2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
40 #include "ofp-print.h"
42 #include "ofproto-sflow.h"
44 #include "openflow/nicira-ext.h"
45 #include "openflow/openflow.h"
46 #include "openvswitch/datapath-protocol.h"
50 #include "poll-loop.h"
54 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto)
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER = 1
74 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
75 struct netdev *netdev;
76 struct ofp_phy_port opp; /* In host byte order. */
80 static void ofport_free(struct ofport *);
81 static void hton_ofp_phy_port(struct ofp_phy_port *);
83 static int xlate_actions(const union ofp_action *in, size_t n_in,
84 const flow_t *flow, struct ofproto *ofproto,
85 const struct ofpbuf *packet,
86 struct odp_actions *out, tag_type *tags,
87 bool *may_set_up_flow, uint16_t *nf_output_iface);
92 uint64_t flow_cookie; /* Controller-issued identifier.
93 (Kept in network-byte order.) */
94 uint16_t idle_timeout; /* In seconds from time of last use. */
95 uint16_t hard_timeout; /* In seconds from time of creation. */
96 bool send_flow_removed; /* Send a flow removed message? */
97 long long int used; /* Last-used time (0 if never used). */
98 long long int created; /* Creation time. */
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
101 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
102 tag_type tags; /* Tags (set only by hooks). */
103 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
105 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
106 * exact-match rule (having cr.wc.wildcards of 0) generated from the
107 * wildcard rule 'super'. In this case, 'list' is an element of the
110 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
111 * a list of subrules. A super-rule with no wildcards (where
112 * cr.wc.wildcards is 0) will never have any subrules. */
118 * 'n_actions' is the number of elements in the 'actions' array. A single
119 * action may take up more more than one element's worth of space.
121 * A subrule has no actions (it uses the super-rule's actions). */
123 union ofp_action *actions;
127 * A super-rule with wildcard fields never has ODP actions (since the
128 * datapath only supports exact-match flows). */
129 bool installed; /* Installed in datapath? */
130 bool may_install; /* True ordinarily; false if actions must
131 * be reassessed for every packet. */
133 union odp_action *odp_actions;
137 rule_is_hidden(const struct rule *rule)
139 /* Subrules are merely an implementation detail, so hide them from the
141 if (rule->super != NULL) {
145 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
146 * (e.g. by in-band control) and are intentionally hidden from the
148 if (rule->cr.priority > UINT16_MAX) {
155 static struct rule *rule_create(struct ofproto *, struct rule *super,
156 const union ofp_action *, size_t n_actions,
157 uint16_t idle_timeout, uint16_t hard_timeout,
158 uint64_t flow_cookie, bool send_flow_removed);
159 static void rule_free(struct rule *);
160 static void rule_destroy(struct ofproto *, struct rule *);
161 static struct rule *rule_from_cls_rule(const struct cls_rule *);
162 static void rule_insert(struct ofproto *, struct rule *,
163 struct ofpbuf *packet, uint16_t in_port);
164 static void rule_remove(struct ofproto *, struct rule *);
165 static bool rule_make_actions(struct ofproto *, struct rule *,
166 const struct ofpbuf *packet);
167 static void rule_install(struct ofproto *, struct rule *,
168 struct rule *displaced_rule);
169 static void rule_uninstall(struct ofproto *, struct rule *);
170 static void rule_post_uninstall(struct ofproto *, struct rule *);
171 static void send_flow_removed(struct ofproto *p, struct rule *rule,
172 long long int now, uint8_t reason);
174 /* ofproto supports two kinds of OpenFlow connections:
176 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
177 * maintains persistent connections to these controllers and by default
178 * sends them asynchronous messages such as packet-ins.
180 * - "Service" connections, e.g. from ovs-ofctl. When these connections
181 * drop, it is the other side's responsibility to reconnect them if
182 * necessary. ofproto does not send them asynchronous messages by default.
184 * Currently, active (tcp, ssl, unix) connections are always "primary"
185 * connections and passive (ptcp, pssl, punix) connections are always "service"
186 * connections. There is no inherent reason for this, but it reflects the
190 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
191 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
194 /* A listener for incoming OpenFlow "service" connections. */
196 struct hmap_node node; /* In struct ofproto's "services" hmap. */
197 struct pvconn *pvconn; /* OpenFlow connection listener. */
199 /* These are not used by ofservice directly. They are settings for
200 * accepted "struct ofconn"s from the pvconn. */
201 int probe_interval; /* Max idle time before probing, in seconds. */
202 int rate_limit; /* Max packet-in rate in packets per second. */
203 int burst_limit; /* Limit on accumulating packet credits. */
206 static struct ofservice *ofservice_lookup(struct ofproto *,
208 static int ofservice_create(struct ofproto *,
209 const struct ofproto_controller *);
210 static void ofservice_reconfigure(struct ofservice *,
211 const struct ofproto_controller *);
212 static void ofservice_destroy(struct ofproto *, struct ofservice *);
214 /* An OpenFlow connection. */
216 struct ofproto *ofproto; /* The ofproto that owns this connection. */
217 struct list node; /* In struct ofproto's "all_conns" list. */
218 struct rconn *rconn; /* OpenFlow connection. */
219 enum ofconn_type type; /* Type. */
221 /* OFPT_PACKET_IN related data. */
222 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
223 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
224 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
225 int miss_send_len; /* Bytes to send of buffered packets. */
227 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
228 * requests, and the maximum number before we stop reading OpenFlow
230 #define OFCONN_REPLY_MAX 100
231 struct rconn_packet_counter *reply_counter;
233 /* type == OFCONN_PRIMARY only. */
234 enum nx_role role; /* Role. */
235 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
236 struct discovery *discovery; /* Controller discovery object, if enabled. */
237 struct status_category *ss; /* Switch status category. */
238 enum ofproto_band band; /* In-band or out-of-band? */
241 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
242 * "schedulers" array. Their values are 0 and 1, and their meanings and values
243 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
244 * case anything ever changes, check their values here. */
245 #define N_SCHEDULERS 2
246 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
247 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
248 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
249 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
251 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
253 static void ofconn_destroy(struct ofconn *);
254 static void ofconn_run(struct ofconn *, struct ofproto *);
255 static void ofconn_wait(struct ofconn *);
256 static bool ofconn_receives_async_msgs(const struct ofconn *);
257 static char *ofconn_make_name(const struct ofproto *, const char *target);
258 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
260 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
261 struct rconn_packet_counter *counter);
263 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
264 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
268 uint64_t datapath_id; /* Datapath ID. */
269 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
270 char *mfr_desc; /* Manufacturer. */
271 char *hw_desc; /* Hardware. */
272 char *sw_desc; /* Software version. */
273 char *serial_desc; /* Serial number. */
274 char *dp_desc; /* Datapath description. */
278 struct netdev_monitor *netdev_monitor;
279 struct hmap ports; /* Contains "struct ofport"s. */
280 struct shash port_by_name;
284 struct switch_status *switch_status;
285 struct fail_open *fail_open;
286 struct netflow *netflow;
287 struct ofproto_sflow *sflow;
289 /* In-band control. */
290 struct in_band *in_band;
291 long long int next_in_band_update;
292 struct sockaddr_in *extra_in_band_remotes;
293 size_t n_extra_remotes;
296 struct classifier cls;
297 bool need_revalidate;
298 long long int next_expiration;
299 struct tag_set revalidate_set;
300 bool tun_id_from_cookie;
302 /* OpenFlow connections. */
303 struct hmap controllers; /* Controller "struct ofconn"s. */
304 struct list all_conns; /* Contains "struct ofconn"s. */
305 enum ofproto_fail_mode fail_mode;
307 /* OpenFlow listeners. */
308 struct hmap services; /* Contains "struct ofservice"s. */
309 struct pvconn **snoops;
312 /* Hooks for ovs-vswitchd. */
313 const struct ofhooks *ofhooks;
316 /* Used by default ofhooks. */
317 struct mac_learning *ml;
320 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
322 static const struct ofhooks default_ofhooks;
324 static uint64_t pick_datapath_id(const struct ofproto *);
325 static uint64_t pick_fallback_dpid(void);
327 static void update_used(struct ofproto *);
328 static void update_stats(struct ofproto *, struct rule *,
329 const struct odp_flow_stats *);
330 static void expire_rule(struct cls_rule *, void *ofproto);
331 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
332 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
333 static void revalidate_cb(struct cls_rule *rule_, void *p_);
335 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
337 static void handle_openflow(struct ofconn *, struct ofproto *,
340 static void refresh_port_groups(struct ofproto *);
342 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
343 static void update_port(struct ofproto *, const char *devname);
344 static int init_ports(struct ofproto *);
345 static void reinit_ports(struct ofproto *);
348 ofproto_create(const char *datapath, const char *datapath_type,
349 const struct ofhooks *ofhooks, void *aux,
350 struct ofproto **ofprotop)
352 struct odp_stats stats;
359 /* Connect to datapath and start listening for messages. */
360 error = dpif_open(datapath, datapath_type, &dpif);
362 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
365 error = dpif_get_dp_stats(dpif, &stats);
367 VLOG_ERR("failed to obtain stats for datapath %s: %s",
368 datapath, strerror(error));
372 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
374 VLOG_ERR("failed to listen on datapath %s: %s",
375 datapath, strerror(error));
379 dpif_flow_flush(dpif);
380 dpif_recv_purge(dpif);
382 /* Initialize settings. */
383 p = xzalloc(sizeof *p);
384 p->fallback_dpid = pick_fallback_dpid();
385 p->datapath_id = p->fallback_dpid;
386 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
387 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
388 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
389 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
390 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
392 /* Initialize datapath. */
394 p->netdev_monitor = netdev_monitor_create();
395 hmap_init(&p->ports);
396 shash_init(&p->port_by_name);
397 p->max_ports = stats.max_ports;
399 /* Initialize submodules. */
400 p->switch_status = switch_status_create(p);
406 /* Initialize flow table. */
407 classifier_init(&p->cls);
408 p->need_revalidate = false;
409 p->next_expiration = time_msec() + 1000;
410 tag_set_init(&p->revalidate_set);
412 /* Initialize OpenFlow connections. */
413 list_init(&p->all_conns);
414 hmap_init(&p->controllers);
415 hmap_init(&p->services);
419 /* Initialize hooks. */
421 p->ofhooks = ofhooks;
425 p->ofhooks = &default_ofhooks;
427 p->ml = mac_learning_create();
430 /* Pick final datapath ID. */
431 p->datapath_id = pick_datapath_id(p);
432 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
439 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
441 uint64_t old_dpid = p->datapath_id;
442 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
443 if (p->datapath_id != old_dpid) {
444 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
446 /* Force all active connections to reconnect, since there is no way to
447 * notify a controller that the datapath ID has changed. */
448 ofproto_reconnect_controllers(p);
453 is_discovery_controller(const struct ofproto_controller *c)
455 return !strcmp(c->target, "discover");
459 is_in_band_controller(const struct ofproto_controller *c)
461 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
464 /* Creates a new controller in 'ofproto'. Some of the settings are initially
465 * drawn from 'c', but update_controller() needs to be called later to finish
466 * the new ofconn's configuration. */
468 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
470 struct discovery *discovery;
471 struct ofconn *ofconn;
473 if (is_discovery_controller(c)) {
474 int error = discovery_create(c->accept_re, c->update_resolv_conf,
475 ofproto->dpif, ofproto->switch_status,
484 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
485 ofconn->pktbuf = pktbuf_create();
486 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
488 ofconn->discovery = discovery;
490 char *name = ofconn_make_name(ofproto, c->target);
491 rconn_connect(ofconn->rconn, c->target, name);
494 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
495 hash_string(c->target, 0));
498 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
499 * target or turn discovery on or off (these are done by creating new ofconns
500 * and deleting old ones), but it can update the rest of an ofconn's
503 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
507 ofconn->band = (is_in_band_controller(c)
508 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
510 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
512 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
513 rconn_set_probe_interval(ofconn->rconn, probe_interval);
515 if (ofconn->discovery) {
516 discovery_set_update_resolv_conf(ofconn->discovery,
517 c->update_resolv_conf);
518 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
521 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
525 ofconn_get_target(const struct ofconn *ofconn)
527 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
530 static struct ofconn *
531 find_controller_by_target(struct ofproto *ofproto, const char *target)
533 struct ofconn *ofconn;
535 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
536 hash_string(target, 0), &ofproto->controllers) {
537 if (!strcmp(ofconn_get_target(ofconn), target)) {
545 update_in_band_remotes(struct ofproto *ofproto)
547 const struct ofconn *ofconn;
548 struct sockaddr_in *addrs;
549 size_t max_addrs, n_addrs;
553 /* Allocate enough memory for as many remotes as we could possibly have. */
554 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
555 addrs = xmalloc(max_addrs * sizeof *addrs);
558 /* Add all the remotes. */
560 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
561 struct sockaddr_in *sin = &addrs[n_addrs];
563 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
567 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
568 if (sin->sin_addr.s_addr) {
569 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
572 if (ofconn->discovery) {
576 for (i = 0; i < ofproto->n_extra_remotes; i++) {
577 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
580 /* Create or update or destroy in-band.
582 * Ordinarily we only enable in-band if there's at least one remote
583 * address, but discovery needs the in-band rules for DHCP to be installed
584 * even before we know any remote addresses. */
585 if (n_addrs || discovery) {
586 if (!ofproto->in_band) {
587 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
590 if (ofproto->in_band) {
591 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
593 ofproto->next_in_band_update = time_msec() + 1000;
595 in_band_destroy(ofproto->in_band);
596 ofproto->in_band = NULL;
604 update_fail_open(struct ofproto *p)
606 struct ofconn *ofconn;
608 if (!hmap_is_empty(&p->controllers)
609 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
610 struct rconn **rconns;
614 p->fail_open = fail_open_create(p, p->switch_status);
618 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
619 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
620 rconns[n++] = ofconn->rconn;
623 fail_open_set_controllers(p->fail_open, rconns, n);
624 /* p->fail_open takes ownership of 'rconns'. */
626 fail_open_destroy(p->fail_open);
632 ofproto_set_controllers(struct ofproto *p,
633 const struct ofproto_controller *controllers,
634 size_t n_controllers)
636 struct shash new_controllers;
637 struct ofconn *ofconn, *next_ofconn;
638 struct ofservice *ofservice, *next_ofservice;
642 /* Create newly configured controllers and services.
643 * Create a name to ofproto_controller mapping in 'new_controllers'. */
644 shash_init(&new_controllers);
645 for (i = 0; i < n_controllers; i++) {
646 const struct ofproto_controller *c = &controllers[i];
648 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
649 if (!find_controller_by_target(p, c->target)) {
650 add_controller(p, c);
652 } else if (!pvconn_verify_name(c->target)) {
653 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
657 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
658 dpif_name(p->dpif), c->target);
662 shash_add_once(&new_controllers, c->target, &controllers[i]);
665 /* Delete controllers that are no longer configured.
666 * Update configuration of all now-existing controllers. */
668 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node,
670 struct ofproto_controller *c;
672 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
674 ofconn_destroy(ofconn);
676 update_controller(ofconn, c);
683 /* Delete services that are no longer configured.
684 * Update configuration of all now-existing services. */
685 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
687 struct ofproto_controller *c;
689 c = shash_find_data(&new_controllers,
690 pvconn_get_name(ofservice->pvconn));
692 ofservice_destroy(p, ofservice);
694 ofservice_reconfigure(ofservice, c);
698 shash_destroy(&new_controllers);
700 update_in_band_remotes(p);
703 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
704 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
705 struct ofconn, hmap_node);
706 ofconn->ss = switch_status_register(p->switch_status, "remote",
707 rconn_status_cb, ofconn->rconn);
712 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
714 p->fail_mode = fail_mode;
718 /* Drops the connections between 'ofproto' and all of its controllers, forcing
719 * them to reconnect. */
721 ofproto_reconnect_controllers(struct ofproto *ofproto)
723 struct ofconn *ofconn;
725 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
726 rconn_reconnect(ofconn->rconn);
731 any_extras_changed(const struct ofproto *ofproto,
732 const struct sockaddr_in *extras, size_t n)
736 if (n != ofproto->n_extra_remotes) {
740 for (i = 0; i < n; i++) {
741 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
742 const struct sockaddr_in *new = &extras[i];
744 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
745 old->sin_port != new->sin_port) {
753 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
754 * in-band control should guarantee access, in the same way that in-band
755 * control guarantees access to OpenFlow controllers. */
757 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
758 const struct sockaddr_in *extras, size_t n)
760 if (!any_extras_changed(ofproto, extras, n)) {
764 free(ofproto->extra_in_band_remotes);
765 ofproto->n_extra_remotes = n;
766 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
768 update_in_band_remotes(ofproto);
772 ofproto_set_desc(struct ofproto *p,
773 const char *mfr_desc, const char *hw_desc,
774 const char *sw_desc, const char *serial_desc,
777 struct ofp_desc_stats *ods;
780 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
781 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
782 sizeof ods->mfr_desc);
785 p->mfr_desc = xstrdup(mfr_desc);
788 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
789 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
790 sizeof ods->hw_desc);
793 p->hw_desc = xstrdup(hw_desc);
796 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
797 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
798 sizeof ods->sw_desc);
801 p->sw_desc = xstrdup(sw_desc);
804 if (strlen(serial_desc) >= sizeof ods->serial_num) {
805 VLOG_WARN("truncating serial_desc, must be less than %zu "
807 sizeof ods->serial_num);
809 free(p->serial_desc);
810 p->serial_desc = xstrdup(serial_desc);
813 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
814 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
815 sizeof ods->dp_desc);
818 p->dp_desc = xstrdup(dp_desc);
823 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
824 const struct svec *svec)
826 struct pvconn **pvconns = *pvconnsp;
827 size_t n_pvconns = *n_pvconnsp;
831 for (i = 0; i < n_pvconns; i++) {
832 pvconn_close(pvconns[i]);
836 pvconns = xmalloc(svec->n * sizeof *pvconns);
838 for (i = 0; i < svec->n; i++) {
839 const char *name = svec->names[i];
840 struct pvconn *pvconn;
843 error = pvconn_open(name, &pvconn);
845 pvconns[n_pvconns++] = pvconn;
847 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
855 *n_pvconnsp = n_pvconns;
861 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
863 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
867 ofproto_set_netflow(struct ofproto *ofproto,
868 const struct netflow_options *nf_options)
870 if (nf_options && nf_options->collectors.n) {
871 if (!ofproto->netflow) {
872 ofproto->netflow = netflow_create();
874 return netflow_set_options(ofproto->netflow, nf_options);
876 netflow_destroy(ofproto->netflow);
877 ofproto->netflow = NULL;
883 ofproto_set_sflow(struct ofproto *ofproto,
884 const struct ofproto_sflow_options *oso)
886 struct ofproto_sflow *os = ofproto->sflow;
889 struct ofport *ofport;
891 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
892 refresh_port_groups(ofproto);
893 HMAP_FOR_EACH (ofport, struct ofport, hmap_node, &ofproto->ports) {
894 ofproto_sflow_add_port(os, ofport->odp_port,
895 netdev_get_name(ofport->netdev));
898 ofproto_sflow_set_options(os, oso);
900 ofproto_sflow_destroy(os);
901 ofproto->sflow = NULL;
906 ofproto_get_datapath_id(const struct ofproto *ofproto)
908 return ofproto->datapath_id;
912 ofproto_has_primary_controller(const struct ofproto *ofproto)
914 return !hmap_is_empty(&ofproto->controllers);
917 enum ofproto_fail_mode
918 ofproto_get_fail_mode(const struct ofproto *p)
924 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
928 for (i = 0; i < ofproto->n_snoops; i++) {
929 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
934 ofproto_destroy(struct ofproto *p)
936 struct ofservice *ofservice, *next_ofservice;
937 struct ofconn *ofconn, *next_ofconn;
938 struct ofport *ofport, *next_ofport;
945 /* Destroy fail-open and in-band early, since they touch the classifier. */
946 fail_open_destroy(p->fail_open);
949 in_band_destroy(p->in_band);
951 free(p->extra_in_band_remotes);
953 ofproto_flush_flows(p);
954 classifier_destroy(&p->cls);
956 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
958 ofconn_destroy(ofconn);
960 hmap_destroy(&p->controllers);
963 netdev_monitor_destroy(p->netdev_monitor);
964 HMAP_FOR_EACH_SAFE (ofport, next_ofport, struct ofport, hmap_node,
966 hmap_remove(&p->ports, &ofport->hmap_node);
969 shash_destroy(&p->port_by_name);
971 switch_status_destroy(p->switch_status);
972 netflow_destroy(p->netflow);
973 ofproto_sflow_destroy(p->sflow);
975 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
977 ofservice_destroy(p, ofservice);
979 hmap_destroy(&p->services);
981 for (i = 0; i < p->n_snoops; i++) {
982 pvconn_close(p->snoops[i]);
986 mac_learning_destroy(p->ml);
991 free(p->serial_desc);
994 hmap_destroy(&p->ports);
1000 ofproto_run(struct ofproto *p)
1002 int error = ofproto_run1(p);
1004 error = ofproto_run2(p, false);
1010 process_port_change(struct ofproto *ofproto, int error, char *devname)
1012 if (error == ENOBUFS) {
1013 reinit_ports(ofproto);
1014 } else if (!error) {
1015 update_port(ofproto, devname);
1020 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1021 * means that 'ofconn' is more interesting for monitoring than a lower return
1024 snoop_preference(const struct ofconn *ofconn)
1026 switch (ofconn->role) {
1027 case NX_ROLE_MASTER:
1034 /* Shouldn't happen. */
1039 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1040 * Connects this vconn to a controller. */
1042 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1044 struct ofconn *ofconn, *best;
1046 /* Pick a controller for monitoring. */
1048 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1049 if (ofconn->type == OFCONN_PRIMARY
1050 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1056 rconn_add_monitor(best->rconn, vconn);
1058 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1064 ofproto_run1(struct ofproto *p)
1066 struct ofconn *ofconn, *next_ofconn;
1067 struct ofservice *ofservice;
1072 if (shash_is_empty(&p->port_by_name)) {
1076 for (i = 0; i < 50; i++) {
1079 error = dpif_recv(p->dpif, &buf);
1081 if (error == ENODEV) {
1082 /* Someone destroyed the datapath behind our back. The caller
1083 * better destroy us and give up, because we're just going to
1084 * spin from here on out. */
1085 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1086 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1087 dpif_name(p->dpif));
1093 handle_odp_msg(p, buf);
1096 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1097 process_port_change(p, error, devname);
1099 while ((error = netdev_monitor_poll(p->netdev_monitor,
1100 &devname)) != EAGAIN) {
1101 process_port_change(p, error, devname);
1105 if (time_msec() >= p->next_in_band_update) {
1106 update_in_band_remotes(p);
1108 in_band_run(p->in_band);
1111 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1113 ofconn_run(ofconn, p);
1116 /* Fail-open maintenance. Do this after processing the ofconns since
1117 * fail-open checks the status of the controller rconn. */
1119 fail_open_run(p->fail_open);
1122 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1123 struct vconn *vconn;
1126 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1128 struct rconn *rconn;
1131 rconn = rconn_create(ofservice->probe_interval, 0);
1132 name = ofconn_make_name(p, vconn_get_name(vconn));
1133 rconn_connect_unreliably(rconn, vconn, name);
1136 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1137 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1138 ofservice->burst_limit);
1139 } else if (retval != EAGAIN) {
1140 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1144 for (i = 0; i < p->n_snoops; i++) {
1145 struct vconn *vconn;
1148 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1150 add_snooper(p, vconn);
1151 } else if (retval != EAGAIN) {
1152 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1156 if (time_msec() >= p->next_expiration) {
1157 COVERAGE_INC(ofproto_expiration);
1158 p->next_expiration = time_msec() + 1000;
1161 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1163 /* Let the hook know that we're at a stable point: all outstanding data
1164 * in existing flows has been accounted to the account_cb. Thus, the
1165 * hook can now reasonably do operations that depend on having accurate
1166 * flow volume accounting (currently, that's just bond rebalancing). */
1167 if (p->ofhooks->account_checkpoint_cb) {
1168 p->ofhooks->account_checkpoint_cb(p->aux);
1173 netflow_run(p->netflow);
1176 ofproto_sflow_run(p->sflow);
1182 struct revalidate_cbdata {
1183 struct ofproto *ofproto;
1184 bool revalidate_all; /* Revalidate all exact-match rules? */
1185 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1186 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1190 ofproto_run2(struct ofproto *p, bool revalidate_all)
1192 if (p->need_revalidate || revalidate_all
1193 || !tag_set_is_empty(&p->revalidate_set)) {
1194 struct revalidate_cbdata cbdata;
1196 cbdata.revalidate_all = revalidate_all;
1197 cbdata.revalidate_subrules = p->need_revalidate;
1198 cbdata.revalidate_set = p->revalidate_set;
1199 tag_set_init(&p->revalidate_set);
1200 COVERAGE_INC(ofproto_revalidate);
1201 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1202 p->need_revalidate = false;
1209 ofproto_wait(struct ofproto *p)
1211 struct ofservice *ofservice;
1212 struct ofconn *ofconn;
1215 dpif_recv_wait(p->dpif);
1216 dpif_port_poll_wait(p->dpif);
1217 netdev_monitor_poll_wait(p->netdev_monitor);
1218 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1219 ofconn_wait(ofconn);
1222 poll_timer_wait_until(p->next_in_band_update);
1223 in_band_wait(p->in_band);
1226 fail_open_wait(p->fail_open);
1229 ofproto_sflow_wait(p->sflow);
1231 if (!tag_set_is_empty(&p->revalidate_set)) {
1232 poll_immediate_wake();
1234 if (p->need_revalidate) {
1235 /* Shouldn't happen, but if it does just go around again. */
1236 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1237 poll_immediate_wake();
1238 } else if (p->next_expiration != LLONG_MAX) {
1239 poll_timer_wait_until(p->next_expiration);
1241 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1242 pvconn_wait(ofservice->pvconn);
1244 for (i = 0; i < p->n_snoops; i++) {
1245 pvconn_wait(p->snoops[i]);
1250 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1252 tag_set_add(&ofproto->revalidate_set, tag);
1256 ofproto_get_revalidate_set(struct ofproto *ofproto)
1258 return &ofproto->revalidate_set;
1262 ofproto_is_alive(const struct ofproto *p)
1264 return !hmap_is_empty(&p->controllers);
1268 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1269 const union ofp_action *actions, size_t n_actions,
1270 const struct ofpbuf *packet)
1272 struct odp_actions odp_actions;
1275 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1281 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1283 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1284 odp_actions.n_actions, packet);
1289 ofproto_add_flow(struct ofproto *p,
1290 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1291 const union ofp_action *actions, size_t n_actions,
1295 rule = rule_create(p, NULL, actions, n_actions,
1296 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1298 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1299 rule_insert(p, rule, NULL, 0);
1303 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1304 uint32_t wildcards, unsigned int priority)
1308 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1312 rule_remove(ofproto, rule);
1317 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1319 struct rule *rule = rule_from_cls_rule(rule_);
1320 struct ofproto *ofproto = ofproto_;
1322 /* Mark the flow as not installed, even though it might really be
1323 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1324 * There is no point in uninstalling it individually since we are about to
1325 * blow away all the flows with dpif_flow_flush(). */
1326 rule->installed = false;
1328 rule_remove(ofproto, rule);
1332 ofproto_flush_flows(struct ofproto *ofproto)
1334 COVERAGE_INC(ofproto_flush);
1335 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1336 dpif_flow_flush(ofproto->dpif);
1337 if (ofproto->in_band) {
1338 in_band_flushed(ofproto->in_band);
1340 if (ofproto->fail_open) {
1341 fail_open_flushed(ofproto->fail_open);
1346 reinit_ports(struct ofproto *p)
1348 struct svec devnames;
1349 struct ofport *ofport;
1350 struct odp_port *odp_ports;
1354 svec_init(&devnames);
1355 HMAP_FOR_EACH (ofport, struct ofport, hmap_node, &p->ports) {
1356 svec_add (&devnames, (char *) ofport->opp.name);
1358 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1359 for (i = 0; i < n_odp_ports; i++) {
1360 svec_add (&devnames, odp_ports[i].devname);
1364 svec_sort_unique(&devnames);
1365 for (i = 0; i < devnames.n; i++) {
1366 update_port(p, devnames.names[i]);
1368 svec_destroy(&devnames);
1372 refresh_port_group(struct ofproto *p, unsigned int group)
1376 struct ofport *port;
1378 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1380 ports = xmalloc(hmap_count(&p->ports) * sizeof *ports);
1382 HMAP_FOR_EACH (port, struct ofport, hmap_node, &p->ports) {
1383 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1384 ports[n_ports++] = port->odp_port;
1387 dpif_port_group_set(p->dpif, group, ports, n_ports);
1394 refresh_port_groups(struct ofproto *p)
1396 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1397 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1399 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1403 static struct ofport *
1404 make_ofport(const struct odp_port *odp_port)
1406 struct netdev_options netdev_options;
1407 enum netdev_flags flags;
1408 struct ofport *ofport;
1409 struct netdev *netdev;
1413 memset(&netdev_options, 0, sizeof netdev_options);
1414 netdev_options.name = odp_port->devname;
1415 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1417 error = netdev_open(&netdev_options, &netdev);
1419 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1420 "cannot be opened (%s)",
1421 odp_port->devname, odp_port->port,
1422 odp_port->devname, strerror(error));
1426 ofport = xmalloc(sizeof *ofport);
1427 ofport->netdev = netdev;
1428 ofport->odp_port = odp_port->port;
1429 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1430 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1431 memcpy(ofport->opp.name, odp_port->devname,
1432 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1433 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1435 netdev_get_flags(netdev, &flags);
1436 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1438 netdev_get_carrier(netdev, &carrier);
1439 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1441 netdev_get_features(netdev,
1442 &ofport->opp.curr, &ofport->opp.advertised,
1443 &ofport->opp.supported, &ofport->opp.peer);
1448 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1450 if (get_port(p, odp_port->port)) {
1451 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1454 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1455 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1464 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1466 const struct ofp_phy_port *a = &a_->opp;
1467 const struct ofp_phy_port *b = &b_->opp;
1469 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1470 return (a->port_no == b->port_no
1471 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1472 && !strcmp((char *) a->name, (char *) b->name)
1473 && a->state == b->state
1474 && a->config == b->config
1475 && a->curr == b->curr
1476 && a->advertised == b->advertised
1477 && a->supported == b->supported
1478 && a->peer == b->peer);
1482 send_port_status(struct ofproto *p, const struct ofport *ofport,
1485 /* XXX Should limit the number of queued port status change messages. */
1486 struct ofconn *ofconn;
1487 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1488 struct ofp_port_status *ops;
1491 if (!ofconn_receives_async_msgs(ofconn)) {
1495 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1496 ops->reason = reason;
1497 ops->desc = ofport->opp;
1498 hton_ofp_phy_port(&ops->desc);
1499 queue_tx(b, ofconn, NULL);
1501 if (p->ofhooks->port_changed_cb) {
1502 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1507 ofport_install(struct ofproto *p, struct ofport *ofport)
1509 const char *netdev_name = (const char *) ofport->opp.name;
1511 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1512 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1513 shash_add(&p->port_by_name, netdev_name, ofport);
1515 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1520 ofport_remove(struct ofproto *p, struct ofport *ofport)
1522 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1523 hmap_remove(&p->ports, &ofport->hmap_node);
1524 shash_delete(&p->port_by_name,
1525 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1527 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1532 ofport_free(struct ofport *ofport)
1535 netdev_close(ofport->netdev);
1540 static struct ofport *
1541 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1543 struct ofport *port;
1545 HMAP_FOR_EACH_IN_BUCKET (port, struct ofport, hmap_node,
1546 hash_int(odp_port, 0), &ofproto->ports) {
1547 if (port->odp_port == odp_port) {
1555 update_port(struct ofproto *p, const char *devname)
1557 struct odp_port odp_port;
1558 struct ofport *old_ofport;
1559 struct ofport *new_ofport;
1562 COVERAGE_INC(ofproto_update_port);
1564 /* Query the datapath for port information. */
1565 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1567 /* Find the old ofport. */
1568 old_ofport = shash_find_data(&p->port_by_name, devname);
1571 /* There's no port named 'devname' but there might be a port with
1572 * the same port number. This could happen if a port is deleted
1573 * and then a new one added in its place very quickly, or if a port
1574 * is renamed. In the former case we want to send an OFPPR_DELETE
1575 * and an OFPPR_ADD, and in the latter case we want to send a
1576 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1577 * the old port's ifindex against the new port, or perhaps less
1578 * reliably but more portably by comparing the old port's MAC
1579 * against the new port's MAC. However, this code isn't that smart
1580 * and always sends an OFPPR_MODIFY (XXX). */
1581 old_ofport = get_port(p, odp_port.port);
1583 } else if (error != ENOENT && error != ENODEV) {
1584 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1585 "%s", strerror(error));
1589 /* Create a new ofport. */
1590 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1592 /* Eliminate a few pathological cases. */
1593 if (!old_ofport && !new_ofport) {
1595 } else if (old_ofport && new_ofport) {
1596 /* Most of the 'config' bits are OpenFlow soft state, but
1597 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1598 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1599 * leaves the other bits 0.) */
1600 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1602 if (ofport_equal(old_ofport, new_ofport)) {
1603 /* False alarm--no change. */
1604 ofport_free(new_ofport);
1609 /* Now deal with the normal cases. */
1611 ofport_remove(p, old_ofport);
1614 ofport_install(p, new_ofport);
1616 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1617 (!old_ofport ? OFPPR_ADD
1618 : !new_ofport ? OFPPR_DELETE
1620 ofport_free(old_ofport);
1622 /* Update port groups. */
1623 refresh_port_groups(p);
1627 init_ports(struct ofproto *p)
1629 struct odp_port *ports;
1634 error = dpif_port_list(p->dpif, &ports, &n_ports);
1639 for (i = 0; i < n_ports; i++) {
1640 const struct odp_port *odp_port = &ports[i];
1641 if (!ofport_conflicts(p, odp_port)) {
1642 struct ofport *ofport = make_ofport(odp_port);
1644 ofport_install(p, ofport);
1649 refresh_port_groups(p);
1653 static struct ofconn *
1654 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1656 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1657 ofconn->ofproto = p;
1658 list_push_back(&p->all_conns, &ofconn->node);
1659 ofconn->rconn = rconn;
1660 ofconn->type = type;
1661 ofconn->role = NX_ROLE_OTHER;
1662 ofconn->packet_in_counter = rconn_packet_counter_create ();
1663 ofconn->pktbuf = NULL;
1664 ofconn->miss_send_len = 0;
1665 ofconn->reply_counter = rconn_packet_counter_create ();
1670 ofconn_destroy(struct ofconn *ofconn)
1672 if (ofconn->type == OFCONN_PRIMARY) {
1673 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1675 discovery_destroy(ofconn->discovery);
1677 list_remove(&ofconn->node);
1678 switch_status_unregister(ofconn->ss);
1679 rconn_destroy(ofconn->rconn);
1680 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1681 rconn_packet_counter_destroy(ofconn->reply_counter);
1682 pktbuf_destroy(ofconn->pktbuf);
1687 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1692 if (ofconn->discovery) {
1693 char *controller_name;
1694 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1695 discovery_question_connectivity(ofconn->discovery);
1697 if (discovery_run(ofconn->discovery, &controller_name)) {
1698 if (controller_name) {
1699 char *ofconn_name = ofconn_make_name(p, controller_name);
1700 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1703 rconn_disconnect(ofconn->rconn);
1708 for (i = 0; i < N_SCHEDULERS; i++) {
1709 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1712 rconn_run(ofconn->rconn);
1714 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1715 /* Limit the number of iterations to prevent other tasks from
1717 for (iteration = 0; iteration < 50; iteration++) {
1718 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1723 fail_open_maybe_recover(p->fail_open);
1725 handle_openflow(ofconn, p, of_msg);
1726 ofpbuf_delete(of_msg);
1730 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1731 ofconn_destroy(ofconn);
1736 ofconn_wait(struct ofconn *ofconn)
1740 if (ofconn->discovery) {
1741 discovery_wait(ofconn->discovery);
1743 for (i = 0; i < N_SCHEDULERS; i++) {
1744 pinsched_wait(ofconn->schedulers[i]);
1746 rconn_run_wait(ofconn->rconn);
1747 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1748 rconn_recv_wait(ofconn->rconn);
1750 COVERAGE_INC(ofproto_ofconn_stuck);
1754 /* Returns true if 'ofconn' should receive asynchronous messages. */
1756 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1758 if (ofconn->type == OFCONN_PRIMARY) {
1759 /* Primary controllers always get asynchronous messages unless they
1760 * have configured themselves as "slaves". */
1761 return ofconn->role != NX_ROLE_SLAVE;
1763 /* Service connections don't get asynchronous messages unless they have
1764 * explicitly asked for them by setting a nonzero miss send length. */
1765 return ofconn->miss_send_len > 0;
1769 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1770 * and 'target', suitable for use in log messages for identifying the
1773 * The name is dynamically allocated. The caller should free it (with free())
1774 * when it is no longer needed. */
1776 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1778 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1782 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1786 for (i = 0; i < N_SCHEDULERS; i++) {
1787 struct pinsched **s = &ofconn->schedulers[i];
1791 *s = pinsched_create(rate, burst,
1792 ofconn->ofproto->switch_status);
1794 pinsched_set_limits(*s, rate, burst);
1797 pinsched_destroy(*s);
1804 ofservice_reconfigure(struct ofservice *ofservice,
1805 const struct ofproto_controller *c)
1807 ofservice->probe_interval = c->probe_interval;
1808 ofservice->rate_limit = c->rate_limit;
1809 ofservice->burst_limit = c->burst_limit;
1812 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1813 * positive errno value. */
1815 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1817 struct ofservice *ofservice;
1818 struct pvconn *pvconn;
1821 error = pvconn_open(c->target, &pvconn);
1826 ofservice = xzalloc(sizeof *ofservice);
1827 hmap_insert(&ofproto->services, &ofservice->node,
1828 hash_string(c->target, 0));
1829 ofservice->pvconn = pvconn;
1831 ofservice_reconfigure(ofservice, c);
1837 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1839 hmap_remove(&ofproto->services, &ofservice->node);
1840 pvconn_close(ofservice->pvconn);
1844 /* Finds and returns the ofservice within 'ofproto' that has the given
1845 * 'target', or a null pointer if none exists. */
1846 static struct ofservice *
1847 ofservice_lookup(struct ofproto *ofproto, const char *target)
1849 struct ofservice *ofservice;
1851 HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
1852 hash_string(target, 0), &ofproto->services) {
1853 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1860 /* Caller is responsible for initializing the 'cr' member of the returned
1862 static struct rule *
1863 rule_create(struct ofproto *ofproto, struct rule *super,
1864 const union ofp_action *actions, size_t n_actions,
1865 uint16_t idle_timeout, uint16_t hard_timeout,
1866 uint64_t flow_cookie, bool send_flow_removed)
1868 struct rule *rule = xzalloc(sizeof *rule);
1869 rule->idle_timeout = idle_timeout;
1870 rule->hard_timeout = hard_timeout;
1871 rule->flow_cookie = flow_cookie;
1872 rule->used = rule->created = time_msec();
1873 rule->send_flow_removed = send_flow_removed;
1874 rule->super = super;
1876 list_push_back(&super->list, &rule->list);
1878 list_init(&rule->list);
1880 rule->n_actions = n_actions;
1881 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1882 netflow_flow_clear(&rule->nf_flow);
1883 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1888 static struct rule *
1889 rule_from_cls_rule(const struct cls_rule *cls_rule)
1891 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1895 rule_free(struct rule *rule)
1897 free(rule->actions);
1898 free(rule->odp_actions);
1902 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1903 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1904 * through all of its subrules and revalidates them, destroying any that no
1905 * longer has a super-rule (which is probably all of them).
1907 * Before calling this function, the caller must make have removed 'rule' from
1908 * the classifier. If 'rule' is an exact-match rule, the caller is also
1909 * responsible for ensuring that it has been uninstalled from the datapath. */
1911 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1914 struct rule *subrule, *next;
1915 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1916 revalidate_rule(ofproto, subrule);
1919 list_remove(&rule->list);
1925 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1927 const union ofp_action *oa;
1928 struct actions_iterator i;
1930 if (out_port == htons(OFPP_NONE)) {
1933 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1934 oa = actions_next(&i)) {
1935 if (action_outputs_to_port(oa, out_port)) {
1942 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1943 * 'packet', which arrived on 'in_port'.
1945 * Takes ownership of 'packet'. */
1947 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1948 const union odp_action *actions, size_t n_actions,
1949 struct ofpbuf *packet)
1951 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1952 /* As an optimization, avoid a round-trip from userspace to kernel to
1953 * userspace. This also avoids possibly filling up kernel packet
1954 * buffers along the way. */
1955 struct odp_msg *msg;
1957 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1958 msg->type = _ODPL_ACTION_NR;
1959 msg->length = sizeof(struct odp_msg) + packet->size;
1960 msg->port = in_port;
1962 msg->arg = actions[0].controller.arg;
1964 send_packet_in(ofproto, packet);
1970 error = dpif_execute(ofproto->dpif, in_port,
1971 actions, n_actions, packet);
1972 ofpbuf_delete(packet);
1977 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1978 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1979 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1981 * The flow that 'packet' actually contains does not need to actually match
1982 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1983 * the packet and byte counters for 'rule' will be credited for the packet sent
1984 * out whether or not the packet actually matches 'rule'.
1986 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1987 * the caller must already have accurately composed ODP actions for it given
1988 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1989 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1990 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1991 * actions and apply them to 'packet'.
1993 * Takes ownership of 'packet'. */
1995 rule_execute(struct ofproto *ofproto, struct rule *rule,
1996 struct ofpbuf *packet, const flow_t *flow)
1998 const union odp_action *actions;
1999 struct odp_flow_stats stats;
2001 struct odp_actions a;
2003 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2005 /* Grab or compose the ODP actions.
2007 * The special case for an exact-match 'rule' where 'flow' is not the
2008 * rule's flow is important to avoid, e.g., sending a packet out its input
2009 * port simply because the ODP actions were composed for the wrong
2011 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
2012 struct rule *super = rule->super ? rule->super : rule;
2013 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
2014 packet, &a, NULL, 0, NULL)) {
2015 ofpbuf_delete(packet);
2018 actions = a.actions;
2019 n_actions = a.n_actions;
2021 actions = rule->odp_actions;
2022 n_actions = rule->n_odp_actions;
2025 /* Execute the ODP actions. */
2026 flow_extract_stats(flow, packet, &stats);
2027 if (execute_odp_actions(ofproto, flow->in_port,
2028 actions, n_actions, packet)) {
2029 update_stats(ofproto, rule, &stats);
2030 rule->used = time_msec();
2031 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2035 /* Inserts 'rule' into 'p''s flow table.
2037 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2038 * actions on it and credits the statistics for sending the packet to 'rule'.
2039 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2042 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2045 struct rule *displaced_rule;
2047 /* Insert the rule in the classifier. */
2048 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2049 if (!rule->cr.wc.wildcards) {
2050 rule_make_actions(p, rule, packet);
2053 /* Send the packet and credit it to the rule. */
2056 flow_extract(packet, 0, in_port, &flow);
2057 rule_execute(p, rule, packet, &flow);
2060 /* Install the rule in the datapath only after sending the packet, to
2061 * avoid packet reordering. */
2062 if (rule->cr.wc.wildcards) {
2063 COVERAGE_INC(ofproto_add_wc_flow);
2064 p->need_revalidate = true;
2066 rule_install(p, rule, displaced_rule);
2069 /* Free the rule that was displaced, if any. */
2070 if (displaced_rule) {
2071 rule_destroy(p, displaced_rule);
2075 static struct rule *
2076 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2079 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2080 rule->idle_timeout, rule->hard_timeout,
2082 COVERAGE_INC(ofproto_subrule_create);
2083 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2084 : rule->cr.priority), &subrule->cr);
2085 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2091 rule_remove(struct ofproto *ofproto, struct rule *rule)
2093 if (rule->cr.wc.wildcards) {
2094 COVERAGE_INC(ofproto_del_wc_flow);
2095 ofproto->need_revalidate = true;
2097 rule_uninstall(ofproto, rule);
2099 classifier_remove(&ofproto->cls, &rule->cr);
2100 rule_destroy(ofproto, rule);
2103 /* Returns true if the actions changed, false otherwise. */
2105 rule_make_actions(struct ofproto *p, struct rule *rule,
2106 const struct ofpbuf *packet)
2108 const struct rule *super;
2109 struct odp_actions a;
2112 assert(!rule->cr.wc.wildcards);
2114 super = rule->super ? rule->super : rule;
2116 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2117 packet, &a, &rule->tags, &rule->may_install,
2118 &rule->nf_flow.output_iface);
2120 actions_len = a.n_actions * sizeof *a.actions;
2121 if (rule->n_odp_actions != a.n_actions
2122 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2123 COVERAGE_INC(ofproto_odp_unchanged);
2124 free(rule->odp_actions);
2125 rule->n_odp_actions = a.n_actions;
2126 rule->odp_actions = xmemdup(a.actions, actions_len);
2134 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2135 struct odp_flow_put *put)
2137 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2138 put->flow.key = rule->cr.flow;
2139 put->flow.actions = rule->odp_actions;
2140 put->flow.n_actions = rule->n_odp_actions;
2141 put->flow.flags = 0;
2143 return dpif_flow_put(ofproto->dpif, put);
2147 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2149 assert(!rule->cr.wc.wildcards);
2151 if (rule->may_install) {
2152 struct odp_flow_put put;
2153 if (!do_put_flow(p, rule,
2154 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2156 rule->installed = true;
2157 if (displaced_rule) {
2158 update_stats(p, displaced_rule, &put.flow.stats);
2159 rule_post_uninstall(p, displaced_rule);
2162 } else if (displaced_rule) {
2163 rule_uninstall(p, displaced_rule);
2168 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2170 if (rule->installed) {
2171 struct odp_flow_put put;
2172 COVERAGE_INC(ofproto_dp_missed);
2173 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2175 rule_install(ofproto, rule, NULL);
2180 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2182 bool actions_changed;
2183 uint16_t new_out_iface, old_out_iface;
2185 old_out_iface = rule->nf_flow.output_iface;
2186 actions_changed = rule_make_actions(ofproto, rule, NULL);
2188 if (rule->may_install) {
2189 if (rule->installed) {
2190 if (actions_changed) {
2191 struct odp_flow_put put;
2192 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2193 | ODPPF_ZERO_STATS, &put);
2194 update_stats(ofproto, rule, &put.flow.stats);
2196 /* Temporarily set the old output iface so that NetFlow
2197 * messages have the correct output interface for the old
2199 new_out_iface = rule->nf_flow.output_iface;
2200 rule->nf_flow.output_iface = old_out_iface;
2201 rule_post_uninstall(ofproto, rule);
2202 rule->nf_flow.output_iface = new_out_iface;
2205 rule_install(ofproto, rule, NULL);
2208 rule_uninstall(ofproto, rule);
2213 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2215 uint64_t total_bytes = rule->byte_count + extra_bytes;
2217 if (ofproto->ofhooks->account_flow_cb
2218 && total_bytes > rule->accounted_bytes)
2220 ofproto->ofhooks->account_flow_cb(
2221 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2222 total_bytes - rule->accounted_bytes, ofproto->aux);
2223 rule->accounted_bytes = total_bytes;
2228 rule_uninstall(struct ofproto *p, struct rule *rule)
2230 assert(!rule->cr.wc.wildcards);
2231 if (rule->installed) {
2232 struct odp_flow odp_flow;
2234 odp_flow.key = rule->cr.flow;
2235 odp_flow.actions = NULL;
2236 odp_flow.n_actions = 0;
2238 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2239 update_stats(p, rule, &odp_flow.stats);
2241 rule->installed = false;
2243 rule_post_uninstall(p, rule);
2248 is_controller_rule(struct rule *rule)
2250 /* If the only action is send to the controller then don't report
2251 * NetFlow expiration messages since it is just part of the control
2252 * logic for the network and not real traffic. */
2256 && rule->super->n_actions == 1
2257 && action_outputs_to_port(&rule->super->actions[0],
2258 htons(OFPP_CONTROLLER)));
2262 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2264 struct rule *super = rule->super;
2266 rule_account(ofproto, rule, 0);
2268 if (ofproto->netflow && !is_controller_rule(rule)) {
2269 struct ofexpired expired;
2270 expired.flow = rule->cr.flow;
2271 expired.packet_count = rule->packet_count;
2272 expired.byte_count = rule->byte_count;
2273 expired.used = rule->used;
2274 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2277 super->packet_count += rule->packet_count;
2278 super->byte_count += rule->byte_count;
2280 /* Reset counters to prevent double counting if the rule ever gets
2282 rule->packet_count = 0;
2283 rule->byte_count = 0;
2284 rule->accounted_bytes = 0;
2286 netflow_flow_clear(&rule->nf_flow);
2291 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2292 struct rconn_packet_counter *counter)
2294 update_openflow_length(msg);
2295 if (rconn_send(ofconn->rconn, msg, counter)) {
2301 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2302 int error, const void *data, size_t len)
2305 struct ofp_error_msg *oem;
2307 if (!(error >> 16)) {
2308 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2313 COVERAGE_INC(ofproto_error);
2314 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2315 oh ? oh->xid : 0, &buf);
2316 oem->type = htons((unsigned int) error >> 16);
2317 oem->code = htons(error & 0xffff);
2318 memcpy(oem->data, data, len);
2319 queue_tx(buf, ofconn, ofconn->reply_counter);
2323 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2326 size_t oh_length = ntohs(oh->length);
2327 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2331 hton_ofp_phy_port(struct ofp_phy_port *opp)
2333 opp->port_no = htons(opp->port_no);
2334 opp->config = htonl(opp->config);
2335 opp->state = htonl(opp->state);
2336 opp->curr = htonl(opp->curr);
2337 opp->advertised = htonl(opp->advertised);
2338 opp->supported = htonl(opp->supported);
2339 opp->peer = htonl(opp->peer);
2343 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2345 struct ofp_header *rq = oh;
2346 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2351 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2352 struct ofp_header *oh)
2354 struct ofp_switch_features *osf;
2356 struct ofport *port;
2358 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2359 osf->datapath_id = htonll(p->datapath_id);
2360 osf->n_buffers = htonl(pktbuf_capacity());
2362 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2363 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2364 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2365 (1u << OFPAT_SET_VLAN_VID) |
2366 (1u << OFPAT_SET_VLAN_PCP) |
2367 (1u << OFPAT_STRIP_VLAN) |
2368 (1u << OFPAT_SET_DL_SRC) |
2369 (1u << OFPAT_SET_DL_DST) |
2370 (1u << OFPAT_SET_NW_SRC) |
2371 (1u << OFPAT_SET_NW_DST) |
2372 (1u << OFPAT_SET_NW_TOS) |
2373 (1u << OFPAT_SET_TP_SRC) |
2374 (1u << OFPAT_SET_TP_DST) |
2375 (1u << OFPAT_ENQUEUE));
2377 HMAP_FOR_EACH (port, struct ofport, hmap_node, &p->ports) {
2378 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2381 queue_tx(buf, ofconn, ofconn->reply_counter);
2386 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2387 struct ofp_header *oh)
2390 struct ofp_switch_config *osc;
2394 /* Figure out flags. */
2395 dpif_get_drop_frags(p->dpif, &drop_frags);
2396 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2399 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2400 osc->flags = htons(flags);
2401 osc->miss_send_len = htons(ofconn->miss_send_len);
2402 queue_tx(buf, ofconn, ofconn->reply_counter);
2408 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2409 struct ofp_switch_config *osc)
2414 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2418 flags = ntohs(osc->flags);
2420 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2421 switch (flags & OFPC_FRAG_MASK) {
2422 case OFPC_FRAG_NORMAL:
2423 dpif_set_drop_frags(p->dpif, false);
2425 case OFPC_FRAG_DROP:
2426 dpif_set_drop_frags(p->dpif, true);
2429 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2435 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2441 add_output_group_action(struct odp_actions *actions, uint16_t group,
2442 uint16_t *nf_output_iface)
2444 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2446 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2447 *nf_output_iface = NF_OUT_FLOOD;
2452 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2454 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2455 a->controller.arg = max_len;
2458 struct action_xlate_ctx {
2460 flow_t flow; /* Flow to which these actions correspond. */
2461 int recurse; /* Recursion level, via xlate_table_action. */
2462 struct ofproto *ofproto;
2463 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2464 * null pointer if we are revalidating
2465 * without a packet to refer to. */
2468 struct odp_actions *out; /* Datapath actions. */
2469 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2470 bool may_set_up_flow; /* True ordinarily; false if the actions must
2471 * be reassessed for every packet. */
2472 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2475 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2476 * flow translation. */
2477 #define MAX_RESUBMIT_RECURSION 8
2479 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2480 struct action_xlate_ctx *ctx);
2483 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2485 const struct ofport *ofport = get_port(ctx->ofproto, port);
2488 if (ofport->opp.config & OFPPC_NO_FWD) {
2489 /* Forwarding disabled on port. */
2494 * We don't have an ofport record for this port, but it doesn't hurt to
2495 * allow forwarding to it anyhow. Maybe such a port will appear later
2496 * and we're pre-populating the flow table.
2500 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2501 ctx->nf_output_iface = port;
2504 static struct rule *
2505 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2508 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2510 /* The rule we found might not be valid, since we could be in need of
2511 * revalidation. If it is not valid, don't return it. */
2514 && ofproto->need_revalidate
2515 && !revalidate_rule(ofproto, rule)) {
2516 COVERAGE_INC(ofproto_invalidated);
2524 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2526 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2527 uint16_t old_in_port;
2530 /* Look up a flow with 'in_port' as the input port. Then restore the
2531 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2532 * have surprising behavior). */
2533 old_in_port = ctx->flow.in_port;
2534 ctx->flow.in_port = in_port;
2535 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2536 ctx->flow.in_port = old_in_port;
2544 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2548 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2550 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2551 MAX_RESUBMIT_RECURSION);
2556 xlate_output_action__(struct action_xlate_ctx *ctx,
2557 uint16_t port, uint16_t max_len)
2560 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2562 ctx->nf_output_iface = NF_OUT_DROP;
2566 add_output_action(ctx, ctx->flow.in_port);
2569 xlate_table_action(ctx, ctx->flow.in_port);
2572 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2573 ctx->out, ctx->tags,
2574 &ctx->nf_output_iface,
2575 ctx->ofproto->aux)) {
2576 COVERAGE_INC(ofproto_uninstallable);
2577 ctx->may_set_up_flow = false;
2581 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2582 &ctx->nf_output_iface);
2585 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2587 case OFPP_CONTROLLER:
2588 add_controller_action(ctx->out, max_len);
2591 add_output_action(ctx, ODPP_LOCAL);
2594 odp_port = ofp_port_to_odp_port(port);
2595 if (odp_port != ctx->flow.in_port) {
2596 add_output_action(ctx, odp_port);
2601 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2602 ctx->nf_output_iface = NF_OUT_FLOOD;
2603 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2604 ctx->nf_output_iface = prev_nf_output_iface;
2605 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2606 ctx->nf_output_iface != NF_OUT_FLOOD) {
2607 ctx->nf_output_iface = NF_OUT_MULTI;
2612 xlate_output_action(struct action_xlate_ctx *ctx,
2613 const struct ofp_action_output *oao)
2615 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2618 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2619 * optimization, because we're going to add another action that sets the
2620 * priority immediately after, or because there are no actions following the
2623 remove_pop_action(struct action_xlate_ctx *ctx)
2625 size_t n = ctx->out->n_actions;
2626 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2627 ctx->out->n_actions--;
2632 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2633 const struct ofp_action_enqueue *oae)
2635 uint16_t ofp_port, odp_port;
2639 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2642 /* Fall back to ordinary output action. */
2643 xlate_output_action__(ctx, ntohs(oae->port), 0);
2647 /* Figure out ODP output port. */
2648 ofp_port = ntohs(oae->port);
2649 if (ofp_port != OFPP_IN_PORT) {
2650 odp_port = ofp_port_to_odp_port(ofp_port);
2652 odp_port = ctx->flow.in_port;
2655 /* Add ODP actions. */
2656 remove_pop_action(ctx);
2657 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2659 add_output_action(ctx, odp_port);
2660 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2662 /* Update NetFlow output port. */
2663 if (ctx->nf_output_iface == NF_OUT_DROP) {
2664 ctx->nf_output_iface = odp_port;
2665 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2666 ctx->nf_output_iface = NF_OUT_MULTI;
2671 xlate_nicira_action(struct action_xlate_ctx *ctx,
2672 const struct nx_action_header *nah)
2674 const struct nx_action_resubmit *nar;
2675 const struct nx_action_set_tunnel *nast;
2676 union odp_action *oa;
2677 int subtype = ntohs(nah->subtype);
2679 assert(nah->vendor == htonl(NX_VENDOR_ID));
2681 case NXAST_RESUBMIT:
2682 nar = (const struct nx_action_resubmit *) nah;
2683 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2686 case NXAST_SET_TUNNEL:
2687 nast = (const struct nx_action_set_tunnel *) nah;
2688 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2689 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2692 case NXAST_DROP_SPOOFED_ARP:
2693 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2694 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2698 /* If you add a new action here that modifies flow data, don't forget to
2699 * update the flow key in ctx->flow at the same time. */
2702 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2708 do_xlate_actions(const union ofp_action *in, size_t n_in,
2709 struct action_xlate_ctx *ctx)
2711 struct actions_iterator iter;
2712 const union ofp_action *ia;
2713 const struct ofport *port;
2715 port = get_port(ctx->ofproto, ctx->flow.in_port);
2716 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2717 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2718 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2719 /* Drop this flow. */
2723 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2724 uint16_t type = ntohs(ia->type);
2725 union odp_action *oa;
2729 xlate_output_action(ctx, &ia->output);
2732 case OFPAT_SET_VLAN_VID:
2733 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2734 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2737 case OFPAT_SET_VLAN_PCP:
2738 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2739 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2742 case OFPAT_STRIP_VLAN:
2743 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2744 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2745 ctx->flow.dl_vlan_pcp = 0;
2748 case OFPAT_SET_DL_SRC:
2749 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2750 memcpy(oa->dl_addr.dl_addr,
2751 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2752 memcpy(ctx->flow.dl_src,
2753 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2756 case OFPAT_SET_DL_DST:
2757 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2758 memcpy(oa->dl_addr.dl_addr,
2759 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2760 memcpy(ctx->flow.dl_dst,
2761 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2764 case OFPAT_SET_NW_SRC:
2765 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2766 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2769 case OFPAT_SET_NW_DST:
2770 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2771 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2774 case OFPAT_SET_NW_TOS:
2775 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2776 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2779 case OFPAT_SET_TP_SRC:
2780 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2781 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2784 case OFPAT_SET_TP_DST:
2785 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2786 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2790 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2794 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2798 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2805 xlate_actions(const union ofp_action *in, size_t n_in,
2806 const flow_t *flow, struct ofproto *ofproto,
2807 const struct ofpbuf *packet,
2808 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2809 uint16_t *nf_output_iface)
2811 tag_type no_tags = 0;
2812 struct action_xlate_ctx ctx;
2813 COVERAGE_INC(ofproto_ofp2odp);
2814 odp_actions_init(out);
2817 ctx.ofproto = ofproto;
2818 ctx.packet = packet;
2820 ctx.tags = tags ? tags : &no_tags;
2821 ctx.may_set_up_flow = true;
2822 ctx.nf_output_iface = NF_OUT_DROP;
2823 do_xlate_actions(in, n_in, &ctx);
2824 remove_pop_action(&ctx);
2826 /* Check with in-band control to see if we're allowed to set up this
2828 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2829 ctx.may_set_up_flow = false;
2832 if (may_set_up_flow) {
2833 *may_set_up_flow = ctx.may_set_up_flow;
2835 if (nf_output_iface) {
2836 *nf_output_iface = ctx.nf_output_iface;
2838 if (odp_actions_overflow(out)) {
2839 COVERAGE_INC(odp_overflow);
2840 odp_actions_init(out);
2841 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2846 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2847 * error message code (composed with ofp_mkerr()) for the caller to propagate
2848 * upward. Otherwise, returns 0.
2850 * 'oh' is used to make log messages more informative. */
2852 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2854 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2855 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2858 type_name = ofp_message_type_to_string(oh->type);
2859 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2863 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2870 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2871 struct ofp_header *oh)
2873 struct ofp_packet_out *opo;
2874 struct ofpbuf payload, *buffer;
2875 struct odp_actions actions;
2881 error = reject_slave_controller(ofconn, oh);
2886 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2890 opo = (struct ofp_packet_out *) oh;
2892 COVERAGE_INC(ofproto_packet_out);
2893 if (opo->buffer_id != htonl(UINT32_MAX)) {
2894 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2896 if (error || !buffer) {
2904 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2905 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2906 &flow, p, &payload, &actions, NULL, NULL, NULL);
2911 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2913 ofpbuf_delete(buffer);
2919 update_port_config(struct ofproto *p, struct ofport *port,
2920 uint32_t config, uint32_t mask)
2922 mask &= config ^ port->opp.config;
2923 if (mask & OFPPC_PORT_DOWN) {
2924 if (config & OFPPC_PORT_DOWN) {
2925 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2927 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2930 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2931 if (mask & REVALIDATE_BITS) {
2932 COVERAGE_INC(ofproto_costly_flags);
2933 port->opp.config ^= mask & REVALIDATE_BITS;
2934 p->need_revalidate = true;
2936 #undef REVALIDATE_BITS
2937 if (mask & OFPPC_NO_FLOOD) {
2938 port->opp.config ^= OFPPC_NO_FLOOD;
2939 refresh_port_groups(p);
2941 if (mask & OFPPC_NO_PACKET_IN) {
2942 port->opp.config ^= OFPPC_NO_PACKET_IN;
2947 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2948 struct ofp_header *oh)
2950 const struct ofp_port_mod *opm;
2951 struct ofport *port;
2954 error = reject_slave_controller(ofconn, oh);
2958 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2962 opm = (struct ofp_port_mod *) oh;
2964 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2966 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2967 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2968 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2970 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2971 if (opm->advertise) {
2972 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2978 static struct ofpbuf *
2979 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2981 struct ofp_stats_reply *osr;
2984 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2985 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2987 osr->flags = htons(0);
2991 static struct ofpbuf *
2992 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2994 return make_stats_reply(request->header.xid, request->type, body_len);
2998 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
3000 struct ofpbuf *msg = *msgp;
3001 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3002 if (nbytes + msg->size > UINT16_MAX) {
3003 struct ofp_stats_reply *reply = msg->data;
3004 reply->flags = htons(OFPSF_REPLY_MORE);
3005 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3006 queue_tx(msg, ofconn, ofconn->reply_counter);
3008 return ofpbuf_put_uninit(*msgp, nbytes);
3012 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3013 struct ofp_stats_request *request)
3015 struct ofp_desc_stats *ods;
3018 msg = start_stats_reply(request, sizeof *ods);
3019 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3020 memset(ods, 0, sizeof *ods);
3021 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3022 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3023 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3024 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3025 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3026 queue_tx(msg, ofconn, ofconn->reply_counter);
3032 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
3034 struct rule *rule = rule_from_cls_rule(cls_rule);
3035 int *n_subrules = n_subrules_;
3043 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3044 struct ofp_stats_request *request)
3046 struct ofp_table_stats *ots;
3048 struct odp_stats dpstats;
3049 int n_exact, n_subrules, n_wild;
3051 msg = start_stats_reply(request, sizeof *ots * 2);
3053 /* Count rules of various kinds. */
3055 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
3056 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3057 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3060 dpif_get_dp_stats(p->dpif, &dpstats);
3061 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3062 memset(ots, 0, sizeof *ots);
3063 ots->table_id = TABLEID_HASH;
3064 strcpy(ots->name, "hash");
3065 ots->wildcards = htonl(0);
3066 ots->max_entries = htonl(dpstats.max_capacity);
3067 ots->active_count = htonl(n_exact);
3068 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3070 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3072 /* Classifier table. */
3073 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3074 memset(ots, 0, sizeof *ots);
3075 ots->table_id = TABLEID_CLASSIFIER;
3076 strcpy(ots->name, "classifier");
3077 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3079 ots->max_entries = htonl(65536);
3080 ots->active_count = htonl(n_wild);
3081 ots->lookup_count = htonll(0); /* XXX */
3082 ots->matched_count = htonll(0); /* XXX */
3084 queue_tx(msg, ofconn, ofconn->reply_counter);
3089 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3090 struct ofpbuf **msgp)
3092 struct netdev_stats stats;
3093 struct ofp_port_stats *ops;
3095 /* Intentionally ignore return value, since errors will set
3096 * 'stats' to all-1s, which is correct for OpenFlow, and
3097 * netdev_get_stats() will log errors. */
3098 netdev_get_stats(port->netdev, &stats);
3100 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3101 ops->port_no = htons(port->opp.port_no);
3102 memset(ops->pad, 0, sizeof ops->pad);
3103 ops->rx_packets = htonll(stats.rx_packets);
3104 ops->tx_packets = htonll(stats.tx_packets);
3105 ops->rx_bytes = htonll(stats.rx_bytes);
3106 ops->tx_bytes = htonll(stats.tx_bytes);
3107 ops->rx_dropped = htonll(stats.rx_dropped);
3108 ops->tx_dropped = htonll(stats.tx_dropped);
3109 ops->rx_errors = htonll(stats.rx_errors);
3110 ops->tx_errors = htonll(stats.tx_errors);
3111 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3112 ops->rx_over_err = htonll(stats.rx_over_errors);
3113 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3114 ops->collisions = htonll(stats.collisions);
3118 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3119 struct ofp_stats_request *osr,
3122 struct ofp_port_stats_request *psr;
3123 struct ofp_port_stats *ops;
3125 struct ofport *port;
3127 if (arg_size != sizeof *psr) {
3128 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3130 psr = (struct ofp_port_stats_request *) osr->body;
3132 msg = start_stats_reply(osr, sizeof *ops * 16);
3133 if (psr->port_no != htons(OFPP_NONE)) {
3134 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3136 append_port_stat(port, ofconn, &msg);
3139 HMAP_FOR_EACH (port, struct ofport, hmap_node, &p->ports) {
3140 append_port_stat(port, ofconn, &msg);
3144 queue_tx(msg, ofconn, ofconn->reply_counter);
3148 struct flow_stats_cbdata {
3149 struct ofproto *ofproto;
3150 struct ofconn *ofconn;
3155 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3156 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3157 * returned statistic include statistics for all of 'rule''s subrules. */
3159 query_stats(struct ofproto *p, struct rule *rule,
3160 uint64_t *packet_countp, uint64_t *byte_countp)
3162 uint64_t packet_count, byte_count;
3163 struct rule *subrule;
3164 struct odp_flow *odp_flows;
3167 /* Start from historical data for 'rule' itself that are no longer tracked
3168 * by the datapath. This counts, for example, subrules that have
3170 packet_count = rule->packet_count;
3171 byte_count = rule->byte_count;
3173 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3174 * wildcarded then on all of its subrules.
3176 * Also, add any statistics that are not tracked by the datapath for each
3177 * subrule. This includes, for example, statistics for packets that were
3178 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3180 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3181 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3182 if (rule->cr.wc.wildcards) {
3184 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3185 odp_flows[i++].key = subrule->cr.flow;
3186 packet_count += subrule->packet_count;
3187 byte_count += subrule->byte_count;
3190 odp_flows[0].key = rule->cr.flow;
3193 /* Fetch up-to-date statistics from the datapath and add them in. */
3194 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3196 for (i = 0; i < n_odp_flows; i++) {
3197 struct odp_flow *odp_flow = &odp_flows[i];
3198 packet_count += odp_flow->stats.n_packets;
3199 byte_count += odp_flow->stats.n_bytes;
3204 /* Return the stats to the caller. */
3205 *packet_countp = packet_count;
3206 *byte_countp = byte_count;
3210 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3212 struct rule *rule = rule_from_cls_rule(rule_);
3213 struct flow_stats_cbdata *cbdata = cbdata_;
3214 struct ofp_flow_stats *ofs;
3215 uint64_t packet_count, byte_count;
3216 size_t act_len, len;
3217 long long int tdiff = time_msec() - rule->created;
3218 uint32_t sec = tdiff / 1000;
3219 uint32_t msec = tdiff - (sec * 1000);
3221 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3225 act_len = sizeof *rule->actions * rule->n_actions;
3226 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3228 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3230 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3231 ofs->length = htons(len);
3232 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3234 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3235 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3236 ofs->duration_sec = htonl(sec);
3237 ofs->duration_nsec = htonl(msec * 1000000);
3238 ofs->cookie = rule->flow_cookie;
3239 ofs->priority = htons(rule->cr.priority);
3240 ofs->idle_timeout = htons(rule->idle_timeout);
3241 ofs->hard_timeout = htons(rule->hard_timeout);
3242 memset(ofs->pad2, 0, sizeof ofs->pad2);
3243 ofs->packet_count = htonll(packet_count);
3244 ofs->byte_count = htonll(byte_count);
3245 memcpy(ofs->actions, rule->actions, act_len);
3249 table_id_to_include(uint8_t table_id)
3251 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3252 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3253 : table_id == 0xff ? CLS_INC_ALL
3258 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3259 const struct ofp_stats_request *osr,
3262 struct ofp_flow_stats_request *fsr;
3263 struct flow_stats_cbdata cbdata;
3264 struct cls_rule target;
3266 if (arg_size != sizeof *fsr) {
3267 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3269 fsr = (struct ofp_flow_stats_request *) osr->body;
3271 COVERAGE_INC(ofproto_flows_req);
3273 cbdata.ofconn = ofconn;
3274 cbdata.out_port = fsr->out_port;
3275 cbdata.msg = start_stats_reply(osr, 1024);
3276 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3277 classifier_for_each_match(&p->cls, &target,
3278 table_id_to_include(fsr->table_id),
3279 flow_stats_cb, &cbdata);
3280 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3284 struct flow_stats_ds_cbdata {
3285 struct ofproto *ofproto;
3290 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3292 struct rule *rule = rule_from_cls_rule(rule_);
3293 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3294 struct ds *results = cbdata->results;
3295 struct ofp_match match;
3296 uint64_t packet_count, byte_count;
3297 size_t act_len = sizeof *rule->actions * rule->n_actions;
3299 /* Don't report on subrules. */
3300 if (rule->super != NULL) {
3304 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3305 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3306 cbdata->ofproto->tun_id_from_cookie, &match);
3308 ds_put_format(results, "duration=%llds, ",
3309 (time_msec() - rule->created) / 1000);
3310 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3311 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3312 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3313 ofp_print_match(results, &match, true);
3314 ofp_print_actions(results, &rule->actions->header, act_len);
3315 ds_put_cstr(results, "\n");
3318 /* Adds a pretty-printed description of all flows to 'results', including
3319 * those marked hidden by secchan (e.g., by in-band control). */
3321 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3323 struct ofp_match match;
3324 struct cls_rule target;
3325 struct flow_stats_ds_cbdata cbdata;
3327 memset(&match, 0, sizeof match);
3328 match.wildcards = htonl(OVSFW_ALL);
3331 cbdata.results = results;
3333 cls_rule_from_match(&match, 0, false, 0, &target);
3334 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3335 flow_stats_ds_cb, &cbdata);
3338 struct aggregate_stats_cbdata {
3339 struct ofproto *ofproto;
3341 uint64_t packet_count;
3342 uint64_t byte_count;
3347 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3349 struct rule *rule = rule_from_cls_rule(rule_);
3350 struct aggregate_stats_cbdata *cbdata = cbdata_;
3351 uint64_t packet_count, byte_count;
3353 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3357 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3359 cbdata->packet_count += packet_count;
3360 cbdata->byte_count += byte_count;
3365 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3366 const struct ofp_stats_request *osr,
3369 struct ofp_aggregate_stats_request *asr;
3370 struct ofp_aggregate_stats_reply *reply;
3371 struct aggregate_stats_cbdata cbdata;
3372 struct cls_rule target;
3375 if (arg_size != sizeof *asr) {
3376 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3378 asr = (struct ofp_aggregate_stats_request *) osr->body;
3380 COVERAGE_INC(ofproto_agg_request);
3382 cbdata.out_port = asr->out_port;
3383 cbdata.packet_count = 0;
3384 cbdata.byte_count = 0;
3386 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3387 classifier_for_each_match(&p->cls, &target,
3388 table_id_to_include(asr->table_id),
3389 aggregate_stats_cb, &cbdata);
3391 msg = start_stats_reply(osr, sizeof *reply);
3392 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3393 reply->flow_count = htonl(cbdata.n_flows);
3394 reply->packet_count = htonll(cbdata.packet_count);
3395 reply->byte_count = htonll(cbdata.byte_count);
3396 queue_tx(msg, ofconn, ofconn->reply_counter);
3400 struct queue_stats_cbdata {
3401 struct ofconn *ofconn;
3402 struct ofport *ofport;
3407 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3408 const struct netdev_queue_stats *stats)
3410 struct ofp_queue_stats *reply;
3412 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3413 reply->port_no = htons(cbdata->ofport->opp.port_no);
3414 memset(reply->pad, 0, sizeof reply->pad);
3415 reply->queue_id = htonl(queue_id);
3416 reply->tx_bytes = htonll(stats->tx_bytes);
3417 reply->tx_packets = htonll(stats->tx_packets);
3418 reply->tx_errors = htonll(stats->tx_errors);
3422 handle_queue_stats_dump_cb(uint32_t queue_id,
3423 struct netdev_queue_stats *stats,
3426 struct queue_stats_cbdata *cbdata = cbdata_;
3428 put_queue_stats(cbdata, queue_id, stats);
3432 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3433 struct queue_stats_cbdata *cbdata)
3435 cbdata->ofport = port;
3436 if (queue_id == OFPQ_ALL) {
3437 netdev_dump_queue_stats(port->netdev,
3438 handle_queue_stats_dump_cb, cbdata);
3440 struct netdev_queue_stats stats;
3442 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3443 put_queue_stats(cbdata, queue_id, &stats);
3448 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3449 const struct ofp_stats_request *osr,
3452 struct ofp_queue_stats_request *qsr;
3453 struct queue_stats_cbdata cbdata;
3454 struct ofport *port;
3455 unsigned int port_no;
3458 if (arg_size != sizeof *qsr) {
3459 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3461 qsr = (struct ofp_queue_stats_request *) osr->body;
3463 COVERAGE_INC(ofproto_queue_req);
3465 cbdata.ofconn = ofconn;
3466 cbdata.msg = start_stats_reply(osr, 128);
3468 port_no = ntohs(qsr->port_no);
3469 queue_id = ntohl(qsr->queue_id);
3470 if (port_no == OFPP_ALL) {
3471 HMAP_FOR_EACH (port, struct ofport, hmap_node, &ofproto->ports) {
3472 handle_queue_stats_for_port(port, queue_id, &cbdata);
3474 } else if (port_no < ofproto->max_ports) {
3475 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3477 handle_queue_stats_for_port(port, queue_id, &cbdata);
3480 ofpbuf_delete(cbdata.msg);
3481 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3483 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3489 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3490 struct ofp_header *oh)
3492 struct ofp_stats_request *osr;
3496 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3501 osr = (struct ofp_stats_request *) oh;
3503 switch (ntohs(osr->type)) {
3505 return handle_desc_stats_request(p, ofconn, osr);
3508 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3510 case OFPST_AGGREGATE:
3511 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3514 return handle_table_stats_request(p, ofconn, osr);
3517 return handle_port_stats_request(p, ofconn, osr, arg_size);
3520 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3523 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3526 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3530 static long long int
3531 msec_from_nsec(uint64_t sec, uint32_t nsec)
3533 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3537 update_time(struct ofproto *ofproto, struct rule *rule,
3538 const struct odp_flow_stats *stats)
3540 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3541 if (used > rule->used) {
3543 if (rule->super && used > rule->super->used) {
3544 rule->super->used = used;
3546 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3551 update_stats(struct ofproto *ofproto, struct rule *rule,
3552 const struct odp_flow_stats *stats)
3554 if (stats->n_packets) {
3555 update_time(ofproto, rule, stats);
3556 rule->packet_count += stats->n_packets;
3557 rule->byte_count += stats->n_bytes;
3558 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3562 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3563 * in which no matching flow already exists in the flow table.
3565 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3566 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3567 * code as encoded by ofp_mkerr() on failure.
3569 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3572 add_flow(struct ofproto *p, struct ofconn *ofconn,
3573 const struct ofp_flow_mod *ofm, size_t n_actions)
3575 struct ofpbuf *packet;
3580 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3584 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3586 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3587 ntohs(ofm->priority))) {
3588 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3592 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3593 n_actions, ntohs(ofm->idle_timeout),
3594 ntohs(ofm->hard_timeout), ofm->cookie,
3595 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3596 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3597 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3600 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3601 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3605 in_port = UINT16_MAX;
3608 rule_insert(p, rule, packet, in_port);
3612 static struct rule *
3613 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3618 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3620 return rule_from_cls_rule(classifier_find_rule_exactly(
3621 &p->cls, &flow, wildcards,
3622 ntohs(ofm->priority)));
3626 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3627 struct rule *rule, const struct ofp_flow_mod *ofm)
3629 struct ofpbuf *packet;
3634 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3638 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3644 flow_extract(packet, 0, in_port, &flow);
3645 rule_execute(ofproto, rule, packet, &flow);
3650 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3652 struct modify_flows_cbdata {
3653 struct ofproto *ofproto;
3654 const struct ofp_flow_mod *ofm;
3659 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3660 size_t n_actions, struct rule *);
3661 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3663 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3664 * encoded by ofp_mkerr() on failure.
3666 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3669 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3670 const struct ofp_flow_mod *ofm, size_t n_actions)
3672 struct modify_flows_cbdata cbdata;
3673 struct cls_rule target;
3677 cbdata.n_actions = n_actions;
3678 cbdata.match = NULL;
3680 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3683 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3684 modify_flows_cb, &cbdata);
3686 /* This credits the packet to whichever flow happened to happened to
3687 * match last. That's weird. Maybe we should do a lookup for the
3688 * flow that actually matches the packet? Who knows. */
3689 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3692 return add_flow(p, ofconn, ofm, n_actions);
3696 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3697 * code as encoded by ofp_mkerr() on failure.
3699 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3702 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3703 struct ofp_flow_mod *ofm, size_t n_actions)
3705 struct rule *rule = find_flow_strict(p, ofm);
3706 if (rule && !rule_is_hidden(rule)) {
3707 modify_flow(p, ofm, n_actions, rule);
3708 return send_buffered_packet(p, ofconn, rule, ofm);
3710 return add_flow(p, ofconn, ofm, n_actions);
3714 /* Callback for modify_flows_loose(). */
3716 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3718 struct rule *rule = rule_from_cls_rule(rule_);
3719 struct modify_flows_cbdata *cbdata = cbdata_;
3721 if (!rule_is_hidden(rule)) {
3722 cbdata->match = rule;
3723 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3727 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3728 * been identified as a flow in 'p''s flow table to be modified, by changing
3729 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3730 * ofp_action[] structures). */
3732 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3733 size_t n_actions, struct rule *rule)
3735 size_t actions_len = n_actions * sizeof *rule->actions;
3737 rule->flow_cookie = ofm->cookie;
3739 /* If the actions are the same, do nothing. */
3740 if (n_actions == rule->n_actions
3741 && !memcmp(ofm->actions, rule->actions, actions_len))
3746 /* Replace actions. */
3747 free(rule->actions);
3748 rule->actions = xmemdup(ofm->actions, actions_len);
3749 rule->n_actions = n_actions;
3751 /* Make sure that the datapath gets updated properly. */
3752 if (rule->cr.wc.wildcards) {
3753 COVERAGE_INC(ofproto_mod_wc_flow);
3754 p->need_revalidate = true;
3756 rule_update_actions(p, rule);
3762 /* OFPFC_DELETE implementation. */
3764 struct delete_flows_cbdata {
3765 struct ofproto *ofproto;
3769 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3770 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3772 /* Implements OFPFC_DELETE. */
3774 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3776 struct delete_flows_cbdata cbdata;
3777 struct cls_rule target;
3780 cbdata.out_port = ofm->out_port;
3782 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3785 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3786 delete_flows_cb, &cbdata);
3789 /* Implements OFPFC_DELETE_STRICT. */
3791 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3793 struct rule *rule = find_flow_strict(p, ofm);
3795 delete_flow(p, rule, ofm->out_port);
3799 /* Callback for delete_flows_loose(). */
3801 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3803 struct rule *rule = rule_from_cls_rule(rule_);
3804 struct delete_flows_cbdata *cbdata = cbdata_;
3806 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3809 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3810 * been identified as a flow to delete from 'p''s flow table, by deleting the
3811 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3814 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3815 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3816 * specified 'out_port'. */
3818 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3820 if (rule_is_hidden(rule)) {
3824 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3828 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3829 rule_remove(p, rule);
3833 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3834 struct ofp_flow_mod *ofm)
3836 struct ofp_match orig_match;
3840 error = reject_slave_controller(ofconn, &ofm->header);
3844 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3845 sizeof *ofm->actions, &n_actions);
3850 /* We do not support the emergency flow cache. It will hopefully
3851 * get dropped from OpenFlow in the near future. */
3852 if (ofm->flags & htons(OFPFF_EMERG)) {
3853 /* There isn't a good fit for an error code, so just state that the
3854 * flow table is full. */
3855 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3858 /* Normalize ofp->match. If normalization actually changes anything, then
3859 * log the differences. */
3860 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3861 orig_match = ofm->match;
3862 normalize_match(&ofm->match);
3863 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3864 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3865 if (!VLOG_DROP_INFO(&normal_rl)) {
3866 char *old = ofp_match_to_literal_string(&orig_match);
3867 char *new = ofp_match_to_literal_string(&ofm->match);
3868 VLOG_INFO("%s: normalization changed ofp_match, details:",
3869 rconn_get_name(ofconn->rconn));
3870 VLOG_INFO(" pre: %s", old);
3871 VLOG_INFO("post: %s", new);
3877 if (!ofm->match.wildcards) {
3878 ofm->priority = htons(UINT16_MAX);
3881 error = validate_actions((const union ofp_action *) ofm->actions,
3882 n_actions, p->max_ports);
3887 switch (ntohs(ofm->command)) {
3889 return add_flow(p, ofconn, ofm, n_actions);
3892 return modify_flows_loose(p, ofconn, ofm, n_actions);
3894 case OFPFC_MODIFY_STRICT:
3895 return modify_flow_strict(p, ofconn, ofm, n_actions);
3898 delete_flows_loose(p, ofm);
3901 case OFPFC_DELETE_STRICT:
3902 delete_flow_strict(p, ofm);
3906 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3911 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3915 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3920 p->tun_id_from_cookie = !!msg->set;
3925 handle_role_request(struct ofproto *ofproto,
3926 struct ofconn *ofconn, struct nicira_header *msg)
3928 struct nx_role_request *nrr;
3929 struct nx_role_request *reply;
3933 if (ntohs(msg->header.length) != sizeof *nrr) {
3934 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3935 ntohs(msg->header.length), sizeof *nrr);
3936 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3938 nrr = (struct nx_role_request *) msg;
3940 if (ofconn->type != OFCONN_PRIMARY) {
3941 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3943 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3946 role = ntohl(nrr->role);
3947 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3948 && role != NX_ROLE_SLAVE) {
3949 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3951 /* There's no good error code for this. */
3952 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3955 if (role == NX_ROLE_MASTER) {
3956 struct ofconn *other;
3958 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3959 &ofproto->controllers) {
3960 if (other->role == NX_ROLE_MASTER) {
3961 other->role = NX_ROLE_SLAVE;
3965 ofconn->role = role;
3967 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3969 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3970 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3971 reply->role = htonl(role);
3972 queue_tx(buf, ofconn, ofconn->reply_counter);
3978 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3980 struct ofp_vendor_header *ovh = msg;
3981 struct nicira_header *nh;
3983 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3984 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3985 "(expected at least %zu)",
3986 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3987 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3989 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3990 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3992 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3993 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3994 "(expected at least %zu)",
3995 ntohs(ovh->header.length), sizeof(struct nicira_header));
3996 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4000 switch (ntohl(nh->subtype)) {
4001 case NXT_STATUS_REQUEST:
4002 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4005 case NXT_TUN_ID_FROM_COOKIE:
4006 return handle_tun_id_from_cookie(p, msg);
4008 case NXT_ROLE_REQUEST:
4009 return handle_role_request(p, ofconn, msg);
4012 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4016 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4018 struct ofp_header *ob;
4021 /* Currently, everything executes synchronously, so we can just
4022 * immediately send the barrier reply. */
4023 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4024 queue_tx(buf, ofconn, ofconn->reply_counter);
4029 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4030 struct ofpbuf *ofp_msg)
4032 struct ofp_header *oh = ofp_msg->data;
4035 COVERAGE_INC(ofproto_recv_openflow);
4037 case OFPT_ECHO_REQUEST:
4038 error = handle_echo_request(ofconn, oh);
4041 case OFPT_ECHO_REPLY:
4045 case OFPT_FEATURES_REQUEST:
4046 error = handle_features_request(p, ofconn, oh);
4049 case OFPT_GET_CONFIG_REQUEST:
4050 error = handle_get_config_request(p, ofconn, oh);
4053 case OFPT_SET_CONFIG:
4054 error = handle_set_config(p, ofconn, ofp_msg->data);
4057 case OFPT_PACKET_OUT:
4058 error = handle_packet_out(p, ofconn, ofp_msg->data);
4062 error = handle_port_mod(p, ofconn, oh);
4066 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4069 case OFPT_STATS_REQUEST:
4070 error = handle_stats_request(p, ofconn, oh);
4074 error = handle_vendor(p, ofconn, ofp_msg->data);
4077 case OFPT_BARRIER_REQUEST:
4078 error = handle_barrier_request(ofconn, oh);
4082 if (VLOG_IS_WARN_ENABLED()) {
4083 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4084 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4087 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4092 send_error_oh(ofconn, ofp_msg->data, error);
4097 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4099 struct odp_msg *msg = packet->data;
4101 struct ofpbuf payload;
4104 payload.data = msg + 1;
4105 payload.size = msg->length - sizeof *msg;
4106 flow_extract(&payload, msg->arg, msg->port, &flow);
4108 /* Check with in-band control to see if this packet should be sent
4109 * to the local port regardless of the flow table. */
4110 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4111 union odp_action action;
4113 memset(&action, 0, sizeof(action));
4114 action.output.type = ODPAT_OUTPUT;
4115 action.output.port = ODPP_LOCAL;
4116 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
4119 rule = lookup_valid_rule(p, &flow);
4121 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4122 struct ofport *port = get_port(p, msg->port);
4124 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4125 COVERAGE_INC(ofproto_no_packet_in);
4126 /* XXX install 'drop' flow entry */
4127 ofpbuf_delete(packet);
4131 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4134 COVERAGE_INC(ofproto_packet_in);
4135 send_packet_in(p, packet);
4139 if (rule->cr.wc.wildcards) {
4140 rule = rule_create_subrule(p, rule, &flow);
4141 rule_make_actions(p, rule, packet);
4143 if (!rule->may_install) {
4144 /* The rule is not installable, that is, we need to process every
4145 * packet, so process the current packet and set its actions into
4147 rule_make_actions(p, rule, packet);
4149 /* XXX revalidate rule if it needs it */
4153 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4155 * Extra-special case for fail-open mode.
4157 * We are in fail-open mode and the packet matched the fail-open rule,
4158 * but we are connected to a controller too. We should send the packet
4159 * up to the controller in the hope that it will try to set up a flow
4160 * and thereby allow us to exit fail-open.
4162 * See the top-level comment in fail-open.c for more information.
4164 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4165 DPIF_RECV_MSG_PADDING));
4168 ofpbuf_pull(packet, sizeof *msg);
4169 rule_execute(p, rule, packet, &flow);
4170 rule_reinstall(p, rule);
4174 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4176 struct odp_msg *msg = packet->data;
4178 switch (msg->type) {
4179 case _ODPL_ACTION_NR:
4180 COVERAGE_INC(ofproto_ctlr_action);
4181 send_packet_in(p, packet);
4184 case _ODPL_SFLOW_NR:
4186 ofproto_sflow_received(p->sflow, msg);
4188 ofpbuf_delete(packet);
4192 handle_odp_miss_msg(p, packet);
4196 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4203 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4205 struct rule *sub = rule_from_cls_rule(sub_);
4206 struct revalidate_cbdata *cbdata = cbdata_;
4208 if (cbdata->revalidate_all
4209 || (cbdata->revalidate_subrules && sub->super)
4210 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4211 revalidate_rule(cbdata->ofproto, sub);
4216 revalidate_rule(struct ofproto *p, struct rule *rule)
4218 const flow_t *flow = &rule->cr.flow;
4220 COVERAGE_INC(ofproto_revalidate_rule);
4223 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4225 rule_remove(p, rule);
4227 } else if (super != rule->super) {
4228 COVERAGE_INC(ofproto_revalidate_moved);
4229 list_remove(&rule->list);
4230 list_push_back(&super->list, &rule->list);
4231 rule->super = super;
4232 rule->hard_timeout = super->hard_timeout;
4233 rule->idle_timeout = super->idle_timeout;
4234 rule->created = super->created;
4239 rule_update_actions(p, rule);
4243 static struct ofpbuf *
4244 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4245 long long int now, uint8_t reason)
4247 struct ofp_flow_removed *ofr;
4249 long long int tdiff = now - rule->created;
4250 uint32_t sec = tdiff / 1000;
4251 uint32_t msec = tdiff - (sec * 1000);
4253 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4254 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4256 ofr->cookie = rule->flow_cookie;
4257 ofr->priority = htons(rule->cr.priority);
4258 ofr->reason = reason;
4259 ofr->duration_sec = htonl(sec);
4260 ofr->duration_nsec = htonl(msec * 1000000);
4261 ofr->idle_timeout = htons(rule->idle_timeout);
4262 ofr->packet_count = htonll(rule->packet_count);
4263 ofr->byte_count = htonll(rule->byte_count);
4269 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4271 assert(rule->installed);
4272 assert(!rule->cr.wc.wildcards);
4275 rule_remove(ofproto, rule);
4277 rule_uninstall(ofproto, rule);
4282 send_flow_removed(struct ofproto *p, struct rule *rule,
4283 long long int now, uint8_t reason)
4285 struct ofconn *ofconn;
4286 struct ofconn *prev;
4287 struct ofpbuf *buf = NULL;
4289 /* We limit the maximum number of queued flow expirations it by accounting
4290 * them under the counter for replies. That works because preventing
4291 * OpenFlow requests from being processed also prevents new flows from
4292 * being added (and expiring). (It also prevents processing OpenFlow
4293 * requests that would not add new flows, so it is imperfect.) */
4296 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4297 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4298 && ofconn_receives_async_msgs(ofconn)) {
4300 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4302 buf = compose_flow_removed(p, rule, now, reason);
4308 queue_tx(buf, prev, prev->reply_counter);
4314 expire_rule(struct cls_rule *cls_rule, void *p_)
4316 struct ofproto *p = p_;
4317 struct rule *rule = rule_from_cls_rule(cls_rule);
4318 long long int hard_expire, idle_expire, expire, now;
4320 hard_expire = (rule->hard_timeout
4321 ? rule->created + rule->hard_timeout * 1000
4323 idle_expire = (rule->idle_timeout
4324 && (rule->super || list_is_empty(&rule->list))
4325 ? rule->used + rule->idle_timeout * 1000
4327 expire = MIN(hard_expire, idle_expire);
4331 if (rule->installed && now >= rule->used + 5000) {
4332 uninstall_idle_flow(p, rule);
4333 } else if (!rule->cr.wc.wildcards) {
4334 active_timeout(p, rule);
4340 COVERAGE_INC(ofproto_expired);
4342 /* Update stats. This code will be a no-op if the rule expired
4343 * due to an idle timeout. */
4344 if (rule->cr.wc.wildcards) {
4345 struct rule *subrule, *next;
4346 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4347 rule_remove(p, subrule);
4350 rule_uninstall(p, rule);
4353 if (!rule_is_hidden(rule)) {
4354 send_flow_removed(p, rule, now,
4356 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4358 rule_remove(p, rule);
4362 active_timeout(struct ofproto *ofproto, struct rule *rule)
4364 if (ofproto->netflow && !is_controller_rule(rule) &&
4365 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4366 struct ofexpired expired;
4367 struct odp_flow odp_flow;
4369 /* Get updated flow stats. */
4370 memset(&odp_flow, 0, sizeof odp_flow);
4371 if (rule->installed) {
4372 odp_flow.key = rule->cr.flow;
4373 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4374 dpif_flow_get(ofproto->dpif, &odp_flow);
4376 if (odp_flow.stats.n_packets) {
4377 update_time(ofproto, rule, &odp_flow.stats);
4378 netflow_flow_update_flags(&rule->nf_flow,
4379 odp_flow.stats.tcp_flags);
4383 expired.flow = rule->cr.flow;
4384 expired.packet_count = rule->packet_count +
4385 odp_flow.stats.n_packets;
4386 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4387 expired.used = rule->used;
4389 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4391 /* Schedule us to send the accumulated records once we have
4392 * collected all of them. */
4393 poll_immediate_wake();
4398 update_used(struct ofproto *p)
4400 struct odp_flow *flows;
4405 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4410 for (i = 0; i < n_flows; i++) {
4411 struct odp_flow *f = &flows[i];
4414 rule = rule_from_cls_rule(
4415 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4416 if (!rule || !rule->installed) {
4417 COVERAGE_INC(ofproto_unexpected_rule);
4418 dpif_flow_del(p->dpif, f);
4422 update_time(p, rule, &f->stats);
4423 rule_account(p, rule, f->stats.n_bytes);
4428 /* pinsched callback for sending 'packet' on 'ofconn'. */
4430 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4432 struct ofconn *ofconn = ofconn_;
4434 rconn_send_with_limit(ofconn->rconn, packet,
4435 ofconn->packet_in_counter, 100);
4438 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4439 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4440 * packet scheduler for sending.
4442 * 'max_len' specifies the maximum number of bytes of the packet to send on
4443 * 'ofconn' (INT_MAX specifies no limit).
4445 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4446 * ownership is transferred to this function. */
4448 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4451 struct ofproto *ofproto = ofconn->ofproto;
4452 struct ofp_packet_in *opi = packet->data;
4453 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4454 int send_len, trim_size;
4458 if (opi->reason == OFPR_ACTION) {
4459 buffer_id = UINT32_MAX;
4460 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4461 buffer_id = pktbuf_get_null();
4462 } else if (!ofconn->pktbuf) {
4463 buffer_id = UINT32_MAX;
4465 struct ofpbuf payload;
4466 payload.data = opi->data;
4467 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4468 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4471 /* Figure out how much of the packet to send. */
4472 send_len = ntohs(opi->total_len);
4473 if (buffer_id != UINT32_MAX) {
4474 send_len = MIN(send_len, ofconn->miss_send_len);
4476 send_len = MIN(send_len, max_len);
4478 /* Adjust packet length and clone if necessary. */
4479 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4481 packet = ofpbuf_clone_data(packet->data, trim_size);
4484 packet->size = trim_size;
4487 /* Update packet headers. */
4488 opi->buffer_id = htonl(buffer_id);
4489 update_openflow_length(packet);
4491 /* Hand over to packet scheduler. It might immediately call into
4492 * do_send_packet_in() or it might buffer it for a while (until a later
4493 * call to pinsched_run()). */
4494 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4495 packet, do_send_packet_in, ofconn);
4498 /* Replace struct odp_msg header in 'packet' by equivalent struct
4499 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4500 * returned by dpif_recv()).
4502 * The conversion is not complete: the caller still needs to trim any unneeded
4503 * payload off the end of the buffer, set the length in the OpenFlow header,
4504 * and set buffer_id. Those require us to know the controller settings and so
4505 * must be done on a per-controller basis.
4507 * Returns the maximum number of bytes of the packet that should be sent to
4508 * the controller (INT_MAX if no limit). */
4510 do_convert_to_packet_in(struct ofpbuf *packet)
4512 struct odp_msg *msg = packet->data;
4513 struct ofp_packet_in *opi;
4519 /* Extract relevant header fields */
4520 if (msg->type == _ODPL_ACTION_NR) {
4521 reason = OFPR_ACTION;
4524 reason = OFPR_NO_MATCH;
4527 total_len = msg->length - sizeof *msg;
4528 in_port = odp_port_to_ofp_port(msg->port);
4530 /* Repurpose packet buffer by overwriting header. */
4531 ofpbuf_pull(packet, sizeof(struct odp_msg));
4532 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4533 opi->header.version = OFP_VERSION;
4534 opi->header.type = OFPT_PACKET_IN;
4535 opi->total_len = htons(total_len);
4536 opi->in_port = htons(in_port);
4537 opi->reason = reason;
4542 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4543 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4544 * as necessary according to their individual configurations.
4546 * 'packet' must have sufficient headroom to convert it into a struct
4547 * ofp_packet_in (e.g. as returned by dpif_recv()).
4549 * Takes ownership of 'packet'. */
4551 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4553 struct ofconn *ofconn, *prev;
4556 max_len = do_convert_to_packet_in(packet);
4559 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4560 if (ofconn_receives_async_msgs(ofconn)) {
4562 schedule_packet_in(prev, packet, max_len, true);
4568 schedule_packet_in(prev, packet, max_len, false);
4570 ofpbuf_delete(packet);
4575 pick_datapath_id(const struct ofproto *ofproto)
4577 const struct ofport *port;
4579 port = get_port(ofproto, ODPP_LOCAL);
4581 uint8_t ea[ETH_ADDR_LEN];
4584 error = netdev_get_etheraddr(port->netdev, ea);
4586 return eth_addr_to_uint64(ea);
4588 VLOG_WARN("could not get MAC address for %s (%s)",
4589 netdev_get_name(port->netdev), strerror(error));
4591 return ofproto->fallback_dpid;
4595 pick_fallback_dpid(void)
4597 uint8_t ea[ETH_ADDR_LEN];
4598 eth_addr_nicira_random(ea);
4599 return eth_addr_to_uint64(ea);
4603 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4604 struct odp_actions *actions, tag_type *tags,
4605 uint16_t *nf_output_iface, void *ofproto_)
4607 struct ofproto *ofproto = ofproto_;
4610 /* Drop frames for reserved multicast addresses. */
4611 if (eth_addr_is_reserved(flow->dl_dst)) {
4615 /* Learn source MAC (but don't try to learn from revalidation). */
4616 if (packet != NULL) {
4617 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4619 GRAT_ARP_LOCK_NONE);
4621 /* The log messages here could actually be useful in debugging,
4622 * so keep the rate limit relatively high. */
4623 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4624 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4625 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4626 ofproto_revalidate(ofproto, rev_tag);
4630 /* Determine output port. */
4631 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4634 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4635 } else if (out_port != flow->in_port) {
4636 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4637 *nf_output_iface = out_port;
4645 static const struct ofhooks default_ofhooks = {
4647 default_normal_ofhook_cb,