC:
2793 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2794 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2797 case OFPAT_SET_NW_DST:
2798 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2799 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2802 case OFPAT_SET_NW_TOS:
2803 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2804 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2807 case OFPAT_SET_TP_SRC:
2808 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2809 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2812 case OFPAT_SET_TP_DST:
2813 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2814 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2818 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2822 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2826 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2833 xlate_actions(const union ofp_action *in, size_t n_in,
2834 const struct flow *flow, struct ofproto *ofproto,
2835 const struct ofpbuf *packet,
2836 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2837 uint16_t *nf_output_iface)
2839 tag_type no_tags = 0;
2840 struct action_xlate_ctx ctx;
2841 COVERAGE_INC(ofproto_ofp2odp);
2842 odp_actions_init(out);
2845 ctx.ofproto = ofproto;
2846 ctx.packet = packet;
2848 ctx.tags = tags ? tags : &no_tags;
2849 ctx.may_set_up_flow = true;
2850 ctx.nf_output_iface = NF_OUT_DROP;
2851 do_xlate_actions(in, n_in, &ctx);
2852 remove_pop_action(&ctx);
2854 /* Check with in-band control to see if we're allowed to set up this
2856 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2857 ctx.may_set_up_flow = false;
2860 if (may_set_up_flow) {
2861 *may_set_up_flow = ctx.may_set_up_flow;
2863 if (nf_output_iface) {
2864 *nf_output_iface = ctx.nf_output_iface;
2866 if (odp_actions_overflow(out)) {
2867 COVERAGE_INC(odp_overflow);
2868 odp_actions_init(out);
2869 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2874 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2875 * error message code (composed with ofp_mkerr()) for the caller to propagate
2876 * upward. Otherwise, returns 0.
2878 * The log message mentions 'msg_type'. */
2880 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2882 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2883 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2884 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2887 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2894 handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh)
2896 struct ofproto *p = ofconn->ofproto;
2897 struct ofp_packet_out *opo;
2898 struct ofpbuf payload, *buffer;
2899 union ofp_action *ofp_actions;
2900 struct odp_actions odp_actions;
2901 struct ofpbuf request;
2903 size_t n_ofp_actions;
2907 COVERAGE_INC(ofproto_packet_out);
2909 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2914 /* Get ofp_packet_out. */
2916 request.size = ntohs(oh->length);
2917 opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions));
2919 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2923 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2924 &ofp_actions, &n_ofp_actions);
2930 if (opo->buffer_id != htonl(UINT32_MAX)) {
2931 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2933 if (error || !buffer) {
2942 /* Extract flow, check actions. */
2943 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2945 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2951 error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload,
2952 &odp_actions, NULL, NULL, NULL);
2954 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions,
2959 ofpbuf_delete(buffer);
2964 update_port_config(struct ofproto *p, struct ofport *port,
2965 uint32_t config, uint32_t mask)
2967 mask &= config ^ port->opp.config;
2968 if (mask & OFPPC_PORT_DOWN) {
2969 if (config & OFPPC_PORT_DOWN) {
2970 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2972 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2975 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2976 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2977 if (mask & REVALIDATE_BITS) {
2978 COVERAGE_INC(ofproto_costly_flags);
2979 port->opp.config ^= mask & REVALIDATE_BITS;
2980 p->need_revalidate = true;
2982 #undef REVALIDATE_BITS
2983 if (mask & OFPPC_NO_PACKET_IN) {
2984 port->opp.config ^= OFPPC_NO_PACKET_IN;
2989 handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh)
2991 struct ofproto *p = ofconn->ofproto;
2992 const struct ofp_port_mod *opm;
2993 struct ofport *port;
2996 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3000 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
3004 opm = (struct ofp_port_mod *) oh;
3006 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3008 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3009 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3010 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3012 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3013 if (opm->advertise) {
3014 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3020 static struct ofpbuf *
3021 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3023 struct ofp_stats_reply *osr;
3026 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3027 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3029 osr->flags = htons(0);
3033 static struct ofpbuf *
3034 start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3036 return make_ofp_stats_reply(request->header.xid, request->type, body_len);
3040 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3041 struct ofpbuf **msgp)
3043 struct ofpbuf *msg = *msgp;
3044 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3045 if (nbytes + msg->size > UINT16_MAX) {
3046 struct ofp_stats_reply *reply = msg->data;
3047 reply->flags = htons(OFPSF_REPLY_MORE);
3048 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3049 queue_tx(msg, ofconn, ofconn->reply_counter);
3051 return ofpbuf_put_uninit(*msgp, nbytes);
3054 static struct ofpbuf *
3055 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3057 struct nicira_stats_msg *nsm;
3060 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3061 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3062 nsm->type = htons(OFPST_VENDOR);
3063 nsm->flags = htons(0);
3064 nsm->vendor = htonl(NX_VENDOR_ID);
3065 nsm->subtype = htonl(subtype);
3069 static struct ofpbuf *
3070 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3072 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3076 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3077 struct ofpbuf **msgp)
3079 struct ofpbuf *msg = *msgp;
3080 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3081 if (nbytes + msg->size > UINT16_MAX) {
3082 struct nicira_stats_msg *reply = msg->data;
3083 reply->flags = htons(OFPSF_REPLY_MORE);
3084 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3085 queue_tx(msg, ofconn, ofconn->reply_counter);
3087 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3091 handle_desc_stats_request(struct ofconn *ofconn,
3092 struct ofp_stats_request *request)
3094 struct ofproto *p = ofconn->ofproto;
3095 struct ofp_desc_stats *ods;
3098 msg = start_ofp_stats_reply(request, sizeof *ods);
3099 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3100 memset(ods, 0, sizeof *ods);
3101 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3102 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3103 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3104 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3105 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3106 queue_tx(msg, ofconn, ofconn->reply_counter);
3112 handle_table_stats_request(struct ofconn *ofconn,
3113 struct ofp_stats_request *request)
3115 struct ofproto *p = ofconn->ofproto;
3116 struct ofp_table_stats *ots;
3121 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3123 /* Count rules other than subrules. */
3124 n_rules = classifier_count(&p->cls);
3125 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) {
3131 /* Classifier table. */
3132 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3133 memset(ots, 0, sizeof *ots);
3134 strcpy(ots->name, "classifier");
3135 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3136 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3137 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3138 ots->active_count = htonl(n_rules);
3139 ots->lookup_count = htonll(0); /* XXX */
3140 ots->matched_count = htonll(0); /* XXX */
3142 queue_tx(msg, ofconn, ofconn->reply_counter);
3147 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3148 struct ofpbuf **msgp)
3150 struct netdev_stats stats;
3151 struct ofp_port_stats *ops;
3153 /* Intentionally ignore return value, since errors will set
3154 * 'stats' to all-1s, which is correct for OpenFlow, and
3155 * netdev_get_stats() will log errors. */
3156 netdev_get_stats(port->netdev, &stats);
3158 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3159 ops->port_no = htons(port->opp.port_no);
3160 memset(ops->pad, 0, sizeof ops->pad);
3161 ops->rx_packets = htonll(stats.rx_packets);
3162 ops->tx_packets = htonll(stats.tx_packets);
3163 ops->rx_bytes = htonll(stats.rx_bytes);
3164 ops->tx_bytes = htonll(stats.tx_bytes);
3165 ops->rx_dropped = htonll(stats.rx_dropped);
3166 ops->tx_dropped = htonll(stats.tx_dropped);
3167 ops->rx_errors = htonll(stats.rx_errors);
3168 ops->tx_errors = htonll(stats.tx_errors);
3169 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3170 ops->rx_over_err = htonll(stats.rx_over_errors);
3171 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3172 ops->collisions = htonll(stats.collisions);
3176 handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr,
3179 struct ofproto *p = ofconn->ofproto;
3180 struct ofp_port_stats_request *psr;
3181 struct ofp_port_stats *ops;
3183 struct ofport *port;
3185 if (arg_size != sizeof *psr) {
3186 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3188 psr = (struct ofp_port_stats_request *) osr->body;
3190 msg = start_ofp_stats_reply(osr, sizeof *ops * 16);
3191 if (psr->port_no != htons(OFPP_NONE)) {
3192 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3194 append_port_stat(port, ofconn, &msg);
3197 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3198 append_port_stat(port, ofconn, &msg);
3202 queue_tx(msg, ofconn, ofconn->reply_counter);
3206 struct flow_stats_cbdata {
3207 struct ofconn *ofconn;
3212 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3213 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3214 * returned statistic include statistics for all of 'rule''s subrules. */
3216 query_stats(struct ofproto *p, struct rule *rule,
3217 uint64_t *packet_countp, uint64_t *byte_countp)
3219 uint64_t packet_count, byte_count;
3220 struct rule *subrule;
3221 struct odp_flow *odp_flows;
3224 /* Start from historical data for 'rule' itself that are no longer tracked
3225 * by the datapath. This counts, for example, subrules that have
3227 packet_count = rule->packet_count;
3228 byte_count = rule->byte_count;
3230 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3231 * wildcarded then on all of its subrules.
3233 * Also, add any statistics that are not tracked by the datapath for each
3234 * subrule. This includes, for example, statistics for packets that were
3235 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3237 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3238 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3239 if (rule->cr.wc.wildcards) {
3241 LIST_FOR_EACH (subrule, list, &rule->list) {
3242 odp_flow_key_from_flow(&odp_flows[i++].key, &subrule->cr.flow);
3243 packet_count += subrule->packet_count;
3244 byte_count += subrule->byte_count;
3247 odp_flow_key_from_flow(&odp_flows[0].key, &rule->cr.flow);
3250 /* Fetch up-to-date statistics from the datapath and add them in. */
3251 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3253 for (i = 0; i < n_odp_flows; i++) {
3254 struct odp_flow *odp_flow = &odp_flows[i];
3255 packet_count += odp_flow->stats.n_packets;
3256 byte_count += odp_flow->stats.n_bytes;
3261 /* Return the stats to the caller. */
3262 *packet_countp = packet_count;
3263 *byte_countp = byte_count;
3267 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3269 long long int msecs = time_msec() - start;
3270 *sec = htonl(msecs / 1000);
3271 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3275 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3277 struct rule *rule = rule_from_cls_rule(rule_);
3278 struct flow_stats_cbdata *cbdata = cbdata_;
3279 struct ofp_flow_stats *ofs;
3280 uint64_t packet_count, byte_count;
3281 size_t act_len, len;
3283 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3287 act_len = sizeof *rule->actions * rule->n_actions;
3288 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3290 query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count);
3292 ofs = append_ofp_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3293 ofs->length = htons(len);
3296 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3297 cbdata->ofconn->flow_format, &ofs->match);
3298 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3299 ofs->cookie = rule->flow_cookie;
3300 ofs->priority = htons(rule->cr.priority);
3301 ofs->idle_timeout = htons(rule->idle_timeout);
3302 ofs->hard_timeout = htons(rule->hard_timeout);
3303 memset(ofs->pad2, 0, sizeof ofs->pad2);
3304 ofs->packet_count = htonll(packet_count);
3305 ofs->byte_count = htonll(byte_count);
3306 if (rule->n_actions > 0) {
3307 memcpy(ofs->actions, rule->actions, act_len);
3312 table_id_to_include(uint8_t table_id)
3314 return table_id == 0 || table_id == 0xff ? CLS_INC_ALL : 0;
3318 handle_flow_stats_request(struct ofconn *ofconn,
3319 const struct ofp_stats_request *osr, size_t arg_size)
3321 struct ofp_flow_stats_request *fsr;
3322 struct flow_stats_cbdata cbdata;
3323 struct cls_rule target;
3325 if (arg_size != sizeof *fsr) {
3326 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3328 fsr = (struct ofp_flow_stats_request *) osr->body;
3330 COVERAGE_INC(ofproto_flows_req);
3331 cbdata.ofconn = ofconn;
3332 cbdata.out_port = fsr->out_port;
3333 cbdata.msg = start_ofp_stats_reply(osr, 1024);
3334 cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, &target);
3335 classifier_for_each_match(&ofconn->ofproto->cls, &target,
3336 table_id_to_include(fsr->table_id),
3337 flow_stats_cb, &cbdata);
3338 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3343 nx_flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3345 struct rule *rule = rule_from_cls_rule(rule_);
3346 struct flow_stats_cbdata *cbdata = cbdata_;
3347 struct nx_flow_stats *nfs;
3348 uint64_t packet_count, byte_count;
3349 size_t act_len, start_len;
3351 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3355 query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count);
3357 act_len = sizeof *rule->actions * rule->n_actions;
3359 start_len = cbdata->msg->size;
3360 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len,
3361 cbdata->ofconn, &cbdata->msg);
3362 nfs = ofpbuf_put_uninit(cbdata->msg, sizeof *nfs);
3365 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3366 nfs->cookie = rule->flow_cookie;
3367 nfs->priority = htons(rule->cr.priority);
3368 nfs->idle_timeout = htons(rule->idle_timeout);
3369 nfs->hard_timeout = htons(rule->hard_timeout);
3370 nfs->match_len = htons(nx_put_match(cbdata->msg, &rule->cr));
3371 memset(nfs->pad2, 0, sizeof nfs->pad2);
3372 nfs->packet_count = htonll(packet_count);
3373 nfs->byte_count = htonll(byte_count);
3374 if (rule->n_actions > 0) {
3375 ofpbuf_put(cbdata->msg, rule->actions, act_len);
3377 nfs->length = htons(cbdata->msg->size - start_len);
3381 handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b)
3383 struct nx_flow_stats_request *nfsr;
3384 struct flow_stats_cbdata cbdata;
3385 struct cls_rule target;
3388 /* Dissect the message. */
3389 nfsr = ofpbuf_try_pull(b, sizeof *nfsr);
3391 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3393 error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target);
3398 COVERAGE_INC(ofproto_flows_req);
3399 cbdata.ofconn = ofconn;
3400 cbdata.out_port = nfsr->out_port;
3401 cbdata.msg = start_nxstats_reply(&nfsr->nsm, 1024);
3402 classifier_for_each_match(&ofconn->ofproto->cls, &target,
3403 table_id_to_include(nfsr->table_id),
3404 nx_flow_stats_cb, &cbdata);
3405 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3409 struct flow_stats_ds_cbdata {
3410 struct ofproto *ofproto;
3415 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3417 struct rule *rule = rule_from_cls_rule(rule_);
3418 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3419 struct ds *results = cbdata->results;
3420 struct ofp_match match;
3421 uint64_t packet_count, byte_count;
3422 size_t act_len = sizeof *rule->actions * rule->n_actions;
3424 /* Don't report on subrules. */
3425 if (rule->super != NULL) {
3429 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3430 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3431 NXFF_OPENFLOW10, &match);
3433 ds_put_format(results, "duration=%llds, ",
3434 (time_msec() - rule->created) / 1000);
3435 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3436 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3437 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3438 ofp_print_match(results, &match, true);
3440 ofp_print_actions(results, &rule->actions->header, act_len);
3442 ds_put_cstr(results, "drop");
3444 ds_put_cstr(results, "\n");
3447 /* Adds a pretty-printed description of all flows to 'results', including
3448 * those marked hidden by secchan (e.g., by in-band control). */
3450 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3452 struct ofp_match match;
3453 struct cls_rule target;
3454 struct flow_stats_ds_cbdata cbdata;
3456 memset(&match, 0, sizeof match);
3457 match.wildcards = htonl(OVSFW_ALL);
3460 cbdata.results = results;
3462 cls_rule_from_match(&match, 0, NXFF_OPENFLOW10, 0, &target);
3463 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3464 flow_stats_ds_cb, &cbdata);
3467 struct aggregate_stats_cbdata {
3468 struct ofproto *ofproto;
3470 uint64_t packet_count;
3471 uint64_t byte_count;
3476 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3478 struct rule *rule = rule_from_cls_rule(rule_);
3479 struct aggregate_stats_cbdata *cbdata = cbdata_;
3480 uint64_t packet_count, byte_count;
3482 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3486 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3488 cbdata->packet_count += packet_count;
3489 cbdata->byte_count += byte_count;
3494 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3495 ovs_be16 out_port, uint8_t table_id,
3496 struct ofp_aggregate_stats_reply *oasr)
3498 struct aggregate_stats_cbdata cbdata;
3500 COVERAGE_INC(ofproto_agg_request);
3501 cbdata.ofproto = ofproto;
3502 cbdata.out_port = out_port;
3503 cbdata.packet_count = 0;
3504 cbdata.byte_count = 0;
3506 classifier_for_each_match(&ofproto->cls, target,
3507 table_id_to_include(table_id),
3508 aggregate_stats_cb, &cbdata);
3510 oasr->flow_count = htonl(cbdata.n_flows);
3511 oasr->packet_count = htonll(cbdata.packet_count);
3512 oasr->byte_count = htonll(cbdata.byte_count);
3513 memset(oasr->pad, 0, sizeof oasr->pad);
3517 handle_aggregate_stats_request(struct ofconn *ofconn,
3518 const struct ofp_stats_request *osr,
3521 struct ofp_aggregate_stats_request *request;
3522 struct ofp_aggregate_stats_reply *reply;
3523 struct cls_rule target;
3526 if (arg_size != sizeof *request) {
3527 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3529 request = (struct ofp_aggregate_stats_request *) osr->body;
3531 cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, &target);
3533 msg = start_ofp_stats_reply(osr, sizeof *reply);
3534 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3535 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3536 request->table_id, reply);
3537 queue_tx(msg, ofconn, ofconn->reply_counter);
3542 handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b)
3544 struct nx_aggregate_stats_request *request;
3545 struct ofp_aggregate_stats_reply *reply;
3546 struct cls_rule target;
3550 /* Dissect the message. */
3551 request = ofpbuf_try_pull(b, sizeof *request);
3553 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3555 error = nx_pull_match(b, ntohs(request->match_len), 0, &target);
3561 COVERAGE_INC(ofproto_flows_req);
3562 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3563 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3564 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3565 request->table_id, reply);
3566 queue_tx(buf, ofconn, ofconn->reply_counter);
3571 struct queue_stats_cbdata {
3572 struct ofconn *ofconn;
3573 struct ofport *ofport;
3578 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3579 const struct netdev_queue_stats *stats)
3581 struct ofp_queue_stats *reply;
3583 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3584 reply->port_no = htons(cbdata->ofport->opp.port_no);
3585 memset(reply->pad, 0, sizeof reply->pad);
3586 reply->queue_id = htonl(queue_id);
3587 reply->tx_bytes = htonll(stats->tx_bytes);
3588 reply->tx_packets = htonll(stats->tx_packets);
3589 reply->tx_errors = htonll(stats->tx_errors);
3593 handle_queue_stats_dump_cb(uint32_t queue_id,
3594 struct netdev_queue_stats *stats,
3597 struct queue_stats_cbdata *cbdata = cbdata_;
3599 put_queue_stats(cbdata, queue_id, stats);
3603 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3604 struct queue_stats_cbdata *cbdata)
3606 cbdata->ofport = port;
3607 if (queue_id == OFPQ_ALL) {
3608 netdev_dump_queue_stats(port->netdev,
3609 handle_queue_stats_dump_cb, cbdata);
3611 struct netdev_queue_stats stats;
3613 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3614 put_queue_stats(cbdata, queue_id, &stats);
3620 handle_queue_stats_request(struct ofconn *ofconn,
3621 const struct ofp_stats_request *osr,
3624 struct ofproto *ofproto = ofconn->ofproto;
3625 struct ofp_queue_stats_request *qsr;
3626 struct queue_stats_cbdata cbdata;
3627 struct ofport *port;
3628 unsigned int port_no;
3631 if (arg_size != sizeof *qsr) {
3632 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3634 qsr = (struct ofp_queue_stats_request *) osr->body;
3636 COVERAGE_INC(ofproto_queue_req);
3638 cbdata.ofconn = ofconn;
3639 cbdata.msg = start_ofp_stats_reply(osr, 128);
3641 port_no = ntohs(qsr->port_no);
3642 queue_id = ntohl(qsr->queue_id);
3643 if (port_no == OFPP_ALL) {
3644 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3645 handle_queue_stats_for_port(port, queue_id, &cbdata);
3647 } else if (port_no < ofproto->max_ports) {
3648 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3650 handle_queue_stats_for_port(port, queue_id, &cbdata);
3653 ofpbuf_delete(cbdata.msg);
3654 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3656 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3662 handle_vendor_stats_request(struct ofconn *ofconn,
3663 struct ofp_stats_request *osr, size_t arg_size)
3665 struct nicira_stats_msg *nsm;
3670 VLOG_WARN_RL(&rl, "truncated vendor stats request body");
3671 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3674 memcpy(&vendor, osr->body, sizeof vendor);
3675 if (vendor != htonl(NX_VENDOR_ID)) {
3676 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3679 if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) {
3680 VLOG_WARN_RL(&rl, "truncated Nicira stats request");
3681 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3684 nsm = (struct nicira_stats_msg *) osr;
3686 b.size = ntohs(nsm->header.length);
3687 switch (ntohl(nsm->subtype)) {
3689 return handle_nxst_flow(ofconn, &b);
3691 case NXST_AGGREGATE:
3692 return handle_nxst_aggregate(ofconn, &b);
3695 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3700 handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh)
3702 struct ofp_stats_request *osr;
3706 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3711 osr = (struct ofp_stats_request *) oh;
3713 switch (ntohs(osr->type)) {
3715 return handle_desc_stats_request(ofconn, osr);
3718 return handle_flow_stats_request(ofconn, osr, arg_size);
3720 case OFPST_AGGREGATE:
3721 return handle_aggregate_stats_request(ofconn, osr, arg_size);
3724 return handle_table_stats_request(ofconn, osr);
3727 return handle_port_stats_request(ofconn, osr, arg_size);
3730 return handle_queue_stats_request(ofconn, osr, arg_size);
3733 return handle_vendor_stats_request(ofconn, osr, arg_size);
3736 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3740 static long long int
3741 msec_from_nsec(uint64_t sec, uint32_t nsec)
3743 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3747 update_time(struct ofproto *ofproto, struct rule *rule,
3748 const struct odp_flow_stats *stats)
3750 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3751 if (used > rule->used) {
3753 if (rule->super && used > rule->super->used) {
3754 rule->super->used = used;
3756 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3761 update_stats(struct ofproto *ofproto, struct rule *rule,
3762 const struct odp_flow_stats *stats)
3764 if (stats->n_packets) {
3765 update_time(ofproto, rule, stats);
3766 rule->packet_count += stats->n_packets;
3767 rule->byte_count += stats->n_bytes;
3768 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3776 uint16_t idle_timeout;
3777 uint16_t hard_timeout;
3781 union ofp_action *actions;
3785 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3786 * in which no matching flow already exists in the flow table.
3788 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3789 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3790 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3792 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3795 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3797 struct ofproto *p = ofconn->ofproto;
3798 struct ofpbuf *packet;
3803 if (fm->flags & OFPFF_CHECK_OVERLAP
3804 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3805 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3808 rule = rule_create(p, NULL, fm->actions, fm->n_actions,
3809 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3810 fm->flags & OFPFF_SEND_FLOW_REM);
3814 if (fm->buffer_id != UINT32_MAX) {
3815 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3819 in_port = UINT16_MAX;
3822 rule_insert(p, rule, packet, in_port);
3826 static struct rule *
3827 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3829 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3833 send_buffered_packet(struct ofconn *ofconn,
3834 struct rule *rule, uint32_t buffer_id)
3836 struct ofpbuf *packet;
3841 if (buffer_id == UINT32_MAX) {
3845 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3850 flow_extract(packet, 0, in_port, &flow);
3851 rule_execute(ofconn->ofproto, rule, packet, &flow);
3856 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3858 struct modify_flows_cbdata {
3859 struct ofproto *ofproto;
3860 const struct flow_mod *fm;
3864 static int modify_flow(struct ofproto *, const struct flow_mod *,
3866 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3868 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3869 * encoded by ofp_mkerr() on failure.
3871 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3874 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3876 struct modify_flows_cbdata cbdata;
3878 cbdata.ofproto = ofconn->ofproto;
3880 cbdata.match = NULL;
3882 classifier_for_each_match(&ofconn->ofproto->cls, &fm->cr, CLS_INC_ALL,
3883 modify_flows_cb, &cbdata);
3885 /* This credits the packet to whichever flow happened to happened to
3886 * match last. That's weird. Maybe we should do a lookup for the
3887 * flow that actually matches the packet? Who knows. */
3888 send_buffered_packet(ofconn, cbdata.match, fm->buffer_id);
3891 return add_flow(ofconn, fm);
3895 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3896 * code as encoded by ofp_mkerr() on failure.
3898 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3901 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3903 struct ofproto *p = ofconn->ofproto;
3904 struct rule *rule = find_flow_strict(p, fm);
3905 if (rule && !rule_is_hidden(rule)) {
3906 modify_flow(p, fm, rule);
3907 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3909 return add_flow(ofconn, fm);
3913 /* Callback for modify_flows_loose(). */
3915 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3917 struct rule *rule = rule_from_cls_rule(rule_);
3918 struct modify_flows_cbdata *cbdata = cbdata_;
3920 if (!rule_is_hidden(rule)) {
3921 cbdata->match = rule;
3922 modify_flow(cbdata->ofproto, cbdata->fm, rule);
3926 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3927 * been identified as a flow in 'p''s flow table to be modified, by changing
3928 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3929 * ofp_action[] structures). */
3931 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3933 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3935 rule->flow_cookie = fm->cookie;
3937 /* If the actions are the same, do nothing. */
3938 if (fm->n_actions == rule->n_actions
3940 || !memcmp(fm->actions, rule->actions, actions_len))) {
3944 /* Replace actions. */
3945 free(rule->actions);
3946 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3947 rule->n_actions = fm->n_actions;
3949 /* Make sure that the datapath gets updated properly. */
3950 if (rule->cr.wc.wildcards) {
3951 COVERAGE_INC(ofproto_mod_wc_flow);
3952 p->need_revalidate = true;
3954 rule_update_actions(p, rule);
3960 /* OFPFC_DELETE implementation. */
3962 struct delete_flows_cbdata {
3963 struct ofproto *ofproto;
3967 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3968 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3970 /* Implements OFPFC_DELETE. */
3972 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3974 struct delete_flows_cbdata cbdata;
3977 cbdata.out_port = htons(fm->out_port);
3979 classifier_for_each_match(&p->cls, &fm->cr, CLS_INC_ALL,
3980 delete_flows_cb, &cbdata);
3983 /* Implements OFPFC_DELETE_STRICT. */
3985 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3987 struct rule *rule = find_flow_strict(p, fm);
3989 delete_flow(p, rule, htons(fm->out_port));
3993 /* Callback for delete_flows_loose(). */
3995 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3997 struct rule *rule = rule_from_cls_rule(rule_);
3998 struct delete_flows_cbdata *cbdata = cbdata_;
4000 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
4003 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4004 * been identified as a flow to delete from 'p''s flow table, by deleting the
4005 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4008 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4009 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4010 * specified 'out_port'. */
4012 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4014 if (rule_is_hidden(rule)) {
4018 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4022 send_flow_removed(p, rule, OFPRR_DELETE);
4023 rule_remove(p, rule);
4027 flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm)
4029 struct ofproto *p = ofconn->ofproto;
4032 error = reject_slave_controller(ofconn, "flow_mod");
4037 error = validate_actions(fm->actions, fm->n_actions,
4038 &fm->cr.flow, p->max_ports);
4043 /* We do not support the emergency flow cache. It will hopefully
4044 * get dropped from OpenFlow in the near future. */
4045 if (fm->flags & OFPFF_EMERG) {
4046 /* There isn't a good fit for an error code, so just state that the
4047 * flow table is full. */
4048 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4051 switch (fm->command) {
4053 return add_flow(ofconn, fm);
4056 return modify_flows_loose(ofconn, fm);
4058 case OFPFC_MODIFY_STRICT:
4059 return modify_flow_strict(ofconn, fm);
4062 delete_flows_loose(p, fm);
4065 case OFPFC_DELETE_STRICT:
4066 delete_flow_strict(p, fm);
4070 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4075 handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4077 struct ofp_match orig_match;
4078 struct ofp_flow_mod *ofm;
4084 b.size = ntohs(oh->length);
4086 /* Dissect the message. */
4087 ofm = ofpbuf_try_pull(&b, sizeof *ofm);
4089 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4091 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4096 /* Normalize ofm->match. If normalization actually changes anything, then
4097 * log the differences. */
4098 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
4099 orig_match = ofm->match;
4100 normalize_match(&ofm->match);
4101 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
4102 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4103 if (!VLOG_DROP_INFO(&normal_rl)) {
4104 char *old = ofp_match_to_literal_string(&orig_match);
4105 char *new = ofp_match_to_literal_string(&ofm->match);
4106 VLOG_INFO("%s: normalization changed ofp_match, details:",
4107 rconn_get_name(ofconn->rconn));
4108 VLOG_INFO(" pre: %s", old);
4109 VLOG_INFO("post: %s", new);
4115 /* Translate the message. */
4116 cls_rule_from_match(&ofm->match, ntohs(ofm->priority), ofconn->flow_format,
4117 ofm->cookie, &fm.cr);
4118 fm.cookie = ofm->cookie;
4119 fm.command = ntohs(ofm->command);
4120 fm.idle_timeout = ntohs(ofm->idle_timeout);
4121 fm.hard_timeout = ntohs(ofm->hard_timeout);
4122 fm.buffer_id = ntohl(ofm->buffer_id);
4123 fm.out_port = ntohs(ofm->out_port);
4124 fm.flags = ntohs(ofm->flags);
4126 /* Execute the command. */
4127 return flow_mod_core(ofconn, &fm);
4131 handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4133 struct nx_flow_mod *nfm;
4139 b.size = ntohs(oh->length);
4141 /* Dissect the message. */
4142 nfm = ofpbuf_try_pull(&b, sizeof *nfm);
4144 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4146 error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority),
4151 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4156 /* Translate the message. */
4157 fm.cookie = nfm->cookie;
4158 fm.command = ntohs(nfm->command);
4159 fm.idle_timeout = ntohs(nfm->idle_timeout);
4160 fm.hard_timeout = ntohs(nfm->hard_timeout);
4161 fm.buffer_id = ntohl(nfm->buffer_id);
4162 fm.out_port = ntohs(nfm->out_port);
4163 fm.flags = ntohs(nfm->flags);
4165 /* Execute the command. */
4166 return flow_mod_core(ofconn, &fm);
4170 handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg)
4174 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4179 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4184 handle_role_request(struct ofconn *ofconn, struct nicira_header *msg)
4186 struct nx_role_request *nrr;
4187 struct nx_role_request *reply;
4191 if (ntohs(msg->header.length) != sizeof *nrr) {
4192 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
4193 ntohs(msg->header.length), sizeof *nrr);
4194 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4196 nrr = (struct nx_role_request *) msg;
4198 if (ofconn->type != OFCONN_PRIMARY) {
4199 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4201 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4204 role = ntohl(nrr->role);
4205 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4206 && role != NX_ROLE_SLAVE) {
4207 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4209 /* There's no good error code for this. */
4210 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4213 if (role == NX_ROLE_MASTER) {
4214 struct ofconn *other;
4216 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4217 if (other->role == NX_ROLE_MASTER) {
4218 other->role = NX_ROLE_SLAVE;
4222 ofconn->role = role;
4224 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid,
4226 reply->role = htonl(role);
4227 queue_tx(buf, ofconn, ofconn->reply_counter);
4233 handle_nxt_set_flow_format(struct ofconn *ofconn,
4234 struct nxt_set_flow_format *msg)
4239 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4244 format = ntohl(msg->format);
4245 if (format == NXFF_OPENFLOW10
4246 || format == NXFF_TUN_ID_FROM_COOKIE
4247 || format == NXFF_NXM) {
4248 ofconn->flow_format = format;
4251 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4256 handle_vendor(struct ofconn *ofconn, void *msg)
4258 struct ofproto *p = ofconn->ofproto;
4259 struct ofp_vendor_header *ovh = msg;
4260 struct nicira_header *nh;
4262 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4263 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4264 "(expected at least %zu)",
4265 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4266 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4268 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4269 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4271 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4272 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4273 "(expected at least %zu)",
4274 ntohs(ovh->header.length), sizeof(struct nicira_header));
4275 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4279 switch (ntohl(nh->subtype)) {
4280 case NXT_STATUS_REQUEST:
4281 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4284 case NXT_TUN_ID_FROM_COOKIE:
4285 return handle_tun_id_from_cookie(ofconn, msg);
4287 case NXT_ROLE_REQUEST:
4288 return handle_role_request(ofconn, msg);
4290 case NXT_SET_FLOW_FORMAT:
4291 return handle_nxt_set_flow_format(ofconn, msg);
4294 return handle_nxt_flow_mod(ofconn, &ovh->header);
4297 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4301 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4303 struct ofp_header *ob;
4306 /* Currently, everything executes synchronously, so we can just
4307 * immediately send the barrier reply. */
4308 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4309 queue_tx(buf, ofconn, ofconn->reply_counter);
4314 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4316 struct ofp_header *oh = ofp_msg->data;
4319 COVERAGE_INC(ofproto_recv_openflow);
4321 case OFPT_ECHO_REQUEST:
4322 error = handle_echo_request(ofconn, oh);
4325 case OFPT_ECHO_REPLY:
4329 case OFPT_FEATURES_REQUEST:
4330 error = handle_features_request(ofconn, oh);
4333 case OFPT_GET_CONFIG_REQUEST:
4334 error = handle_get_config_request(ofconn, oh);
4337 case OFPT_SET_CONFIG:
4338 error = handle_set_config(ofconn, ofp_msg->data);
4341 case OFPT_PACKET_OUT:
4342 error = handle_packet_out(ofconn, ofp_msg->data);
4346 error = handle_port_mod(ofconn, oh);
4350 error = handle_ofpt_flow_mod(ofconn, ofp_msg->data);
4353 case OFPT_STATS_REQUEST:
4354 error = handle_stats_request(ofconn, oh);
4358 error = handle_vendor(ofconn, ofp_msg->data);
4361 case OFPT_BARRIER_REQUEST:
4362 error = handle_barrier_request(ofconn, oh);
4366 if (VLOG_IS_WARN_ENABLED()) {
4367 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4368 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4371 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4376 send_error_oh(ofconn, ofp_msg->data, error);
4381 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4383 struct odp_msg *msg = packet->data;
4385 struct ofpbuf payload;
4388 payload.data = msg + 1;
4389 payload.size = msg->length - sizeof *msg;
4390 flow_extract(&payload, msg->arg, msg->port, &flow);
4392 /* Check with in-band control to see if this packet should be sent
4393 * to the local port regardless of the flow table. */
4394 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4395 union odp_action action;
4397 memset(&action, 0, sizeof(action));
4398 action.output.type = ODPAT_OUTPUT;
4399 action.output.port = ODPP_LOCAL;
4400 dpif_execute(p->dpif, &action, 1, &payload);
4403 rule = lookup_valid_rule(p, &flow);
4405 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4406 struct ofport *port = get_port(p, msg->port);
4408 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4409 COVERAGE_INC(ofproto_no_packet_in);
4410 /* XXX install 'drop' flow entry */
4411 ofpbuf_delete(packet);
4415 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4418 COVERAGE_INC(ofproto_packet_in);
4419 send_packet_in(p, packet);
4423 if (rule->cr.wc.wildcards) {
4424 rule = rule_create_subrule(p, rule, &flow);
4425 rule_make_actions(p, rule, packet);
4427 if (!rule->may_install) {
4428 /* The rule is not installable, that is, we need to process every
4429 * packet, so process the current packet and set its actions into
4431 rule_make_actions(p, rule, packet);
4433 /* XXX revalidate rule if it needs it */
4437 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4439 * Extra-special case for fail-open mode.
4441 * We are in fail-open mode and the packet matched the fail-open rule,
4442 * but we are connected to a controller too. We should send the packet
4443 * up to the controller in the hope that it will try to set up a flow
4444 * and thereby allow us to exit fail-open.
4446 * See the top-level comment in fail-open.c for more information.
4448 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4449 DPIF_RECV_MSG_PADDING));
4452 ofpbuf_pull(packet, sizeof *msg);
4453 rule_execute(p, rule, packet, &flow);
4454 rule_reinstall(p, rule);
4458 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4460 struct odp_msg *msg = packet->data;
4462 switch (msg->type) {
4463 case _ODPL_ACTION_NR:
4464 COVERAGE_INC(ofproto_ctlr_action);
4465 send_packet_in(p, packet);
4468 case _ODPL_SFLOW_NR:
4470 ofproto_sflow_received(p->sflow, msg);
4472 ofpbuf_delete(packet);
4476 handle_odp_miss_msg(p, packet);
4480 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4486 /* Flow expiration. */
4488 struct expire_cbdata {
4489 struct ofproto *ofproto;
4493 static int ofproto_dp_max_idle(const struct ofproto *);
4494 static void ofproto_update_used(struct ofproto *);
4495 static void rule_expire(struct cls_rule *, void *cbdata);
4497 /* This function is called periodically by ofproto_run(). Its job is to
4498 * collect updates for the flows that have been installed into the datapath,
4499 * most importantly when they last were used, and then use that information to
4500 * expire flows that have not been used recently.
4502 * Returns the number of milliseconds after which it should be called again. */
4504 ofproto_expire(struct ofproto *ofproto)
4506 struct expire_cbdata cbdata;
4508 /* Update 'used' for each flow in the datapath. */
4509 ofproto_update_used(ofproto);
4511 /* Expire idle flows.
4513 * A wildcarded flow is idle only when all of its subrules have expired due
4514 * to becoming idle, so iterate through the exact-match flows first. */
4515 cbdata.ofproto = ofproto;
4516 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4517 classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
4518 classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
4520 /* Let the hook know that we're at a stable point: all outstanding data
4521 * in existing flows has been accounted to the account_cb. Thus, the
4522 * hook can now reasonably do operations that depend on having accurate
4523 * flow volume accounting (currently, that's just bond rebalancing). */
4524 if (ofproto->ofhooks->account_checkpoint_cb) {
4525 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4528 return MIN(cbdata.dp_max_idle, 1000);
4531 /* Update 'used' member of each flow currently installed into the datapath. */
4533 ofproto_update_used(struct ofproto *p)
4535 struct odp_flow *flows;
4540 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4545 for (i = 0; i < n_flows; i++) {
4546 struct odp_flow *f = &flows[i];
4547 struct cls_rule target;
4551 odp_flow_key_to_flow(&f->key, &flow);
4552 cls_rule_init_exact(&flow, UINT16_MAX, &target);
4554 rule = rule_from_cls_rule(classifier_find_rule_exactly(&p->cls,
4557 if (rule && rule->installed) {
4558 update_time(p, rule, &f->stats);
4559 rule_account(p, rule, f->stats.n_bytes);
4561 /* There's a flow in the datapath that we know nothing about.
4563 COVERAGE_INC(ofproto_unexpected_rule);
4564 dpif_flow_del(p->dpif, f);
4571 /* Calculates and returns the number of milliseconds of idle time after which
4572 * flows should expire from the datapath and we should fold their statistics
4573 * into their parent rules in userspace. */
4575 ofproto_dp_max_idle(const struct ofproto *ofproto)
4578 * Idle time histogram.
4580 * Most of the time a switch has a relatively small number of flows. When
4581 * this is the case we might as well keep statistics for all of them in
4582 * userspace and to cache them in the kernel datapath for performance as
4585 * As the number of flows increases, the memory required to maintain
4586 * statistics about them in userspace and in the kernel becomes
4587 * significant. However, with a large number of flows it is likely that
4588 * only a few of them are "heavy hitters" that consume a large amount of
4589 * bandwidth. At this point, only heavy hitters are worth caching in the
4590 * kernel and maintaining in userspaces; other flows we can discard.
4592 * The technique used to compute the idle time is to build a histogram with
4593 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4594 * is installed in the kernel gets dropped in the appropriate bucket.
4595 * After the histogram has been built, we compute the cutoff so that only
4596 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4597 * cached. At least the most-recently-used bucket of flows is kept, so
4598 * actually an arbitrary number of flows can be kept in any given
4599 * expiration run (though the next run will delete most of those unless
4600 * they receive additional data).
4602 * This requires a second pass through the exact-match flows, in addition
4603 * to the pass made by ofproto_update_used(), because the former function
4604 * never looks at uninstallable flows.
4606 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4607 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4608 int buckets[N_BUCKETS] = { 0 };
4614 total = classifier_count_exact(&ofproto->cls);
4615 if (total <= 1000) {
4616 return N_BUCKETS * BUCKET_WIDTH;
4619 /* Build histogram. */
4621 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) {
4622 long long int idle = now - rule->used;
4623 int bucket = (idle <= 0 ? 0
4624 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4625 : (unsigned int) idle / BUCKET_WIDTH);
4629 /* Find the first bucket whose flows should be expired. */
4630 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4631 if (buckets[bucket]) {
4634 subtotal += buckets[bucket++];
4635 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4640 if (VLOG_IS_DBG_ENABLED()) {
4644 ds_put_cstr(&s, "keep");
4645 for (i = 0; i < N_BUCKETS; i++) {
4647 ds_put_cstr(&s, ", drop");
4650 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4653 VLOG_INFO("%s: %s (msec:count)",
4654 dpif_name(ofproto->dpif), ds_cstr(&s));
4658 return bucket * BUCKET_WIDTH;
4662 rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
4664 if (ofproto->netflow && !is_controller_rule(rule) &&
4665 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4666 struct ofexpired expired;
4667 struct odp_flow odp_flow;
4669 /* Get updated flow stats.
4671 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4672 * updated TCP flags and (2) the dpif_flow_list_all() in
4673 * ofproto_update_used() zeroed TCP flags. */
4674 memset(&odp_flow, 0, sizeof odp_flow);
4675 if (rule->installed) {
4676 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
4677 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4678 dpif_flow_get(ofproto->dpif, &odp_flow);
4680 if (odp_flow.stats.n_packets) {
4681 update_time(ofproto, rule, &odp_flow.stats);
4682 netflow_flow_update_flags(&rule->nf_flow,
4683 odp_flow.stats.tcp_flags);
4687 expired.flow = rule->cr.flow;
4688 expired.packet_count = rule->packet_count +
4689 odp_flow.stats.n_packets;
4690 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4691 expired.used = rule->used;
4693 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4697 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4698 * rules, then delete it entirely.
4700 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4701 * the datapath and fold its statistics back into its super-rule.
4703 * (This is a callback function for classifier_for_each().) */
4705 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4707 struct expire_cbdata *cbdata = cbdata_;
4708 struct ofproto *ofproto = cbdata->ofproto;
4709 struct rule *rule = rule_from_cls_rule(cls_rule);
4710 long long int hard_expire, idle_expire, expire, now;
4712 /* Calculate OpenFlow expiration times for 'rule'. */
4713 hard_expire = (rule->hard_timeout
4714 ? rule->created + rule->hard_timeout * 1000
4716 idle_expire = (rule->idle_timeout
4717 && (rule->super || list_is_empty(&rule->list))
4718 ? rule->used + rule->idle_timeout * 1000
4720 expire = MIN(hard_expire, idle_expire);
4724 /* 'rule' has not expired according to OpenFlow rules. */
4725 if (!rule->cr.wc.wildcards) {
4726 if (now >= rule->used + cbdata->dp_max_idle) {
4727 /* This rule is idle, so drop it to free up resources. */
4729 /* It's not part of the OpenFlow flow table, so we can
4730 * delete it entirely and fold its statistics into its
4732 rule_remove(ofproto, rule);
4734 /* It is part of the OpenFlow flow table, so we have to
4735 * keep the rule but we can at least uninstall it from the
4737 rule_uninstall(ofproto, rule);
4740 /* Send NetFlow active timeout if appropriate. */
4741 rule_active_timeout(cbdata->ofproto, rule);
4745 /* 'rule' has expired according to OpenFlow rules. */
4746 COVERAGE_INC(ofproto_expired);
4748 /* Update stats. (This is a no-op if the rule expired due to an idle
4749 * timeout, because that only happens when the rule has no subrules
4751 if (rule->cr.wc.wildcards) {
4752 struct rule *subrule, *next;
4753 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
4754 rule_remove(cbdata->ofproto, subrule);
4757 rule_uninstall(cbdata->ofproto, rule);
4760 /* Get rid of the rule. */
4761 if (!rule_is_hidden(rule)) {
4762 send_flow_removed(cbdata->ofproto, rule,
4764 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4766 rule_remove(cbdata->ofproto, rule);
4771 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4773 struct rule *sub = rule_from_cls_rule(sub_);
4774 struct revalidate_cbdata *cbdata = cbdata_;
4776 if (cbdata->revalidate_all
4777 || (cbdata->revalidate_subrules && sub->super)
4778 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4779 revalidate_rule(cbdata->ofproto, sub);
4784 revalidate_rule(struct ofproto *p, struct rule *rule)
4786 const struct flow *flow = &rule->cr.flow;
4788 COVERAGE_INC(ofproto_revalidate_rule);
4791 super = rule_from_cls_rule(classifier_lookup(&p->cls, flow,
4794 rule_remove(p, rule);
4796 } else if (super != rule->super) {
4797 COVERAGE_INC(ofproto_revalidate_moved);
4798 list_remove(&rule->list);
4799 list_push_back(&super->list, &rule->list);
4800 rule->super = super;
4801 rule->hard_timeout = super->hard_timeout;
4802 rule->idle_timeout = super->idle_timeout;
4803 rule->created = super->created;
4808 rule_update_actions(p, rule);
4812 static struct ofpbuf *
4813 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4816 struct ofp_flow_removed *ofr;
4819 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4820 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, ofconn->flow_format,
4822 ofr->cookie = rule->flow_cookie;
4823 ofr->priority = htons(rule->cr.priority);
4824 ofr->reason = reason;
4825 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4826 ofr->idle_timeout = htons(rule->idle_timeout);
4827 ofr->packet_count = htonll(rule->packet_count);
4828 ofr->byte_count = htonll(rule->byte_count);
4833 static struct ofpbuf *
4834 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4836 struct nx_flow_removed *nfr;
4840 nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf);
4842 match_len = nx_put_match(buf, &rule->cr);
4844 nfr->cookie = rule->flow_cookie;
4845 nfr->priority = htons(rule->cr.priority);
4846 nfr->reason = reason;
4847 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4848 nfr->idle_timeout = htons(rule->idle_timeout);
4849 nfr->match_len = htons(match_len);
4850 nfr->packet_count = htonll(rule->packet_count);
4851 nfr->byte_count = htonll(rule->byte_count);
4857 send_flow_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4859 struct ofconn *ofconn;
4861 if (!rule->send_flow_removed) {
4865 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4868 if (!rconn_is_connected(ofconn->rconn)
4869 || !ofconn_receives_async_msgs(ofconn)) {
4873 msg = (ofconn->flow_format == NXFF_NXM
4874 ? compose_nx_flow_removed(rule, reason)
4875 : compose_ofp_flow_removed(ofconn, rule, reason));
4877 /* Account flow expirations under ofconn->reply_counter, the counter
4878 * for replies to OpenFlow requests. That works because preventing
4879 * OpenFlow requests from being processed also prevents new flows from
4880 * being added (and expiring). (It also prevents processing OpenFlow
4881 * requests that would not add new flows, so it is imperfect.) */
4882 queue_tx(msg, ofconn, ofconn->reply_counter);
4886 /* pinsched callback for sending 'packet' on 'ofconn'. */
4888 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4890 struct ofconn *ofconn = ofconn_;
4892 rconn_send_with_limit(ofconn->rconn, packet,
4893 ofconn->packet_in_counter, 100);
4896 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4897 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4898 * packet scheduler for sending.
4900 * 'max_len' specifies the maximum number of bytes of the packet to send on
4901 * 'ofconn' (INT_MAX specifies no limit).
4903 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4904 * ownership is transferred to this function. */
4906 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4909 struct ofproto *ofproto = ofconn->ofproto;
4910 struct ofp_packet_in *opi = packet->data;
4911 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4912 int send_len, trim_size;
4916 if (opi->reason == OFPR_ACTION) {
4917 buffer_id = UINT32_MAX;
4918 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4919 buffer_id = pktbuf_get_null();
4920 } else if (!ofconn->pktbuf) {
4921 buffer_id = UINT32_MAX;
4923 struct ofpbuf payload;
4924 payload.data = opi->data;
4925 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4926 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4929 /* Figure out how much of the packet to send. */
4930 send_len = ntohs(opi->total_len);
4931 if (buffer_id != UINT32_MAX) {
4932 send_len = MIN(send_len, ofconn->miss_send_len);
4934 send_len = MIN(send_len, max_len);
4936 /* Adjust packet length and clone if necessary. */
4937 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4939 packet = ofpbuf_clone_data(packet->data, trim_size);
4942 packet->size = trim_size;
4945 /* Update packet headers. */
4946 opi->buffer_id = htonl(buffer_id);
4947 update_openflow_length(packet);
4949 /* Hand over to packet scheduler. It might immediately call into
4950 * do_send_packet_in() or it might buffer it for a while (until a later
4951 * call to pinsched_run()). */
4952 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4953 packet, do_send_packet_in, ofconn);
4956 /* Replace struct odp_msg header in 'packet' by equivalent struct
4957 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4958 * returned by dpif_recv()).
4960 * The conversion is not complete: the caller still needs to trim any unneeded
4961 * payload off the end of the buffer, set the length in the OpenFlow header,
4962 * and set buffer_id. Those require us to know the controller settings and so
4963 * must be done on a per-controller basis.
4965 * Returns the maximum number of bytes of the packet that should be sent to
4966 * the controller (INT_MAX if no limit). */
4968 do_convert_to_packet_in(struct ofpbuf *packet)
4970 struct odp_msg *msg = packet->data;
4971 struct ofp_packet_in *opi;
4977 /* Extract relevant header fields */
4978 if (msg->type == _ODPL_ACTION_NR) {
4979 reason = OFPR_ACTION;
4982 reason = OFPR_NO_MATCH;
4985 total_len = msg->length - sizeof *msg;
4986 in_port = odp_port_to_ofp_port(msg->port);
4988 /* Repurpose packet buffer by overwriting header. */
4989 ofpbuf_pull(packet, sizeof(struct odp_msg));
4990 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4991 opi->header.version = OFP_VERSION;
4992 opi->header.type = OFPT_PACKET_IN;
4993 opi->total_len = htons(total_len);
4994 opi->in_port = htons(in_port);
4995 opi->reason = reason;
5000 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
5001 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
5002 * as necessary according to their individual configurations.
5004 * 'packet' must have sufficient headroom to convert it into a struct
5005 * ofp_packet_in (e.g. as returned by dpif_recv()).
5007 * Takes ownership of 'packet'. */
5009 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
5011 struct ofconn *ofconn, *prev;
5014 max_len = do_convert_to_packet_in(packet);
5017 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5018 if (ofconn_receives_async_msgs(ofconn)) {
5020 schedule_packet_in(prev, packet, max_len, true);
5026 schedule_packet_in(prev, packet, max_len, false);
5028 ofpbuf_delete(packet);
5033 pick_datapath_id(const struct ofproto *ofproto)
5035 const struct ofport *port;
5037 port = get_port(ofproto, ODPP_LOCAL);
5039 uint8_t ea[ETH_ADDR_LEN];
5042 error = netdev_get_etheraddr(port->netdev, ea);
5044 return eth_addr_to_uint64(ea);
5046 VLOG_WARN("could not get MAC address for %s (%s)",
5047 netdev_get_name(port->netdev), strerror(error));
5049 return ofproto->fallback_dpid;
5053 pick_fallback_dpid(void)
5055 uint8_t ea[ETH_ADDR_LEN];
5056 eth_addr_nicira_random(ea);
5057 return eth_addr_to_uint64(ea);
5061 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5062 struct odp_actions *actions, tag_type *tags,
5063 uint16_t *nf_output_iface, void *ofproto_)
5065 struct ofproto *ofproto = ofproto_;
5068 /* Drop frames for reserved multicast addresses. */
5069 if (eth_addr_is_reserved(flow->dl_dst)) {
5073 /* Learn source MAC (but don't try to learn from revalidation). */
5074 if (packet != NULL) {
5075 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5077 GRAT_ARP_LOCK_NONE);
5079 /* The log messages here could actually be useful in debugging,
5080 * so keep the rate limit relatively high. */
5081 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5082 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5083 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5084 ofproto_revalidate(ofproto, rev_tag);
5088 /* Determine output port. */
5089 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5092 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5093 nf_output_iface, actions);
5094 } else if (out_port != flow->in_port) {
5095 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
5096 *nf_output_iface = out_port;
5104 static const struct ofhooks default_ofhooks = {
5105 default_normal_ofhook_cb,