2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
28 #include "classifier.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
43 #include "ofp-print.h"
45 #include "ofproto-sflow.h"
47 #include "openflow/nicira-ext.h"
48 #include "openflow/openflow.h"
49 #include "openvswitch/datapath-protocol.h"
53 #include "poll-loop.h"
56 #include "stream-ssl.h"
60 #include "unaligned.h"
65 VLOG_DEFINE_THIS_MODULE(ofproto);
67 COVERAGE_DEFINE(facet_changed_rule);
68 COVERAGE_DEFINE(facet_revalidate);
69 COVERAGE_DEFINE(odp_overflow);
70 COVERAGE_DEFINE(ofproto_agg_request);
71 COVERAGE_DEFINE(ofproto_costly_flags);
72 COVERAGE_DEFINE(ofproto_ctlr_action);
73 COVERAGE_DEFINE(ofproto_del_rule);
74 COVERAGE_DEFINE(ofproto_error);
75 COVERAGE_DEFINE(ofproto_expiration);
76 COVERAGE_DEFINE(ofproto_expired);
77 COVERAGE_DEFINE(ofproto_flows_req);
78 COVERAGE_DEFINE(ofproto_flush);
79 COVERAGE_DEFINE(ofproto_invalidated);
80 COVERAGE_DEFINE(ofproto_no_packet_in);
81 COVERAGE_DEFINE(ofproto_ofconn_stuck);
82 COVERAGE_DEFINE(ofproto_ofp2odp);
83 COVERAGE_DEFINE(ofproto_packet_in);
84 COVERAGE_DEFINE(ofproto_packet_out);
85 COVERAGE_DEFINE(ofproto_queue_req);
86 COVERAGE_DEFINE(ofproto_recv_openflow);
87 COVERAGE_DEFINE(ofproto_reinit_ports);
88 COVERAGE_DEFINE(ofproto_unexpected_rule);
89 COVERAGE_DEFINE(ofproto_uninstallable);
90 COVERAGE_DEFINE(ofproto_update_port);
92 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
93 * flow translation. */
94 #define MAX_RESUBMIT_RECURSION 16
99 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
100 struct netdev *netdev;
101 struct ofp_phy_port opp; /* In host byte order. */
105 static void ofport_free(struct ofport *);
106 static void hton_ofp_phy_port(struct ofp_phy_port *);
108 struct action_xlate_ctx {
109 /* action_xlate_ctx_init() initializes these members. */
112 struct ofproto *ofproto;
114 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
115 * this flow when actions change header fields. */
118 /* The packet corresponding to 'flow', or a null pointer if we are
119 * revalidating without a packet to refer to. */
120 const struct ofpbuf *packet;
122 /* If nonnull, called just before executing a resubmit action.
124 * This is normally null so the client has to set it manually after
125 * calling action_xlate_ctx_init(). */
126 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
128 /* If true, the speciality of 'flow' should be checked before executing
129 * its actions. If special_cb returns false on 'flow' rendered
130 * uninstallable and no actions will be executed. */
133 /* xlate_actions() initializes and uses these members. The client might want
134 * to look at them after it returns. */
136 struct ofpbuf *odp_actions; /* Datapath actions. */
137 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
138 bool may_set_up_flow; /* True ordinarily; false if the actions must
139 * be reassessed for every packet. */
140 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
142 /* xlate_actions() initializes and uses these members, but the client has no
143 * reason to look at them. */
145 int recurse; /* Recursion level, via xlate_table_action. */
146 int last_pop_priority; /* Offset in 'odp_actions' just past most
147 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
150 static void action_xlate_ctx_init(struct action_xlate_ctx *,
151 struct ofproto *, const struct flow *,
152 const struct ofpbuf *);
153 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
154 const union ofp_action *in, size_t n_in);
156 /* An OpenFlow flow. */
158 long long int used; /* Time last used; time created if not used. */
159 long long int created; /* Creation time. */
163 * - Do include packets and bytes from facets that have been deleted or
164 * whose own statistics have been folded into the rule.
166 * - Do include packets and bytes sent "by hand" that were accounted to
167 * the rule without any facet being involved (this is a rare corner
168 * case in rule_execute()).
170 * - Do not include packet or bytes that can be obtained from any facet's
171 * packet_count or byte_count member or that can be obtained from the
172 * datapath by, e.g., dpif_flow_get() for any facet.
174 uint64_t packet_count; /* Number of packets received. */
175 uint64_t byte_count; /* Number of bytes received. */
177 ovs_be64 flow_cookie; /* Controller-issued identifier. */
179 struct cls_rule cr; /* In owning ofproto's classifier. */
180 uint16_t idle_timeout; /* In seconds from time of last use. */
181 uint16_t hard_timeout; /* In seconds from time of creation. */
182 bool send_flow_removed; /* Send a flow removed message? */
183 int n_actions; /* Number of elements in actions[]. */
184 union ofp_action *actions; /* OpenFlow actions. */
185 struct list facets; /* List of "struct facet"s. */
188 static struct rule *rule_from_cls_rule(const struct cls_rule *);
189 static bool rule_is_hidden(const struct rule *);
191 static struct rule *rule_create(const struct cls_rule *,
192 const union ofp_action *, size_t n_actions,
193 uint16_t idle_timeout, uint16_t hard_timeout,
194 ovs_be64 flow_cookie, bool send_flow_removed);
195 static void rule_destroy(struct ofproto *, struct rule *);
196 static void rule_free(struct rule *);
198 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
199 static void rule_insert(struct ofproto *, struct rule *);
200 static void rule_remove(struct ofproto *, struct rule *);
202 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
203 static void rule_get_stats(const struct rule *, uint64_t *packets,
206 /* An exact-match instantiation of an OpenFlow flow. */
208 long long int used; /* Time last used; time created if not used. */
212 * - Do include packets and bytes sent "by hand", e.g. with
215 * - Do include packets and bytes that were obtained from the datapath
216 * when a flow was deleted (e.g. dpif_flow_del()) or when its
217 * statistics were reset (e.g. dpif_flow_put() with
218 * DPIF_FP_ZERO_STATS).
220 * - Do not include any packets or bytes that can currently be obtained
221 * from the datapath by, e.g., dpif_flow_get().
223 uint64_t packet_count; /* Number of packets received. */
224 uint64_t byte_count; /* Number of bytes received. */
226 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
227 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
229 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
230 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
231 long long int rs_used; /* Used time pushed to resubmit children. */
233 /* Number of bytes passed to account_cb. This may include bytes that can
234 * currently obtained from the datapath (thus, it can be greater than
236 uint64_t accounted_bytes;
238 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
239 struct list list_node; /* In owning rule's 'facets' list. */
240 struct rule *rule; /* Owning rule. */
241 struct flow flow; /* Exact-match flow. */
242 bool installed; /* Installed in datapath? */
243 bool may_install; /* True ordinarily; false if actions must
244 * be reassessed for every packet. */
245 size_t actions_len; /* Number of bytes in actions[]. */
246 struct nlattr *actions; /* Datapath actions. */
247 tag_type tags; /* Tags (set only by hooks). */
248 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
251 static struct facet *facet_create(struct ofproto *, struct rule *,
253 const struct ofpbuf *packet);
254 static void facet_remove(struct ofproto *, struct facet *);
255 static void facet_free(struct facet *);
257 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
258 static bool facet_revalidate(struct ofproto *, struct facet *);
260 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
261 static void facet_uninstall(struct ofproto *, struct facet *);
262 static void facet_flush_stats(struct ofproto *, struct facet *);
264 static void facet_make_actions(struct ofproto *, struct facet *,
265 const struct ofpbuf *packet);
266 static void facet_update_stats(struct ofproto *, struct facet *,
267 const struct dpif_flow_stats *);
268 static void facet_push_stats(struct ofproto *, struct facet *);
270 /* ofproto supports two kinds of OpenFlow connections:
272 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
273 * maintains persistent connections to these controllers and by default
274 * sends them asynchronous messages such as packet-ins.
276 * - "Service" connections, e.g. from ovs-ofctl. When these connections
277 * drop, it is the other side's responsibility to reconnect them if
278 * necessary. ofproto does not send them asynchronous messages by default.
280 * Currently, active (tcp, ssl, unix) connections are always "primary"
281 * connections and passive (ptcp, pssl, punix) connections are always "service"
282 * connections. There is no inherent reason for this, but it reflects the
286 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
287 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
290 /* A listener for incoming OpenFlow "service" connections. */
292 struct hmap_node node; /* In struct ofproto's "services" hmap. */
293 struct pvconn *pvconn; /* OpenFlow connection listener. */
295 /* These are not used by ofservice directly. They are settings for
296 * accepted "struct ofconn"s from the pvconn. */
297 int probe_interval; /* Max idle time before probing, in seconds. */
298 int rate_limit; /* Max packet-in rate in packets per second. */
299 int burst_limit; /* Limit on accumulating packet credits. */
302 static struct ofservice *ofservice_lookup(struct ofproto *,
304 static int ofservice_create(struct ofproto *,
305 const struct ofproto_controller *);
306 static void ofservice_reconfigure(struct ofservice *,
307 const struct ofproto_controller *);
308 static void ofservice_destroy(struct ofproto *, struct ofservice *);
310 /* An OpenFlow connection. */
312 struct ofproto *ofproto; /* The ofproto that owns this connection. */
313 struct list node; /* In struct ofproto's "all_conns" list. */
314 struct rconn *rconn; /* OpenFlow connection. */
315 enum ofconn_type type; /* Type. */
316 enum nx_flow_format flow_format; /* Currently selected flow format. */
318 /* OFPT_PACKET_IN related data. */
319 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
320 #define N_SCHEDULERS 2
321 struct pinsched *schedulers[N_SCHEDULERS];
322 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
323 int miss_send_len; /* Bytes to send of buffered packets. */
325 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
326 * requests, and the maximum number before we stop reading OpenFlow
328 #define OFCONN_REPLY_MAX 100
329 struct rconn_packet_counter *reply_counter;
331 /* type == OFCONN_PRIMARY only. */
332 enum nx_role role; /* Role. */
333 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
334 enum ofproto_band band; /* In-band or out-of-band? */
338 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
340 static void ofconn_destroy(struct ofconn *);
341 static void ofconn_run(struct ofconn *);
342 static void ofconn_wait(struct ofconn *);
343 static bool ofconn_receives_async_msgs(const struct ofconn *);
344 static char *ofconn_make_name(const struct ofproto *, const char *target);
345 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
347 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
348 struct rconn_packet_counter *counter);
350 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
351 const struct flow *, bool clone);
352 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
356 uint64_t datapath_id; /* Datapath ID. */
357 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
358 char *mfr_desc; /* Manufacturer. */
359 char *hw_desc; /* Hardware. */
360 char *sw_desc; /* Software version. */
361 char *serial_desc; /* Serial number. */
362 char *dp_desc; /* Datapath description. */
366 struct netdev_monitor *netdev_monitor;
367 struct hmap ports; /* Contains "struct ofport"s. */
368 struct shash port_by_name;
372 struct fail_open *fail_open;
373 struct netflow *netflow;
374 struct ofproto_sflow *sflow;
376 /* In-band control. */
377 struct in_band *in_band;
378 long long int next_in_band_update;
379 struct sockaddr_in *extra_in_band_remotes;
380 size_t n_extra_remotes;
384 struct classifier cls;
385 long long int next_expiration;
389 bool need_revalidate;
390 struct tag_set revalidate_set;
392 /* OpenFlow connections. */
393 struct hmap controllers; /* Controller "struct ofconn"s. */
394 struct list all_conns; /* Contains "struct ofconn"s. */
395 enum ofproto_fail_mode fail_mode;
397 /* OpenFlow listeners. */
398 struct hmap services; /* Contains "struct ofservice"s. */
399 struct pvconn **snoops;
402 /* Hooks for ovs-vswitchd. */
403 const struct ofhooks *ofhooks;
406 /* Used by default ofhooks. */
407 struct mac_learning *ml;
410 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
411 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
413 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
415 static const struct ofhooks default_ofhooks;
417 static uint64_t pick_datapath_id(const struct ofproto *);
418 static uint64_t pick_fallback_dpid(void);
420 static int ofproto_expire(struct ofproto *);
421 static void flow_push_stats(struct ofproto *, const struct rule *,
422 struct flow *, uint64_t packets, uint64_t bytes,
425 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
427 static void handle_openflow(struct ofconn *, struct ofpbuf *);
429 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
430 static void update_port(struct ofproto *, const char *devname);
431 static int init_ports(struct ofproto *);
432 static void reinit_ports(struct ofproto *);
434 static void ofproto_unixctl_init(void);
437 ofproto_create(const char *datapath, const char *datapath_type,
438 const struct ofhooks *ofhooks, void *aux,
439 struct ofproto **ofprotop)
447 ofproto_unixctl_init();
449 /* Connect to datapath and start listening for messages. */
450 error = dpif_open(datapath, datapath_type, &dpif);
452 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
455 error = dpif_recv_set_mask(dpif,
456 ((1u << DPIF_UC_MISS) |
457 (1u << DPIF_UC_ACTION) |
458 (1u << DPIF_UC_SAMPLE)));
460 VLOG_ERR("failed to listen on datapath %s: %s",
461 datapath, strerror(error));
465 dpif_flow_flush(dpif);
466 dpif_recv_purge(dpif);
468 /* Initialize settings. */
469 p = xzalloc(sizeof *p);
470 p->fallback_dpid = pick_fallback_dpid();
471 p->datapath_id = p->fallback_dpid;
472 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
473 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
474 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
475 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
476 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
478 /* Initialize datapath. */
480 p->netdev_monitor = netdev_monitor_create();
481 hmap_init(&p->ports);
482 shash_init(&p->port_by_name);
483 p->max_ports = dpif_get_max_ports(dpif);
485 /* Initialize submodules. */
490 /* Initialize in-band control. */
492 p->in_band_queue = -1;
494 /* Initialize flow table. */
495 classifier_init(&p->cls);
496 p->next_expiration = time_msec() + 1000;
498 /* Initialize facet table. */
499 hmap_init(&p->facets);
500 p->need_revalidate = false;
501 tag_set_init(&p->revalidate_set);
503 /* Initialize OpenFlow connections. */
504 list_init(&p->all_conns);
505 hmap_init(&p->controllers);
506 hmap_init(&p->services);
510 /* Initialize hooks. */
512 p->ofhooks = ofhooks;
516 p->ofhooks = &default_ofhooks;
518 p->ml = mac_learning_create();
521 /* Pick final datapath ID. */
522 p->datapath_id = pick_datapath_id(p);
523 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
525 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
532 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
534 uint64_t old_dpid = p->datapath_id;
535 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
536 if (p->datapath_id != old_dpid) {
537 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
539 /* Force all active connections to reconnect, since there is no way to
540 * notify a controller that the datapath ID has changed. */
541 ofproto_reconnect_controllers(p);
545 /* Creates a new controller in 'ofproto'. Some of the settings are initially
546 * drawn from 'c', but update_controller() needs to be called later to finish
547 * the new ofconn's configuration. */
549 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
551 char *name = ofconn_make_name(ofproto, c->target);
552 struct ofconn *ofconn;
554 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
555 ofconn->pktbuf = pktbuf_create();
556 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
557 rconn_connect(ofconn->rconn, c->target, name);
558 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
559 hash_string(c->target, 0));
564 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
565 * target (this is done by creating new ofconns and deleting old ones), but it
566 * can update the rest of an ofconn's settings. */
568 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
572 ofconn->band = c->band;
574 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
576 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
577 rconn_set_probe_interval(ofconn->rconn, probe_interval);
579 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
583 ofconn_get_target(const struct ofconn *ofconn)
585 return rconn_get_target(ofconn->rconn);
588 static struct ofconn *
589 find_controller_by_target(struct ofproto *ofproto, const char *target)
591 struct ofconn *ofconn;
593 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
594 hash_string(target, 0), &ofproto->controllers) {
595 if (!strcmp(ofconn_get_target(ofconn), target)) {
603 update_in_band_remotes(struct ofproto *ofproto)
605 const struct ofconn *ofconn;
606 struct sockaddr_in *addrs;
607 size_t max_addrs, n_addrs;
610 /* Allocate enough memory for as many remotes as we could possibly have. */
611 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
612 addrs = xmalloc(max_addrs * sizeof *addrs);
615 /* Add all the remotes. */
616 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
617 struct sockaddr_in *sin = &addrs[n_addrs];
619 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
623 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
624 if (sin->sin_addr.s_addr) {
625 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
629 for (i = 0; i < ofproto->n_extra_remotes; i++) {
630 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
633 /* Create or update or destroy in-band. */
635 if (!ofproto->in_band) {
636 in_band_create(ofproto, ofproto->dpif, &ofproto->in_band);
638 if (ofproto->in_band) {
639 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
641 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
642 ofproto->next_in_band_update = time_msec() + 1000;
644 in_band_destroy(ofproto->in_band);
645 ofproto->in_band = NULL;
653 update_fail_open(struct ofproto *p)
655 struct ofconn *ofconn;
657 if (!hmap_is_empty(&p->controllers)
658 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
659 struct rconn **rconns;
663 p->fail_open = fail_open_create(p);
667 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
668 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
669 rconns[n++] = ofconn->rconn;
672 fail_open_set_controllers(p->fail_open, rconns, n);
673 /* p->fail_open takes ownership of 'rconns'. */
675 fail_open_destroy(p->fail_open);
681 ofproto_set_controllers(struct ofproto *p,
682 const struct ofproto_controller *controllers,
683 size_t n_controllers)
685 struct shash new_controllers;
686 struct ofconn *ofconn, *next_ofconn;
687 struct ofservice *ofservice, *next_ofservice;
690 /* Create newly configured controllers and services.
691 * Create a name to ofproto_controller mapping in 'new_controllers'. */
692 shash_init(&new_controllers);
693 for (i = 0; i < n_controllers; i++) {
694 const struct ofproto_controller *c = &controllers[i];
696 if (!vconn_verify_name(c->target)) {
697 if (!find_controller_by_target(p, c->target)) {
698 add_controller(p, c);
700 } else if (!pvconn_verify_name(c->target)) {
701 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
705 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
706 dpif_name(p->dpif), c->target);
710 shash_add_once(&new_controllers, c->target, &controllers[i]);
713 /* Delete controllers that are no longer configured.
714 * Update configuration of all now-existing controllers. */
715 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
716 struct ofproto_controller *c;
718 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
720 ofconn_destroy(ofconn);
722 update_controller(ofconn, c);
726 /* Delete services that are no longer configured.
727 * Update configuration of all now-existing services. */
728 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
729 struct ofproto_controller *c;
731 c = shash_find_data(&new_controllers,
732 pvconn_get_name(ofservice->pvconn));
734 ofservice_destroy(p, ofservice);
736 ofservice_reconfigure(ofservice, c);
740 shash_destroy(&new_controllers);
742 update_in_band_remotes(p);
747 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
749 p->fail_mode = fail_mode;
753 /* Drops the connections between 'ofproto' and all of its controllers, forcing
754 * them to reconnect. */
756 ofproto_reconnect_controllers(struct ofproto *ofproto)
758 struct ofconn *ofconn;
760 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
761 rconn_reconnect(ofconn->rconn);
766 any_extras_changed(const struct ofproto *ofproto,
767 const struct sockaddr_in *extras, size_t n)
771 if (n != ofproto->n_extra_remotes) {
775 for (i = 0; i < n; i++) {
776 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
777 const struct sockaddr_in *new = &extras[i];
779 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
780 old->sin_port != new->sin_port) {
788 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
789 * in-band control should guarantee access, in the same way that in-band
790 * control guarantees access to OpenFlow controllers. */
792 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
793 const struct sockaddr_in *extras, size_t n)
795 if (!any_extras_changed(ofproto, extras, n)) {
799 free(ofproto->extra_in_band_remotes);
800 ofproto->n_extra_remotes = n;
801 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
803 update_in_band_remotes(ofproto);
806 /* Sets the OpenFlow queue used by flows set up by in-band control on
807 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
808 * flows will use the default queue. */
810 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
812 if (queue_id != ofproto->in_band_queue) {
813 ofproto->in_band_queue = queue_id;
814 update_in_band_remotes(ofproto);
819 ofproto_set_desc(struct ofproto *p,
820 const char *mfr_desc, const char *hw_desc,
821 const char *sw_desc, const char *serial_desc,
824 struct ofp_desc_stats *ods;
827 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
828 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
829 sizeof ods->mfr_desc);
832 p->mfr_desc = xstrdup(mfr_desc);
835 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
836 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
837 sizeof ods->hw_desc);
840 p->hw_desc = xstrdup(hw_desc);
843 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
844 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
845 sizeof ods->sw_desc);
848 p->sw_desc = xstrdup(sw_desc);
851 if (strlen(serial_desc) >= sizeof ods->serial_num) {
852 VLOG_WARN("truncating serial_desc, must be less than %zu "
854 sizeof ods->serial_num);
856 free(p->serial_desc);
857 p->serial_desc = xstrdup(serial_desc);
860 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
861 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
862 sizeof ods->dp_desc);
865 p->dp_desc = xstrdup(dp_desc);
870 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
871 const struct svec *svec)
873 struct pvconn **pvconns = *pvconnsp;
874 size_t n_pvconns = *n_pvconnsp;
878 for (i = 0; i < n_pvconns; i++) {
879 pvconn_close(pvconns[i]);
883 pvconns = xmalloc(svec->n * sizeof *pvconns);
885 for (i = 0; i < svec->n; i++) {
886 const char *name = svec->names[i];
887 struct pvconn *pvconn;
890 error = pvconn_open(name, &pvconn);
892 pvconns[n_pvconns++] = pvconn;
894 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
902 *n_pvconnsp = n_pvconns;
908 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
910 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
914 ofproto_set_netflow(struct ofproto *ofproto,
915 const struct netflow_options *nf_options)
917 if (nf_options && nf_options->collectors.n) {
918 if (!ofproto->netflow) {
919 ofproto->netflow = netflow_create();
921 return netflow_set_options(ofproto->netflow, nf_options);
923 netflow_destroy(ofproto->netflow);
924 ofproto->netflow = NULL;
930 ofproto_set_sflow(struct ofproto *ofproto,
931 const struct ofproto_sflow_options *oso)
933 struct ofproto_sflow *os = ofproto->sflow;
936 struct ofport *ofport;
938 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
939 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
940 ofproto_sflow_add_port(os, ofport->odp_port,
941 netdev_get_name(ofport->netdev));
944 ofproto_sflow_set_options(os, oso);
946 ofproto_sflow_destroy(os);
947 ofproto->sflow = NULL;
952 ofproto_get_datapath_id(const struct ofproto *ofproto)
954 return ofproto->datapath_id;
958 ofproto_has_primary_controller(const struct ofproto *ofproto)
960 return !hmap_is_empty(&ofproto->controllers);
963 enum ofproto_fail_mode
964 ofproto_get_fail_mode(const struct ofproto *p)
970 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
974 for (i = 0; i < ofproto->n_snoops; i++) {
975 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
980 ofproto_destroy(struct ofproto *p)
982 struct ofservice *ofservice, *next_ofservice;
983 struct ofconn *ofconn, *next_ofconn;
984 struct ofport *ofport, *next_ofport;
991 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
993 /* Destroy fail-open and in-band early, since they touch the classifier. */
994 fail_open_destroy(p->fail_open);
997 in_band_destroy(p->in_band);
999 free(p->extra_in_band_remotes);
1001 ofproto_flush_flows(p);
1002 classifier_destroy(&p->cls);
1003 hmap_destroy(&p->facets);
1005 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1006 ofconn_destroy(ofconn);
1008 hmap_destroy(&p->controllers);
1010 dpif_close(p->dpif);
1011 netdev_monitor_destroy(p->netdev_monitor);
1012 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1013 hmap_remove(&p->ports, &ofport->hmap_node);
1014 ofport_free(ofport);
1016 shash_destroy(&p->port_by_name);
1018 netflow_destroy(p->netflow);
1019 ofproto_sflow_destroy(p->sflow);
1021 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1022 ofservice_destroy(p, ofservice);
1024 hmap_destroy(&p->services);
1026 for (i = 0; i < p->n_snoops; i++) {
1027 pvconn_close(p->snoops[i]);
1031 mac_learning_destroy(p->ml);
1036 free(p->serial_desc);
1039 hmap_destroy(&p->ports);
1045 ofproto_run(struct ofproto *p)
1047 int error = ofproto_run1(p);
1049 error = ofproto_run2(p, false);
1055 process_port_change(struct ofproto *ofproto, int error, char *devname)
1057 if (error == ENOBUFS) {
1058 reinit_ports(ofproto);
1059 } else if (!error) {
1060 update_port(ofproto, devname);
1065 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1066 * means that 'ofconn' is more interesting for monitoring than a lower return
1069 snoop_preference(const struct ofconn *ofconn)
1071 switch (ofconn->role) {
1072 case NX_ROLE_MASTER:
1079 /* Shouldn't happen. */
1084 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1085 * Connects this vconn to a controller. */
1087 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1089 struct ofconn *ofconn, *best;
1091 /* Pick a controller for monitoring. */
1093 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1094 if (ofconn->type == OFCONN_PRIMARY
1095 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1101 rconn_add_monitor(best->rconn, vconn);
1103 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1109 ofproto_run1(struct ofproto *p)
1111 struct ofconn *ofconn, *next_ofconn;
1112 struct ofservice *ofservice;
1117 if (shash_is_empty(&p->port_by_name)) {
1121 for (i = 0; i < 50; i++) {
1122 struct dpif_upcall packet;
1124 error = dpif_recv(p->dpif, &packet);
1126 if (error == ENODEV) {
1127 /* Someone destroyed the datapath behind our back. The caller
1128 * better destroy us and give up, because we're just going to
1129 * spin from here on out. */
1130 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1131 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1132 dpif_name(p->dpif));
1138 handle_upcall(p, &packet);
1141 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1142 process_port_change(p, error, devname);
1144 while ((error = netdev_monitor_poll(p->netdev_monitor,
1145 &devname)) != EAGAIN) {
1146 process_port_change(p, error, devname);
1150 if (time_msec() >= p->next_in_band_update) {
1151 update_in_band_remotes(p);
1153 in_band_run(p->in_band);
1156 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1160 /* Fail-open maintenance. Do this after processing the ofconns since
1161 * fail-open checks the status of the controller rconn. */
1163 fail_open_run(p->fail_open);
1166 HMAP_FOR_EACH (ofservice, node, &p->services) {
1167 struct vconn *vconn;
1170 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1172 struct rconn *rconn;
1175 rconn = rconn_create(ofservice->probe_interval, 0);
1176 name = ofconn_make_name(p, vconn_get_name(vconn));
1177 rconn_connect_unreliably(rconn, vconn, name);
1180 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1181 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1182 ofservice->burst_limit);
1183 } else if (retval != EAGAIN) {
1184 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1188 for (i = 0; i < p->n_snoops; i++) {
1189 struct vconn *vconn;
1192 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1194 add_snooper(p, vconn);
1195 } else if (retval != EAGAIN) {
1196 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1200 if (time_msec() >= p->next_expiration) {
1201 int delay = ofproto_expire(p);
1202 p->next_expiration = time_msec() + delay;
1203 COVERAGE_INC(ofproto_expiration);
1207 netflow_run(p->netflow);
1210 ofproto_sflow_run(p->sflow);
1217 ofproto_run2(struct ofproto *p, bool revalidate_all)
1219 /* Figure out what we need to revalidate now, if anything. */
1220 struct tag_set revalidate_set = p->revalidate_set;
1221 if (p->need_revalidate) {
1222 revalidate_all = true;
1225 /* Clear the revalidation flags. */
1226 tag_set_init(&p->revalidate_set);
1227 p->need_revalidate = false;
1229 /* Now revalidate if there's anything to do. */
1230 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1231 struct facet *facet, *next;
1233 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1235 || tag_set_intersects(&revalidate_set, facet->tags)) {
1236 facet_revalidate(p, facet);
1245 ofproto_wait(struct ofproto *p)
1247 struct ofservice *ofservice;
1248 struct ofconn *ofconn;
1251 dpif_recv_wait(p->dpif);
1252 dpif_port_poll_wait(p->dpif);
1253 netdev_monitor_poll_wait(p->netdev_monitor);
1254 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1255 ofconn_wait(ofconn);
1258 poll_timer_wait_until(p->next_in_band_update);
1259 in_band_wait(p->in_band);
1262 fail_open_wait(p->fail_open);
1265 ofproto_sflow_wait(p->sflow);
1267 if (!tag_set_is_empty(&p->revalidate_set)) {
1268 poll_immediate_wake();
1270 if (p->need_revalidate) {
1271 /* Shouldn't happen, but if it does just go around again. */
1272 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1273 poll_immediate_wake();
1274 } else if (p->next_expiration != LLONG_MAX) {
1275 poll_timer_wait_until(p->next_expiration);
1277 HMAP_FOR_EACH (ofservice, node, &p->services) {
1278 pvconn_wait(ofservice->pvconn);
1280 for (i = 0; i < p->n_snoops; i++) {
1281 pvconn_wait(p->snoops[i]);
1286 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1288 tag_set_add(&ofproto->revalidate_set, tag);
1292 ofproto_get_revalidate_set(struct ofproto *ofproto)
1294 return &ofproto->revalidate_set;
1298 ofproto_is_alive(const struct ofproto *p)
1300 return !hmap_is_empty(&p->controllers);
1304 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1307 const struct ofconn *ofconn;
1311 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1312 const struct rconn *rconn = ofconn->rconn;
1313 time_t now = time_now();
1314 time_t last_connection = rconn_get_last_connection(rconn);
1315 time_t last_disconnect = rconn_get_last_disconnect(rconn);
1316 const int last_error = rconn_get_last_error(rconn);
1317 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1319 shash_add(info, rconn_get_target(rconn), cinfo);
1321 cinfo->is_connected = rconn_is_connected(rconn);
1322 cinfo->role = ofconn->role;
1327 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1328 cinfo->pairs.values[cinfo->pairs.n++] =
1329 xstrdup(ovs_retval_to_string(last_error));
1332 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1333 cinfo->pairs.values[cinfo->pairs.n++] =
1334 xstrdup(rconn_get_state(rconn));
1336 if (last_connection != TIME_MIN) {
1337 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
1338 cinfo->pairs.values[cinfo->pairs.n++]
1339 = xasprintf("%ld", (long int) (now - last_connection));
1342 if (last_disconnect != TIME_MIN) {
1343 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
1344 cinfo->pairs.values[cinfo->pairs.n++]
1345 = xasprintf("%ld", (long int) (now - last_disconnect));
1351 ofproto_free_ofproto_controller_info(struct shash *info)
1353 struct shash_node *node;
1355 SHASH_FOR_EACH (node, info) {
1356 struct ofproto_controller_info *cinfo = node->data;
1357 while (cinfo->pairs.n) {
1358 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1362 shash_destroy(info);
1365 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1367 * This is almost the same as calling dpif_port_del() directly on the
1368 * datapath, but it also makes 'ofproto' close its open netdev for the port
1369 * (if any). This makes it possible to create a new netdev of a different
1370 * type under the same name, which otherwise the netdev library would refuse
1371 * to do because of the conflict. (The netdev would eventually get closed on
1372 * the next trip through ofproto_run(), but this interface is more direct.)
1374 * Returns 0 if successful, otherwise a positive errno. */
1376 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1378 struct ofport *ofport = get_port(ofproto, odp_port);
1379 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1382 error = dpif_port_del(ofproto->dpif, odp_port);
1384 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1385 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1386 } else if (ofport) {
1387 /* 'name' is the netdev's name and update_port() is going to close the
1388 * netdev. Just in case update_port() refers to 'name' after it
1389 * destroys 'ofport', make a copy of it around the update_port()
1391 char *devname = xstrdup(name);
1392 update_port(ofproto, devname);
1398 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1399 * true if 'odp_port' exists and should be included, false otherwise. */
1401 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1403 struct ofport *ofport = get_port(ofproto, odp_port);
1404 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1408 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1409 const union ofp_action *actions, size_t n_actions,
1410 const struct ofpbuf *packet)
1412 struct action_xlate_ctx ctx;
1413 struct ofpbuf *odp_actions;
1415 action_xlate_ctx_init(&ctx, p, flow, packet);
1416 /* Always xlate packets originated in this function. */
1417 ctx.check_special = false;
1418 odp_actions = xlate_actions(&ctx, actions, n_actions);
1420 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1422 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
1424 ofpbuf_delete(odp_actions);
1429 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1430 * performs the 'n_actions' actions in 'actions'. The new flow will not
1433 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1434 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1435 * controllers; otherwise, it will be hidden.
1437 * The caller retains ownership of 'cls_rule' and 'actions'. */
1439 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1440 const union ofp_action *actions, size_t n_actions)
1443 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1444 rule_insert(p, rule);
1448 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1452 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1455 rule_remove(ofproto, rule);
1460 ofproto_flush_flows(struct ofproto *ofproto)
1462 struct facet *facet, *next_facet;
1463 struct rule *rule, *next_rule;
1464 struct cls_cursor cursor;
1466 COVERAGE_INC(ofproto_flush);
1468 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1469 /* Mark the facet as not installed so that facet_remove() doesn't
1470 * bother trying to uninstall it. There is no point in uninstalling it
1471 * individually since we are about to blow away all the facets with
1472 * dpif_flow_flush(). */
1473 facet->installed = false;
1474 facet->dp_packet_count = 0;
1475 facet->dp_byte_count = 0;
1476 facet_remove(ofproto, facet);
1479 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1480 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1481 rule_remove(ofproto, rule);
1484 dpif_flow_flush(ofproto->dpif);
1485 if (ofproto->in_band) {
1486 in_band_flushed(ofproto->in_band);
1488 if (ofproto->fail_open) {
1489 fail_open_flushed(ofproto->fail_open);
1494 reinit_ports(struct ofproto *p)
1496 struct dpif_port_dump dump;
1497 struct shash_node *node;
1498 struct shash devnames;
1499 struct ofport *ofport;
1500 struct dpif_port dpif_port;
1502 COVERAGE_INC(ofproto_reinit_ports);
1504 shash_init(&devnames);
1505 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1506 shash_add_once (&devnames, netdev_get_name(ofport->netdev), NULL);
1508 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1509 shash_add_once (&devnames, dpif_port.name, NULL);
1512 SHASH_FOR_EACH (node, &devnames) {
1513 update_port(p, node->name);
1515 shash_destroy(&devnames);
1518 static struct ofport *
1519 make_ofport(const struct dpif_port *dpif_port)
1521 struct netdev_options netdev_options;
1522 enum netdev_flags flags;
1523 struct ofport *ofport;
1524 struct netdev *netdev;
1527 memset(&netdev_options, 0, sizeof netdev_options);
1528 netdev_options.name = dpif_port->name;
1529 netdev_options.type = dpif_port->type;
1530 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1532 error = netdev_open(&netdev_options, &netdev);
1534 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1535 "cannot be opened (%s)",
1536 dpif_port->name, dpif_port->port_no,
1537 dpif_port->name, strerror(error));
1541 ofport = xzalloc(sizeof *ofport);
1542 ofport->netdev = netdev;
1543 ofport->odp_port = dpif_port->port_no;
1544 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1545 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1546 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1548 netdev_get_flags(netdev, &flags);
1549 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1551 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1553 netdev_get_features(netdev,
1554 &ofport->opp.curr, &ofport->opp.advertised,
1555 &ofport->opp.supported, &ofport->opp.peer);
1560 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1562 if (get_port(p, dpif_port->port_no)) {
1563 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1564 dpif_port->port_no);
1566 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1567 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1576 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1578 const struct ofp_phy_port *a = &a_->opp;
1579 const struct ofp_phy_port *b = &b_->opp;
1581 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1582 return (a->port_no == b->port_no
1583 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1584 && !strcmp(a->name, b->name)
1585 && a->state == b->state
1586 && a->config == b->config
1587 && a->curr == b->curr
1588 && a->advertised == b->advertised
1589 && a->supported == b->supported
1590 && a->peer == b->peer);
1594 send_port_status(struct ofproto *p, const struct ofport *ofport,
1597 /* XXX Should limit the number of queued port status change messages. */
1598 struct ofconn *ofconn;
1599 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1600 struct ofp_port_status *ops;
1603 /* Primary controllers, even slaves, should always get port status
1604 updates. Otherwise obey ofconn_receives_async_msgs(). */
1605 if (ofconn->type != OFCONN_PRIMARY
1606 && !ofconn_receives_async_msgs(ofconn)) {
1610 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1611 ops->reason = reason;
1612 ops->desc = ofport->opp;
1613 hton_ofp_phy_port(&ops->desc);
1614 queue_tx(b, ofconn, NULL);
1619 ofport_install(struct ofproto *p, struct ofport *ofport)
1621 const char *netdev_name = netdev_get_name(ofport->netdev);
1623 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1624 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1625 shash_add(&p->port_by_name, netdev_name, ofport);
1627 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1632 ofport_remove(struct ofproto *p, struct ofport *ofport)
1634 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1635 hmap_remove(&p->ports, &ofport->hmap_node);
1636 shash_delete(&p->port_by_name,
1637 shash_find(&p->port_by_name,
1638 netdev_get_name(ofport->netdev)));
1640 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1645 ofport_free(struct ofport *ofport)
1648 netdev_close(ofport->netdev);
1653 static struct ofport *
1654 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1656 struct ofport *port;
1658 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1659 hash_int(odp_port, 0), &ofproto->ports) {
1660 if (port->odp_port == odp_port) {
1668 update_port(struct ofproto *p, const char *devname)
1670 struct dpif_port dpif_port;
1671 struct ofport *old_ofport;
1672 struct ofport *new_ofport;
1675 COVERAGE_INC(ofproto_update_port);
1677 /* Query the datapath for port information. */
1678 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1680 /* Find the old ofport. */
1681 old_ofport = shash_find_data(&p->port_by_name, devname);
1684 /* There's no port named 'devname' but there might be a port with
1685 * the same port number. This could happen if a port is deleted
1686 * and then a new one added in its place very quickly, or if a port
1687 * is renamed. In the former case we want to send an OFPPR_DELETE
1688 * and an OFPPR_ADD, and in the latter case we want to send a
1689 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1690 * the old port's ifindex against the new port, or perhaps less
1691 * reliably but more portably by comparing the old port's MAC
1692 * against the new port's MAC. However, this code isn't that smart
1693 * and always sends an OFPPR_MODIFY (XXX). */
1694 old_ofport = get_port(p, dpif_port.port_no);
1696 } else if (error != ENOENT && error != ENODEV) {
1697 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1698 "%s", strerror(error));
1702 /* Create a new ofport. */
1703 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1705 /* Eliminate a few pathological cases. */
1706 if (!old_ofport && !new_ofport) {
1708 } else if (old_ofport && new_ofport) {
1709 /* Most of the 'config' bits are OpenFlow soft state, but
1710 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1711 * OpenFlow bits from old_ofport. (make_ofport() only sets
1712 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1713 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1715 if (ofport_equal(old_ofport, new_ofport)) {
1716 /* False alarm--no change. */
1717 ofport_free(new_ofport);
1722 /* Now deal with the normal cases. */
1724 ofport_remove(p, old_ofport);
1727 ofport_install(p, new_ofport);
1729 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1730 (!old_ofport ? OFPPR_ADD
1731 : !new_ofport ? OFPPR_DELETE
1733 ofport_free(old_ofport);
1736 dpif_port_destroy(&dpif_port);
1740 init_ports(struct ofproto *p)
1742 struct dpif_port_dump dump;
1743 struct dpif_port dpif_port;
1745 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1746 if (!ofport_conflicts(p, &dpif_port)) {
1747 struct ofport *ofport = make_ofport(&dpif_port);
1749 ofport_install(p, ofport);
1757 static struct ofconn *
1758 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1760 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1761 ofconn->ofproto = p;
1762 list_push_back(&p->all_conns, &ofconn->node);
1763 ofconn->rconn = rconn;
1764 ofconn->type = type;
1765 ofconn->flow_format = NXFF_OPENFLOW10;
1766 ofconn->role = NX_ROLE_OTHER;
1767 ofconn->packet_in_counter = rconn_packet_counter_create ();
1768 ofconn->pktbuf = NULL;
1769 ofconn->miss_send_len = 0;
1770 ofconn->reply_counter = rconn_packet_counter_create ();
1775 ofconn_destroy(struct ofconn *ofconn)
1777 if (ofconn->type == OFCONN_PRIMARY) {
1778 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1781 list_remove(&ofconn->node);
1782 rconn_destroy(ofconn->rconn);
1783 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1784 rconn_packet_counter_destroy(ofconn->reply_counter);
1785 pktbuf_destroy(ofconn->pktbuf);
1790 ofconn_run(struct ofconn *ofconn)
1792 struct ofproto *p = ofconn->ofproto;
1796 for (i = 0; i < N_SCHEDULERS; i++) {
1797 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1800 rconn_run(ofconn->rconn);
1802 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1803 /* Limit the number of iterations to prevent other tasks from
1805 for (iteration = 0; iteration < 50; iteration++) {
1806 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1811 fail_open_maybe_recover(p->fail_open);
1813 handle_openflow(ofconn, of_msg);
1814 ofpbuf_delete(of_msg);
1818 if (!rconn_is_alive(ofconn->rconn)) {
1819 ofconn_destroy(ofconn);
1824 ofconn_wait(struct ofconn *ofconn)
1828 for (i = 0; i < N_SCHEDULERS; i++) {
1829 pinsched_wait(ofconn->schedulers[i]);
1831 rconn_run_wait(ofconn->rconn);
1832 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1833 rconn_recv_wait(ofconn->rconn);
1835 COVERAGE_INC(ofproto_ofconn_stuck);
1839 /* Returns true if 'ofconn' should receive asynchronous messages. */
1841 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1843 if (ofconn->type == OFCONN_PRIMARY) {
1844 /* Primary controllers always get asynchronous messages unless they
1845 * have configured themselves as "slaves". */
1846 return ofconn->role != NX_ROLE_SLAVE;
1848 /* Service connections don't get asynchronous messages unless they have
1849 * explicitly asked for them by setting a nonzero miss send length. */
1850 return ofconn->miss_send_len > 0;
1854 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1855 * and 'target', suitable for use in log messages for identifying the
1858 * The name is dynamically allocated. The caller should free it (with free())
1859 * when it is no longer needed. */
1861 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1863 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1867 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1871 for (i = 0; i < N_SCHEDULERS; i++) {
1872 struct pinsched **s = &ofconn->schedulers[i];
1876 *s = pinsched_create(rate, burst);
1878 pinsched_set_limits(*s, rate, burst);
1881 pinsched_destroy(*s);
1888 ofservice_reconfigure(struct ofservice *ofservice,
1889 const struct ofproto_controller *c)
1891 ofservice->probe_interval = c->probe_interval;
1892 ofservice->rate_limit = c->rate_limit;
1893 ofservice->burst_limit = c->burst_limit;
1896 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1897 * positive errno value. */
1899 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1901 struct ofservice *ofservice;
1902 struct pvconn *pvconn;
1905 error = pvconn_open(c->target, &pvconn);
1910 ofservice = xzalloc(sizeof *ofservice);
1911 hmap_insert(&ofproto->services, &ofservice->node,
1912 hash_string(c->target, 0));
1913 ofservice->pvconn = pvconn;
1915 ofservice_reconfigure(ofservice, c);
1921 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1923 hmap_remove(&ofproto->services, &ofservice->node);
1924 pvconn_close(ofservice->pvconn);
1928 /* Finds and returns the ofservice within 'ofproto' that has the given
1929 * 'target', or a null pointer if none exists. */
1930 static struct ofservice *
1931 ofservice_lookup(struct ofproto *ofproto, const char *target)
1933 struct ofservice *ofservice;
1935 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1936 &ofproto->services) {
1937 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1944 /* Returns true if 'rule' should be hidden from the controller.
1946 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1947 * (e.g. by in-band control) and are intentionally hidden from the
1950 rule_is_hidden(const struct rule *rule)
1952 return rule->cr.priority > UINT16_MAX;
1955 /* Creates and returns a new rule initialized as specified.
1957 * The caller is responsible for inserting the rule into the classifier (with
1958 * rule_insert()). */
1959 static struct rule *
1960 rule_create(const struct cls_rule *cls_rule,
1961 const union ofp_action *actions, size_t n_actions,
1962 uint16_t idle_timeout, uint16_t hard_timeout,
1963 ovs_be64 flow_cookie, bool send_flow_removed)
1965 struct rule *rule = xzalloc(sizeof *rule);
1966 rule->cr = *cls_rule;
1967 rule->idle_timeout = idle_timeout;
1968 rule->hard_timeout = hard_timeout;
1969 rule->flow_cookie = flow_cookie;
1970 rule->used = rule->created = time_msec();
1971 rule->send_flow_removed = send_flow_removed;
1972 list_init(&rule->facets);
1973 if (n_actions > 0) {
1974 rule->n_actions = n_actions;
1975 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1981 static struct rule *
1982 rule_from_cls_rule(const struct cls_rule *cls_rule)
1984 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1988 rule_free(struct rule *rule)
1990 free(rule->actions);
1994 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1995 * destroying any that no longer has a rule (which is probably all of them).
1997 * The caller must have already removed 'rule' from the classifier. */
1999 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2001 struct facet *facet, *next_facet;
2002 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2003 facet_revalidate(ofproto, facet);
2008 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2009 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2012 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2014 const union ofp_action *oa;
2015 struct actions_iterator i;
2017 if (out_port == htons(OFPP_NONE)) {
2020 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2021 oa = actions_next(&i)) {
2022 if (action_outputs_to_port(oa, out_port)) {
2029 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2030 * 'packet', which arrived on 'in_port'.
2032 * Takes ownership of 'packet'. */
2034 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2035 const struct nlattr *odp_actions, size_t actions_len,
2036 struct ofpbuf *packet)
2038 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2039 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2040 /* As an optimization, avoid a round-trip from userspace to kernel to
2041 * userspace. This also avoids possibly filling up kernel packet
2042 * buffers along the way. */
2043 struct dpif_upcall upcall;
2045 upcall.type = DPIF_UC_ACTION;
2046 upcall.packet = packet;
2049 upcall.userdata = nl_attr_get_u64(odp_actions);
2050 upcall.sample_pool = 0;
2051 upcall.actions = NULL;
2052 upcall.actions_len = 0;
2054 send_packet_in(ofproto, &upcall, flow, false);
2060 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2061 ofpbuf_delete(packet);
2066 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2067 * statistics appropriately. 'packet' must have at least sizeof(struct
2068 * ofp_packet_in) bytes of headroom.
2070 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2071 * applying flow_extract() to 'packet' would yield the same flow as
2074 * 'facet' must have accurately composed ODP actions; that is, it must not be
2075 * in need of revalidation.
2077 * Takes ownership of 'packet'. */
2079 facet_execute(struct ofproto *ofproto, struct facet *facet,
2080 struct ofpbuf *packet)
2082 struct dpif_flow_stats stats;
2084 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2086 flow_extract_stats(&facet->flow, packet, &stats);
2087 stats.used = time_msec();
2088 if (execute_odp_actions(ofproto, &facet->flow,
2089 facet->actions, facet->actions_len, packet)) {
2090 facet_update_stats(ofproto, facet, &stats);
2094 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2095 * statistics (or the statistics for one of its facets) appropriately.
2096 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2098 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2099 * with statistics for 'packet' either way.
2101 * Takes ownership of 'packet'. */
2103 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2104 struct ofpbuf *packet)
2106 struct action_xlate_ctx ctx;
2107 struct ofpbuf *odp_actions;
2108 struct facet *facet;
2112 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2114 flow_extract(packet, 0, in_port, &flow);
2116 /* First look for a related facet. If we find one, account it to that. */
2117 facet = facet_lookup_valid(ofproto, &flow);
2118 if (facet && facet->rule == rule) {
2119 facet_execute(ofproto, facet, packet);
2123 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2124 * create a new facet for it and use that. */
2125 if (rule_lookup(ofproto, &flow) == rule) {
2126 facet = facet_create(ofproto, rule, &flow, packet);
2127 facet_execute(ofproto, facet, packet);
2128 facet_install(ofproto, facet, true);
2132 /* We can't account anything to a facet. If we were to try, then that
2133 * facet would have a non-matching rule, busting our invariants. */
2134 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2135 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2136 size = packet->size;
2137 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2138 odp_actions->size, packet)) {
2139 rule->used = time_msec();
2140 rule->packet_count++;
2141 rule->byte_count += size;
2142 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2144 ofpbuf_delete(odp_actions);
2147 /* Inserts 'rule' into 'p''s flow table. */
2149 rule_insert(struct ofproto *p, struct rule *rule)
2151 struct rule *displaced_rule;
2153 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2154 if (displaced_rule) {
2155 rule_destroy(p, displaced_rule);
2157 p->need_revalidate = true;
2160 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2161 * 'flow' and an example 'packet' within that flow.
2163 * The caller must already have determined that no facet with an identical
2164 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2165 * 'ofproto''s classifier table. */
2166 static struct facet *
2167 facet_create(struct ofproto *ofproto, struct rule *rule,
2168 const struct flow *flow, const struct ofpbuf *packet)
2170 struct facet *facet;
2172 facet = xzalloc(sizeof *facet);
2173 facet->used = time_msec();
2174 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2175 list_push_back(&rule->facets, &facet->list_node);
2177 facet->flow = *flow;
2178 netflow_flow_init(&facet->nf_flow);
2179 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2181 facet_make_actions(ofproto, facet, packet);
2187 facet_free(struct facet *facet)
2189 free(facet->actions);
2193 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2195 * - Removes 'rule' from the classifier.
2197 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2198 * destroys them), via rule_destroy().
2201 rule_remove(struct ofproto *ofproto, struct rule *rule)
2203 COVERAGE_INC(ofproto_del_rule);
2204 ofproto->need_revalidate = true;
2205 classifier_remove(&ofproto->cls, &rule->cr);
2206 rule_destroy(ofproto, rule);
2209 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2211 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2212 * rule's statistics, via facet_uninstall().
2214 * - Removes 'facet' from its rule and from ofproto->facets.
2217 facet_remove(struct ofproto *ofproto, struct facet *facet)
2219 facet_uninstall(ofproto, facet);
2220 facet_flush_stats(ofproto, facet);
2221 hmap_remove(&ofproto->facets, &facet->hmap_node);
2222 list_remove(&facet->list_node);
2226 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2228 facet_make_actions(struct ofproto *p, struct facet *facet,
2229 const struct ofpbuf *packet)
2231 const struct rule *rule = facet->rule;
2232 struct ofpbuf *odp_actions;
2233 struct action_xlate_ctx ctx;
2235 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2236 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2237 facet->tags = ctx.tags;
2238 facet->may_install = ctx.may_set_up_flow;
2239 facet->nf_flow.output_iface = ctx.nf_output_iface;
2241 if (facet->actions_len != odp_actions->size
2242 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2243 free(facet->actions);
2244 facet->actions_len = odp_actions->size;
2245 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2248 ofpbuf_delete(odp_actions);
2252 facet_put__(struct ofproto *ofproto, struct facet *facet,
2253 const struct nlattr *actions, size_t actions_len,
2254 struct dpif_flow_stats *stats)
2256 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2257 enum dpif_flow_put_flags flags;
2260 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2262 flags |= DPIF_FP_ZERO_STATS;
2263 facet->dp_packet_count = 0;
2264 facet->dp_byte_count = 0;
2267 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2268 odp_flow_key_from_flow(&key, &facet->flow);
2269 assert(key.base == keybuf);
2271 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2272 actions, actions_len, stats);
2275 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2276 * 'zero_stats' is true, clears any existing statistics from the datapath for
2279 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2281 struct dpif_flow_stats stats;
2283 if (facet->may_install
2284 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2285 zero_stats ? &stats : NULL)) {
2286 facet->installed = true;
2290 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2291 * to the accounting hook function in the ofhooks structure. */
2293 facet_account(struct ofproto *ofproto,
2294 struct facet *facet, uint64_t extra_bytes)
2296 uint64_t total_bytes = facet->byte_count + extra_bytes;
2298 if (ofproto->ofhooks->account_flow_cb
2299 && total_bytes > facet->accounted_bytes)
2301 ofproto->ofhooks->account_flow_cb(
2302 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2303 total_bytes - facet->accounted_bytes, ofproto->aux);
2304 facet->accounted_bytes = total_bytes;
2308 /* If 'rule' is installed in the datapath, uninstalls it. */
2310 facet_uninstall(struct ofproto *p, struct facet *facet)
2312 if (facet->installed) {
2313 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2314 struct dpif_flow_stats stats;
2317 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2318 odp_flow_key_from_flow(&key, &facet->flow);
2319 assert(key.base == keybuf);
2321 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2322 facet_update_stats(p, facet, &stats);
2324 facet->installed = false;
2325 facet->dp_packet_count = 0;
2326 facet->dp_byte_count = 0;
2328 assert(facet->dp_packet_count == 0);
2329 assert(facet->dp_byte_count == 0);
2333 /* Returns true if the only action for 'facet' is to send to the controller.
2334 * (We don't report NetFlow expiration messages for such facets because they
2335 * are just part of the control logic for the network, not real traffic). */
2337 facet_is_controller_flow(struct facet *facet)
2340 && facet->rule->n_actions == 1
2341 && action_outputs_to_port(&facet->rule->actions[0],
2342 htons(OFPP_CONTROLLER)));
2345 /* Folds all of 'facet''s statistics into its rule. Also updates the
2346 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2347 * 'facet''s statistics in the datapath should have been zeroed and folded into
2348 * its packet and byte counts before this function is called. */
2350 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2352 assert(!facet->dp_byte_count);
2353 assert(!facet->dp_packet_count);
2355 facet_push_stats(ofproto, facet);
2356 facet_account(ofproto, facet, 0);
2358 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2359 struct ofexpired expired;
2360 expired.flow = facet->flow;
2361 expired.packet_count = facet->packet_count;
2362 expired.byte_count = facet->byte_count;
2363 expired.used = facet->used;
2364 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2367 facet->rule->packet_count += facet->packet_count;
2368 facet->rule->byte_count += facet->byte_count;
2370 /* Reset counters to prevent double counting if 'facet' ever gets
2372 facet->packet_count = 0;
2373 facet->byte_count = 0;
2374 facet->rs_packet_count = 0;
2375 facet->rs_byte_count = 0;
2376 facet->accounted_bytes = 0;
2378 netflow_flow_clear(&facet->nf_flow);
2381 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2382 * Returns it if found, otherwise a null pointer.
2384 * The returned facet might need revalidation; use facet_lookup_valid()
2385 * instead if that is important. */
2386 static struct facet *
2387 facet_find(struct ofproto *ofproto, const struct flow *flow)
2389 struct facet *facet;
2391 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2393 if (flow_equal(flow, &facet->flow)) {
2401 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2402 * Returns it if found, otherwise a null pointer.
2404 * The returned facet is guaranteed to be valid. */
2405 static struct facet *
2406 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2408 struct facet *facet = facet_find(ofproto, flow);
2410 /* The facet we found might not be valid, since we could be in need of
2411 * revalidation. If it is not valid, don't return it. */
2413 && ofproto->need_revalidate
2414 && !facet_revalidate(ofproto, facet)) {
2415 COVERAGE_INC(ofproto_invalidated);
2422 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2424 * - If the rule found is different from 'facet''s current rule, moves
2425 * 'facet' to the new rule and recompiles its actions.
2427 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2428 * where it is and recompiles its actions anyway.
2430 * - If there is none, destroys 'facet'.
2432 * Returns true if 'facet' still exists, false if it has been destroyed. */
2434 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2436 struct action_xlate_ctx ctx;
2437 struct ofpbuf *odp_actions;
2438 struct rule *new_rule;
2439 bool actions_changed;
2441 COVERAGE_INC(facet_revalidate);
2443 /* Determine the new rule. */
2444 new_rule = rule_lookup(ofproto, &facet->flow);
2446 /* No new rule, so delete the facet. */
2447 facet_remove(ofproto, facet);
2451 /* Calculate new ODP actions.
2453 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2454 * emit a NetFlow expiration and, if so, we need to have the old state
2455 * around to properly compose it. */
2456 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2457 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2458 actions_changed = (facet->actions_len != odp_actions->size
2459 || memcmp(facet->actions, odp_actions->data,
2460 facet->actions_len));
2462 /* If the ODP actions changed or the installability changed, then we need
2463 * to talk to the datapath. */
2464 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2465 if (ctx.may_set_up_flow) {
2466 struct dpif_flow_stats stats;
2468 facet_put__(ofproto, facet,
2469 odp_actions->data, odp_actions->size, &stats);
2470 facet_update_stats(ofproto, facet, &stats);
2472 facet_uninstall(ofproto, facet);
2475 /* The datapath flow is gone or has zeroed stats, so push stats out of
2476 * 'facet' into 'rule'. */
2477 facet_flush_stats(ofproto, facet);
2480 /* Update 'facet' now that we've taken care of all the old state. */
2481 facet->tags = ctx.tags;
2482 facet->nf_flow.output_iface = ctx.nf_output_iface;
2483 facet->may_install = ctx.may_set_up_flow;
2484 if (actions_changed) {
2485 free(facet->actions);
2486 facet->actions_len = odp_actions->size;
2487 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2489 if (facet->rule != new_rule) {
2490 COVERAGE_INC(facet_changed_rule);
2491 list_remove(&facet->list_node);
2492 list_push_back(&new_rule->facets, &facet->list_node);
2493 facet->rule = new_rule;
2494 facet->used = new_rule->created;
2495 facet->rs_used = facet->used;
2498 ofpbuf_delete(odp_actions);
2504 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2505 struct rconn_packet_counter *counter)
2507 update_openflow_length(msg);
2508 if (rconn_send(ofconn->rconn, msg, counter)) {
2514 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2517 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2519 COVERAGE_INC(ofproto_error);
2520 queue_tx(buf, ofconn, ofconn->reply_counter);
2525 hton_ofp_phy_port(struct ofp_phy_port *opp)
2527 opp->port_no = htons(opp->port_no);
2528 opp->config = htonl(opp->config);
2529 opp->state = htonl(opp->state);
2530 opp->curr = htonl(opp->curr);
2531 opp->advertised = htonl(opp->advertised);
2532 opp->supported = htonl(opp->supported);
2533 opp->peer = htonl(opp->peer);
2537 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2539 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2544 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2546 struct ofp_switch_features *osf;
2548 struct ofport *port;
2550 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2551 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2552 osf->n_buffers = htonl(pktbuf_capacity());
2554 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2555 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2556 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2557 (1u << OFPAT_SET_VLAN_VID) |
2558 (1u << OFPAT_SET_VLAN_PCP) |
2559 (1u << OFPAT_STRIP_VLAN) |
2560 (1u << OFPAT_SET_DL_SRC) |
2561 (1u << OFPAT_SET_DL_DST) |
2562 (1u << OFPAT_SET_NW_SRC) |
2563 (1u << OFPAT_SET_NW_DST) |
2564 (1u << OFPAT_SET_NW_TOS) |
2565 (1u << OFPAT_SET_TP_SRC) |
2566 (1u << OFPAT_SET_TP_DST) |
2567 (1u << OFPAT_ENQUEUE));
2569 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2570 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2573 queue_tx(buf, ofconn, ofconn->reply_counter);
2578 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2581 struct ofp_switch_config *osc;
2585 /* Figure out flags. */
2586 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2587 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2590 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2591 osc->flags = htons(flags);
2592 osc->miss_send_len = htons(ofconn->miss_send_len);
2593 queue_tx(buf, ofconn, ofconn->reply_counter);
2599 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2601 uint16_t flags = ntohs(osc->flags);
2603 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2604 switch (flags & OFPC_FRAG_MASK) {
2605 case OFPC_FRAG_NORMAL:
2606 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2608 case OFPC_FRAG_DROP:
2609 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2612 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2618 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2623 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2624 struct action_xlate_ctx *ctx);
2627 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2629 const struct ofport *ofport = get_port(ctx->ofproto, port);
2632 if (ofport->opp.config & OFPPC_NO_FWD) {
2633 /* Forwarding disabled on port. */
2638 * We don't have an ofport record for this port, but it doesn't hurt to
2639 * allow forwarding to it anyhow. Maybe such a port will appear later
2640 * and we're pre-populating the flow table.
2644 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2645 ctx->nf_output_iface = port;
2648 static struct rule *
2649 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2651 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2655 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2657 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2658 uint16_t old_in_port;
2661 /* Look up a flow with 'in_port' as the input port. Then restore the
2662 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2663 * have surprising behavior). */
2664 old_in_port = ctx->flow.in_port;
2665 ctx->flow.in_port = in_port;
2666 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2667 ctx->flow.in_port = old_in_port;
2669 if (ctx->resubmit_hook) {
2670 ctx->resubmit_hook(ctx, rule);
2675 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2679 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2681 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2682 MAX_RESUBMIT_RECURSION);
2687 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2688 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2690 struct ofport *ofport;
2692 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2693 uint16_t odp_port = ofport->odp_port;
2694 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2695 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2698 *nf_output_iface = NF_OUT_FLOOD;
2702 xlate_output_action__(struct action_xlate_ctx *ctx,
2703 uint16_t port, uint16_t max_len)
2706 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2708 ctx->nf_output_iface = NF_OUT_DROP;
2712 add_output_action(ctx, ctx->flow.in_port);
2715 xlate_table_action(ctx, ctx->flow.in_port);
2718 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2719 ctx->odp_actions, &ctx->tags,
2720 &ctx->nf_output_iface,
2721 ctx->ofproto->aux)) {
2722 COVERAGE_INC(ofproto_uninstallable);
2723 ctx->may_set_up_flow = false;
2727 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2728 &ctx->nf_output_iface, ctx->odp_actions);
2731 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2732 &ctx->nf_output_iface, ctx->odp_actions);
2734 case OFPP_CONTROLLER:
2735 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2738 add_output_action(ctx, ODPP_LOCAL);
2741 odp_port = ofp_port_to_odp_port(port);
2742 if (odp_port != ctx->flow.in_port) {
2743 add_output_action(ctx, odp_port);
2748 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2749 ctx->nf_output_iface = NF_OUT_FLOOD;
2750 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2751 ctx->nf_output_iface = prev_nf_output_iface;
2752 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2753 ctx->nf_output_iface != NF_OUT_FLOOD) {
2754 ctx->nf_output_iface = NF_OUT_MULTI;
2759 xlate_output_action(struct action_xlate_ctx *ctx,
2760 const struct ofp_action_output *oao)
2762 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2765 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2766 * optimization, because we're going to add another action that sets the
2767 * priority immediately after, or because there are no actions following the
2770 remove_pop_action(struct action_xlate_ctx *ctx)
2772 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2773 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2774 ctx->last_pop_priority = -1;
2779 add_pop_action(struct action_xlate_ctx *ctx)
2781 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2782 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2783 ctx->last_pop_priority = ctx->odp_actions->size;
2788 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2789 const struct ofp_action_enqueue *oae)
2791 uint16_t ofp_port, odp_port;
2795 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2798 /* Fall back to ordinary output action. */
2799 xlate_output_action__(ctx, ntohs(oae->port), 0);
2803 /* Figure out ODP output port. */
2804 ofp_port = ntohs(oae->port);
2805 if (ofp_port != OFPP_IN_PORT) {
2806 odp_port = ofp_port_to_odp_port(ofp_port);
2808 odp_port = ctx->flow.in_port;
2811 /* Add ODP actions. */
2812 remove_pop_action(ctx);
2813 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2814 add_output_action(ctx, odp_port);
2815 add_pop_action(ctx);
2817 /* Update NetFlow output port. */
2818 if (ctx->nf_output_iface == NF_OUT_DROP) {
2819 ctx->nf_output_iface = odp_port;
2820 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2821 ctx->nf_output_iface = NF_OUT_MULTI;
2826 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2827 const struct nx_action_set_queue *nasq)
2832 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2835 /* Couldn't translate queue to a priority, so ignore. A warning
2836 * has already been logged. */
2840 remove_pop_action(ctx);
2841 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2845 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2847 ovs_be16 tci = ctx->flow.vlan_tci;
2848 if (!(tci & htons(VLAN_CFI))) {
2849 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2851 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2852 tci & ~htons(VLAN_CFI));
2856 struct xlate_reg_state {
2862 save_reg_state(const struct action_xlate_ctx *ctx,
2863 struct xlate_reg_state *state)
2865 state->vlan_tci = ctx->flow.vlan_tci;
2866 state->tun_id = ctx->flow.tun_id;
2870 update_reg_state(struct action_xlate_ctx *ctx,
2871 const struct xlate_reg_state *state)
2873 if (ctx->flow.vlan_tci != state->vlan_tci) {
2874 xlate_set_dl_tci(ctx);
2876 if (ctx->flow.tun_id != state->tun_id) {
2877 nl_msg_put_be64(ctx->odp_actions,
2878 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2883 xlate_nicira_action(struct action_xlate_ctx *ctx,
2884 const struct nx_action_header *nah)
2886 const struct nx_action_resubmit *nar;
2887 const struct nx_action_set_tunnel *nast;
2888 const struct nx_action_set_queue *nasq;
2889 const struct nx_action_multipath *nam;
2890 enum nx_action_subtype subtype = ntohs(nah->subtype);
2891 struct xlate_reg_state state;
2894 assert(nah->vendor == htonl(NX_VENDOR_ID));
2896 case NXAST_RESUBMIT:
2897 nar = (const struct nx_action_resubmit *) nah;
2898 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2901 case NXAST_SET_TUNNEL:
2902 nast = (const struct nx_action_set_tunnel *) nah;
2903 tun_id = htonll(ntohl(nast->tun_id));
2904 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2905 ctx->flow.tun_id = tun_id;
2908 case NXAST_DROP_SPOOFED_ARP:
2909 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2910 nl_msg_put_flag(ctx->odp_actions,
2911 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2915 case NXAST_SET_QUEUE:
2916 nasq = (const struct nx_action_set_queue *) nah;
2917 xlate_set_queue_action(ctx, nasq);
2920 case NXAST_POP_QUEUE:
2921 add_pop_action(ctx);
2924 case NXAST_REG_MOVE:
2925 save_reg_state(ctx, &state);
2926 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2928 update_reg_state(ctx, &state);
2931 case NXAST_REG_LOAD:
2932 save_reg_state(ctx, &state);
2933 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2935 update_reg_state(ctx, &state);
2939 /* Nothing to do. */
2942 case NXAST_SET_TUNNEL64:
2943 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2944 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2945 ctx->flow.tun_id = tun_id;
2948 case NXAST_MULTIPATH:
2949 nam = (const struct nx_action_multipath *) nah;
2950 multipath_execute(nam, &ctx->flow);
2953 /* If you add a new action here that modifies flow data, don't forget to
2954 * update the flow key in ctx->flow at the same time. */
2956 case NXAST_SNAT__OBSOLETE:
2958 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2964 do_xlate_actions(const union ofp_action *in, size_t n_in,
2965 struct action_xlate_ctx *ctx)
2967 struct actions_iterator iter;
2968 const union ofp_action *ia;
2969 const struct ofport *port;
2971 port = get_port(ctx->ofproto, ctx->flow.in_port);
2972 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2973 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2974 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2975 /* Drop this flow. */
2979 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2980 enum ofp_action_type type = ntohs(ia->type);
2981 const struct ofp_action_dl_addr *oada;
2985 xlate_output_action(ctx, &ia->output);
2988 case OFPAT_SET_VLAN_VID:
2989 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2990 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2991 xlate_set_dl_tci(ctx);
2994 case OFPAT_SET_VLAN_PCP:
2995 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2996 ctx->flow.vlan_tci |= htons(
2997 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2998 xlate_set_dl_tci(ctx);
3001 case OFPAT_STRIP_VLAN:
3002 ctx->flow.vlan_tci = htons(0);
3003 xlate_set_dl_tci(ctx);
3006 case OFPAT_SET_DL_SRC:
3007 oada = ((struct ofp_action_dl_addr *) ia);
3008 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3009 oada->dl_addr, ETH_ADDR_LEN);
3010 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3013 case OFPAT_SET_DL_DST:
3014 oada = ((struct ofp_action_dl_addr *) ia);
3015 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3016 oada->dl_addr, ETH_ADDR_LEN);
3017 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3020 case OFPAT_SET_NW_SRC:
3021 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3022 ia->nw_addr.nw_addr);
3023 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3026 case OFPAT_SET_NW_DST:
3027 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3028 ia->nw_addr.nw_addr);
3029 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3032 case OFPAT_SET_NW_TOS:
3033 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3035 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3038 case OFPAT_SET_TP_SRC:
3039 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3040 ia->tp_port.tp_port);
3041 ctx->flow.tp_src = ia->tp_port.tp_port;
3044 case OFPAT_SET_TP_DST:
3045 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3046 ia->tp_port.tp_port);
3047 ctx->flow.tp_dst = ia->tp_port.tp_port;
3051 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3055 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3059 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3066 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3067 struct ofproto *ofproto, const struct flow *flow,
3068 const struct ofpbuf *packet)
3070 ctx->ofproto = ofproto;
3072 ctx->packet = packet;
3073 ctx->resubmit_hook = NULL;
3074 ctx->check_special = true;
3077 static struct ofpbuf *
3078 xlate_actions(struct action_xlate_ctx *ctx,
3079 const union ofp_action *in, size_t n_in)
3081 COVERAGE_INC(ofproto_ofp2odp);
3083 ctx->odp_actions = ofpbuf_new(512);
3085 ctx->may_set_up_flow = true;
3086 ctx->nf_output_iface = NF_OUT_DROP;
3088 ctx->last_pop_priority = -1;
3090 if (!ctx->check_special
3091 || !ctx->ofproto->ofhooks->special_cb
3092 || ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3093 ctx->ofproto->aux)) {
3094 do_xlate_actions(in, n_in, ctx);
3096 ctx->may_set_up_flow = false;
3099 remove_pop_action(ctx);
3101 /* Check with in-band control to see if we're allowed to set up this
3103 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3104 ctx->odp_actions->data, ctx->odp_actions->size)) {
3105 ctx->may_set_up_flow = false;
3108 return ctx->odp_actions;
3111 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3112 * error message code (composed with ofp_mkerr()) for the caller to propagate
3113 * upward. Otherwise, returns 0.
3115 * The log message mentions 'msg_type'. */
3117 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3119 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3120 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3121 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3124 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3131 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3133 struct ofproto *p = ofconn->ofproto;
3134 struct ofp_packet_out *opo;
3135 struct ofpbuf payload, *buffer;
3136 union ofp_action *ofp_actions;
3137 struct action_xlate_ctx ctx;
3138 struct ofpbuf *odp_actions;
3139 struct ofpbuf request;
3141 size_t n_ofp_actions;
3145 COVERAGE_INC(ofproto_packet_out);
3147 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3152 /* Get ofp_packet_out. */
3153 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3154 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3157 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3158 &ofp_actions, &n_ofp_actions);
3164 if (opo->buffer_id != htonl(UINT32_MAX)) {
3165 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3167 if (error || !buffer) {
3176 /* Extract flow, check actions. */
3177 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3179 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3185 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3186 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3187 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3188 ofpbuf_delete(odp_actions);
3191 ofpbuf_delete(buffer);
3196 update_port_config(struct ofproto *p, struct ofport *port,
3197 uint32_t config, uint32_t mask)
3199 mask &= config ^ port->opp.config;
3200 if (mask & OFPPC_PORT_DOWN) {
3201 if (config & OFPPC_PORT_DOWN) {
3202 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3204 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3207 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3208 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3209 if (mask & REVALIDATE_BITS) {
3210 COVERAGE_INC(ofproto_costly_flags);
3211 port->opp.config ^= mask & REVALIDATE_BITS;
3212 p->need_revalidate = true;
3214 #undef REVALIDATE_BITS
3215 if (mask & OFPPC_NO_PACKET_IN) {
3216 port->opp.config ^= OFPPC_NO_PACKET_IN;
3221 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3223 struct ofproto *p = ofconn->ofproto;
3224 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3225 struct ofport *port;
3228 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3233 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3235 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3236 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3237 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3239 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3240 if (opm->advertise) {
3241 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3247 static struct ofpbuf *
3248 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3250 struct ofp_stats_reply *osr;
3253 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3254 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3256 osr->flags = htons(0);
3260 static struct ofpbuf *
3261 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3263 const struct ofp_stats_request *osr
3264 = (const struct ofp_stats_request *) request;
3265 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3269 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3270 struct ofpbuf **msgp)
3272 struct ofpbuf *msg = *msgp;
3273 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3274 if (nbytes + msg->size > UINT16_MAX) {
3275 struct ofp_stats_reply *reply = msg->data;
3276 reply->flags = htons(OFPSF_REPLY_MORE);
3277 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3278 queue_tx(msg, ofconn, ofconn->reply_counter);
3280 return ofpbuf_put_uninit(*msgp, nbytes);
3283 static struct ofpbuf *
3284 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3286 struct nicira_stats_msg *nsm;
3289 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3290 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3291 nsm->type = htons(OFPST_VENDOR);
3292 nsm->flags = htons(0);
3293 nsm->vendor = htonl(NX_VENDOR_ID);
3294 nsm->subtype = subtype;
3298 static struct ofpbuf *
3299 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3301 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3305 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3306 struct ofpbuf **msgp)
3308 struct ofpbuf *msg = *msgp;
3309 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3310 if (nbytes + msg->size > UINT16_MAX) {
3311 struct nicira_stats_msg *reply = msg->data;
3312 reply->flags = htons(OFPSF_REPLY_MORE);
3313 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3314 queue_tx(msg, ofconn, ofconn->reply_counter);
3316 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3320 handle_desc_stats_request(struct ofconn *ofconn,
3321 const struct ofp_header *request)
3323 struct ofproto *p = ofconn->ofproto;
3324 struct ofp_desc_stats *ods;
3327 msg = start_ofp_stats_reply(request, sizeof *ods);
3328 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3329 memset(ods, 0, sizeof *ods);
3330 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3331 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3332 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3333 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3334 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3335 queue_tx(msg, ofconn, ofconn->reply_counter);
3341 handle_table_stats_request(struct ofconn *ofconn,
3342 const struct ofp_header *request)
3344 struct ofproto *p = ofconn->ofproto;
3345 struct ofp_table_stats *ots;
3348 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3350 /* Classifier table. */
3351 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3352 memset(ots, 0, sizeof *ots);
3353 strcpy(ots->name, "classifier");
3354 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3355 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3356 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3357 ots->active_count = htonl(classifier_count(&p->cls));
3358 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3359 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3361 queue_tx(msg, ofconn, ofconn->reply_counter);
3366 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3367 struct ofpbuf **msgp)
3369 struct netdev_stats stats;
3370 struct ofp_port_stats *ops;
3372 /* Intentionally ignore return value, since errors will set
3373 * 'stats' to all-1s, which is correct for OpenFlow, and
3374 * netdev_get_stats() will log errors. */
3375 netdev_get_stats(port->netdev, &stats);
3377 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3378 ops->port_no = htons(port->opp.port_no);
3379 memset(ops->pad, 0, sizeof ops->pad);
3380 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3381 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3382 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3383 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3384 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3385 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3386 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3387 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3388 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3389 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3390 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3391 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3395 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3397 struct ofproto *p = ofconn->ofproto;
3398 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3399 struct ofp_port_stats *ops;
3401 struct ofport *port;
3403 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3404 if (psr->port_no != htons(OFPP_NONE)) {
3405 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3407 append_port_stat(port, ofconn, &msg);
3410 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3411 append_port_stat(port, ofconn, &msg);
3415 queue_tx(msg, ofconn, ofconn->reply_counter);
3420 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3422 long long int msecs = time_msec() - start;
3423 *sec = htonl(msecs / 1000);
3424 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3428 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3429 ovs_be16 out_port, struct ofpbuf **replyp)
3431 struct ofp_flow_stats *ofs;
3432 uint64_t packet_count, byte_count;
3434 size_t act_len, len;
3436 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3440 act_len = sizeof *rule->actions * rule->n_actions;
3441 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3443 rule_get_stats(rule, &packet_count, &byte_count);
3445 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3446 ofs->length = htons(len);
3449 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3450 rule->flow_cookie, &cookie);
3451 put_32aligned_be64(&ofs->cookie, cookie);
3452 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3453 ofs->priority = htons(rule->cr.priority);
3454 ofs->idle_timeout = htons(rule->idle_timeout);
3455 ofs->hard_timeout = htons(rule->hard_timeout);
3456 memset(ofs->pad2, 0, sizeof ofs->pad2);
3457 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3458 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3459 if (rule->n_actions > 0) {
3460 memcpy(ofs->actions, rule->actions, act_len);
3465 is_valid_table(uint8_t table_id)
3467 if (table_id == 0 || table_id == 0xff) {
3470 /* It would probably be better to reply with an error but there doesn't
3471 * seem to be any appropriate value, so that might just be
3473 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3480 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3482 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3483 struct ofpbuf *reply;
3485 COVERAGE_INC(ofproto_flows_req);
3486 reply = start_ofp_stats_reply(oh, 1024);
3487 if (is_valid_table(fsr->table_id)) {
3488 struct cls_cursor cursor;
3489 struct cls_rule target;
3492 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3494 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3495 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3496 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3499 queue_tx(reply, ofconn, ofconn->reply_counter);
3505 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3506 ovs_be16 out_port, struct ofpbuf **replyp)
3508 struct nx_flow_stats *nfs;
3509 uint64_t packet_count, byte_count;
3510 size_t act_len, start_len;
3511 struct ofpbuf *reply;
3513 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3517 rule_get_stats(rule, &packet_count, &byte_count);
3519 act_len = sizeof *rule->actions * rule->n_actions;
3521 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3522 start_len = (*replyp)->size;
3525 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3528 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3529 nfs->cookie = rule->flow_cookie;
3530 nfs->priority = htons(rule->cr.priority);
3531 nfs->idle_timeout = htons(rule->idle_timeout);
3532 nfs->hard_timeout = htons(rule->hard_timeout);
3533 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3534 memset(nfs->pad2, 0, sizeof nfs->pad2);
3535 nfs->packet_count = htonll(packet_count);
3536 nfs->byte_count = htonll(byte_count);
3537 if (rule->n_actions > 0) {
3538 ofpbuf_put(reply, rule->actions, act_len);
3540 nfs->length = htons(reply->size - start_len);
3544 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3546 struct nx_flow_stats_request *nfsr;
3547 struct cls_rule target;
3548 struct ofpbuf *reply;
3552 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3554 /* Dissect the message. */
3555 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3556 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3561 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3564 COVERAGE_INC(ofproto_flows_req);
3565 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3566 if (is_valid_table(nfsr->table_id)) {
3567 struct cls_cursor cursor;
3570 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3571 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3572 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3575 queue_tx(reply, ofconn, ofconn->reply_counter);
3581 flow_stats_ds(struct rule *rule, struct ds *results)
3583 uint64_t packet_count, byte_count;
3584 size_t act_len = sizeof *rule->actions * rule->n_actions;
3586 rule_get_stats(rule, &packet_count, &byte_count);
3588 ds_put_format(results, "duration=%llds, ",
3589 (time_msec() - rule->created) / 1000);
3590 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3591 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3592 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3593 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3594 cls_rule_format(&rule->cr, results);
3595 ds_put_char(results, ',');
3597 ofp_print_actions(results, &rule->actions->header, act_len);
3599 ds_put_cstr(results, "drop");
3601 ds_put_cstr(results, "\n");
3604 /* Adds a pretty-printed description of all flows to 'results', including
3605 * hidden flows (e.g., set up by in-band control). */
3607 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3609 struct cls_cursor cursor;
3612 cls_cursor_init(&cursor, &p->cls, NULL);
3613 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3614 flow_stats_ds(rule, results);
3619 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3620 ovs_be16 out_port, uint8_t table_id,
3621 struct ofp_aggregate_stats_reply *oasr)
3623 uint64_t total_packets = 0;
3624 uint64_t total_bytes = 0;
3627 COVERAGE_INC(ofproto_agg_request);
3629 if (is_valid_table(table_id)) {
3630 struct cls_cursor cursor;
3633 cls_cursor_init(&cursor, &ofproto->cls, target);
3634 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3635 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3636 uint64_t packet_count;
3637 uint64_t byte_count;
3639 rule_get_stats(rule, &packet_count, &byte_count);
3641 total_packets += packet_count;
3642 total_bytes += byte_count;
3648 oasr->flow_count = htonl(n_flows);
3649 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3650 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3651 memset(oasr->pad, 0, sizeof oasr->pad);
3655 handle_aggregate_stats_request(struct ofconn *ofconn,
3656 const struct ofp_header *oh)
3658 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3659 struct ofp_aggregate_stats_reply *reply;
3660 struct cls_rule target;
3663 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3666 msg = start_ofp_stats_reply(oh, sizeof *reply);
3667 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3668 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3669 request->table_id, reply);
3670 queue_tx(msg, ofconn, ofconn->reply_counter);
3675 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3677 struct nx_aggregate_stats_request *request;
3678 struct ofp_aggregate_stats_reply *reply;
3679 struct cls_rule target;
3684 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3686 /* Dissect the message. */
3687 request = ofpbuf_pull(&b, sizeof *request);
3688 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3693 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3697 COVERAGE_INC(ofproto_flows_req);
3698 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3699 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3700 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3701 request->table_id, reply);
3702 queue_tx(buf, ofconn, ofconn->reply_counter);
3707 struct queue_stats_cbdata {
3708 struct ofconn *ofconn;
3709 struct ofport *ofport;
3714 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3715 const struct netdev_queue_stats *stats)
3717 struct ofp_queue_stats *reply;
3719 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3720 reply->port_no = htons(cbdata->ofport->opp.port_no);
3721 memset(reply->pad, 0, sizeof reply->pad);
3722 reply->queue_id = htonl(queue_id);
3723 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3724 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3725 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3729 handle_queue_stats_dump_cb(uint32_t queue_id,
3730 struct netdev_queue_stats *stats,
3733 struct queue_stats_cbdata *cbdata = cbdata_;
3735 put_queue_stats(cbdata, queue_id, stats);
3739 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3740 struct queue_stats_cbdata *cbdata)
3742 cbdata->ofport = port;
3743 if (queue_id == OFPQ_ALL) {
3744 netdev_dump_queue_stats(port->netdev,
3745 handle_queue_stats_dump_cb, cbdata);
3747 struct netdev_queue_stats stats;
3749 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3750 put_queue_stats(cbdata, queue_id, &stats);
3756 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3758 struct ofproto *ofproto = ofconn->ofproto;
3759 const struct ofp_queue_stats_request *qsr;
3760 struct queue_stats_cbdata cbdata;
3761 struct ofport *port;
3762 unsigned int port_no;
3765 qsr = ofputil_stats_body(oh);
3767 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3770 COVERAGE_INC(ofproto_queue_req);
3772 cbdata.ofconn = ofconn;
3773 cbdata.msg = start_ofp_stats_reply(oh, 128);
3775 port_no = ntohs(qsr->port_no);
3776 queue_id = ntohl(qsr->queue_id);
3777 if (port_no == OFPP_ALL) {
3778 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3779 handle_queue_stats_for_port(port, queue_id, &cbdata);
3781 } else if (port_no < ofproto->max_ports) {
3782 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3784 handle_queue_stats_for_port(port, queue_id, &cbdata);
3787 ofpbuf_delete(cbdata.msg);
3788 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3790 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3795 /* Updates 'facet''s used time. Caller is responsible for calling
3796 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3798 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3801 if (used > facet->used) {
3803 if (used > facet->rule->used) {
3804 facet->rule->used = used;
3806 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3810 /* Folds the statistics from 'stats' into the counters in 'facet'.
3812 * Because of the meaning of a facet's counters, it only makes sense to do this
3813 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3814 * packet that was sent by hand or if it represents statistics that have been
3815 * cleared out of the datapath. */
3817 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3818 const struct dpif_flow_stats *stats)
3820 if (stats->n_packets || stats->used > facet->used) {
3821 facet_update_time(ofproto, facet, stats->used);
3822 facet->packet_count += stats->n_packets;
3823 facet->byte_count += stats->n_bytes;
3824 facet_push_stats(ofproto, facet);
3825 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3830 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3832 uint64_t rs_packets, rs_bytes;
3834 assert(facet->packet_count >= facet->rs_packet_count);
3835 assert(facet->byte_count >= facet->rs_byte_count);
3836 assert(facet->used >= facet->rs_used);
3838 rs_packets = facet->packet_count - facet->rs_packet_count;
3839 rs_bytes = facet->byte_count - facet->rs_byte_count;
3841 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3842 facet->rs_packet_count = facet->packet_count;
3843 facet->rs_byte_count = facet->byte_count;
3844 facet->rs_used = facet->used;
3846 flow_push_stats(ofproto, facet->rule, &facet->flow,
3847 rs_packets, rs_bytes, facet->used);
3851 struct ofproto_push {
3852 struct action_xlate_ctx ctx;
3859 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3861 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3864 rule->packet_count += push->packets;
3865 rule->byte_count += push->bytes;
3866 rule->used = MAX(push->used, rule->used);
3870 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3871 * 'rule''s actions. */
3873 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3874 struct flow *flow, uint64_t packets, uint64_t bytes,
3877 struct ofproto_push push;
3879 push.packets = packets;
3883 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3884 push.ctx.resubmit_hook = push_resubmit;
3885 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3888 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3889 * in which no matching flow already exists in the flow table.
3891 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3892 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3893 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3895 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3898 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3900 struct ofproto *p = ofconn->ofproto;
3901 struct ofpbuf *packet;
3906 if (fm->flags & OFPFF_CHECK_OVERLAP
3907 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3908 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3912 if (fm->buffer_id != UINT32_MAX) {
3913 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3917 in_port = UINT16_MAX;
3920 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3921 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3922 fm->flags & OFPFF_SEND_FLOW_REM);
3923 rule_insert(p, rule);
3925 rule_execute(p, rule, in_port, packet);
3930 static struct rule *
3931 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3933 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3937 send_buffered_packet(struct ofconn *ofconn,
3938 struct rule *rule, uint32_t buffer_id)
3940 struct ofpbuf *packet;
3944 if (buffer_id == UINT32_MAX) {
3948 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3953 rule_execute(ofconn->ofproto, rule, in_port, packet);
3958 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3960 struct modify_flows_cbdata {
3961 struct ofproto *ofproto;
3962 const struct flow_mod *fm;
3966 static int modify_flow(struct ofproto *, const struct flow_mod *,
3969 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3970 * encoded by ofp_mkerr() on failure.
3972 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3975 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3977 struct ofproto *p = ofconn->ofproto;
3978 struct rule *match = NULL;
3979 struct cls_cursor cursor;
3982 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3983 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3984 if (!rule_is_hidden(rule)) {
3986 modify_flow(p, fm, rule);
3991 /* This credits the packet to whichever flow happened to match last.
3992 * That's weird. Maybe we should do a lookup for the flow that
3993 * actually matches the packet? Who knows. */
3994 send_buffered_packet(ofconn, match, fm->buffer_id);
3997 return add_flow(ofconn, fm);
4001 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4002 * code as encoded by ofp_mkerr() on failure.
4004 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4007 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4009 struct ofproto *p = ofconn->ofproto;
4010 struct rule *rule = find_flow_strict(p, fm);
4011 if (rule && !rule_is_hidden(rule)) {
4012 modify_flow(p, fm, rule);
4013 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4015 return add_flow(ofconn, fm);
4019 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4020 * been identified as a flow in 'p''s flow table to be modified, by changing
4021 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4022 * ofp_action[] structures). */
4024 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4026 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4028 rule->flow_cookie = fm->cookie;
4030 /* If the actions are the same, do nothing. */
4031 if (fm->n_actions == rule->n_actions
4033 || !memcmp(fm->actions, rule->actions, actions_len))) {
4037 /* Replace actions. */
4038 free(rule->actions);
4039 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4040 rule->n_actions = fm->n_actions;
4042 p->need_revalidate = true;
4047 /* OFPFC_DELETE implementation. */
4049 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4051 /* Implements OFPFC_DELETE. */
4053 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4055 struct rule *rule, *next_rule;
4056 struct cls_cursor cursor;
4058 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4059 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4060 delete_flow(p, rule, htons(fm->out_port));
4064 /* Implements OFPFC_DELETE_STRICT. */
4066 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4068 struct rule *rule = find_flow_strict(p, fm);
4070 delete_flow(p, rule, htons(fm->out_port));
4074 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4075 * been identified as a flow to delete from 'p''s flow table, by deleting the
4076 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4079 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4080 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4081 * specified 'out_port'. */
4083 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4085 if (rule_is_hidden(rule)) {
4089 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4093 rule_send_removed(p, rule, OFPRR_DELETE);
4094 rule_remove(p, rule);
4098 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4100 struct ofproto *p = ofconn->ofproto;
4104 error = reject_slave_controller(ofconn, "flow_mod");
4109 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4114 /* We do not support the emergency flow cache. It will hopefully get
4115 * dropped from OpenFlow in the near future. */
4116 if (fm.flags & OFPFF_EMERG) {
4117 /* There isn't a good fit for an error code, so just state that the
4118 * flow table is full. */
4119 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4122 error = validate_actions(fm.actions, fm.n_actions,
4123 &fm.cr.flow, p->max_ports);
4128 switch (fm.command) {
4130 return add_flow(ofconn, &fm);
4133 return modify_flows_loose(ofconn, &fm);
4135 case OFPFC_MODIFY_STRICT:
4136 return modify_flow_strict(ofconn, &fm);
4139 delete_flows_loose(p, &fm);
4142 case OFPFC_DELETE_STRICT:
4143 delete_flow_strict(p, &fm);
4147 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4152 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4154 const struct nxt_tun_id_cookie *msg
4155 = (const struct nxt_tun_id_cookie *) oh;
4157 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4162 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4164 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4165 struct nx_role_request *reply;
4169 if (ofconn->type != OFCONN_PRIMARY) {
4170 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4172 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4175 role = ntohl(nrr->role);
4176 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4177 && role != NX_ROLE_SLAVE) {
4178 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4180 /* There's no good error code for this. */
4181 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4184 if (role == NX_ROLE_MASTER) {
4185 struct ofconn *other;
4187 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4188 if (other->role == NX_ROLE_MASTER) {
4189 other->role = NX_ROLE_SLAVE;
4193 ofconn->role = role;
4195 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4196 reply->role = htonl(role);
4197 queue_tx(buf, ofconn, ofconn->reply_counter);
4203 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4205 const struct nxt_set_flow_format *msg
4206 = (const struct nxt_set_flow_format *) oh;
4209 format = ntohl(msg->format);
4210 if (format == NXFF_OPENFLOW10
4211 || format == NXFF_TUN_ID_FROM_COOKIE
4212 || format == NXFF_NXM) {
4213 ofconn->flow_format = format;
4216 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4221 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4223 struct ofp_header *ob;
4226 /* Currently, everything executes synchronously, so we can just
4227 * immediately send the barrier reply. */
4228 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4229 queue_tx(buf, ofconn, ofconn->reply_counter);
4234 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4236 const struct ofp_header *oh = msg->data;
4237 const struct ofputil_msg_type *type;
4240 error = ofputil_decode_msg_type(oh, &type);
4245 switch (ofputil_msg_type_code(type)) {
4246 /* OpenFlow requests. */
4247 case OFPUTIL_OFPT_ECHO_REQUEST:
4248 return handle_echo_request(ofconn, oh);
4250 case OFPUTIL_OFPT_FEATURES_REQUEST:
4251 return handle_features_request(ofconn, oh);
4253 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4254 return handle_get_config_request(ofconn, oh);
4256 case OFPUTIL_OFPT_SET_CONFIG:
4257 return handle_set_config(ofconn, msg->data);
4259 case OFPUTIL_OFPT_PACKET_OUT:
4260 return handle_packet_out(ofconn, oh);
4262 case OFPUTIL_OFPT_PORT_MOD:
4263 return handle_port_mod(ofconn, oh);
4265 case OFPUTIL_OFPT_FLOW_MOD:
4266 return handle_flow_mod(ofconn, oh);
4268 case OFPUTIL_OFPT_BARRIER_REQUEST:
4269 return handle_barrier_request(ofconn, oh);
4271 /* OpenFlow replies. */
4272 case OFPUTIL_OFPT_ECHO_REPLY:
4275 /* Nicira extension requests. */
4276 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4277 return handle_tun_id_from_cookie(ofconn, oh);
4279 case OFPUTIL_NXT_ROLE_REQUEST:
4280 return handle_role_request(ofconn, oh);
4282 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4283 return handle_nxt_set_flow_format(ofconn, oh);
4285 case OFPUTIL_NXT_FLOW_MOD:
4286 return handle_flow_mod(ofconn, oh);
4288 /* OpenFlow statistics requests. */
4289 case OFPUTIL_OFPST_DESC_REQUEST:
4290 return handle_desc_stats_request(ofconn, oh);
4292 case OFPUTIL_OFPST_FLOW_REQUEST:
4293 return handle_flow_stats_request(ofconn, oh);
4295 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4296 return handle_aggregate_stats_request(ofconn, oh);
4298 case OFPUTIL_OFPST_TABLE_REQUEST:
4299 return handle_table_stats_request(ofconn, oh);
4301 case OFPUTIL_OFPST_PORT_REQUEST:
4302 return handle_port_stats_request(ofconn, oh);
4304 case OFPUTIL_OFPST_QUEUE_REQUEST:
4305 return handle_queue_stats_request(ofconn, oh);
4307 /* Nicira extension statistics requests. */
4308 case OFPUTIL_NXST_FLOW_REQUEST:
4309 return handle_nxst_flow(ofconn, oh);
4311 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4312 return handle_nxst_aggregate(ofconn, oh);
4314 case OFPUTIL_INVALID:
4315 case OFPUTIL_OFPT_HELLO:
4316 case OFPUTIL_OFPT_ERROR:
4317 case OFPUTIL_OFPT_FEATURES_REPLY:
4318 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4319 case OFPUTIL_OFPT_PACKET_IN:
4320 case OFPUTIL_OFPT_FLOW_REMOVED:
4321 case OFPUTIL_OFPT_PORT_STATUS:
4322 case OFPUTIL_OFPT_BARRIER_REPLY:
4323 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4324 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4325 case OFPUTIL_OFPST_DESC_REPLY:
4326 case OFPUTIL_OFPST_FLOW_REPLY:
4327 case OFPUTIL_OFPST_QUEUE_REPLY:
4328 case OFPUTIL_OFPST_PORT_REPLY:
4329 case OFPUTIL_OFPST_TABLE_REPLY:
4330 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4331 case OFPUTIL_NXT_ROLE_REPLY:
4332 case OFPUTIL_NXT_FLOW_REMOVED:
4333 case OFPUTIL_NXST_FLOW_REPLY:
4334 case OFPUTIL_NXST_AGGREGATE_REPLY:
4336 if (VLOG_IS_WARN_ENABLED()) {
4337 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4338 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4341 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4342 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4344 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4350 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4352 int error = handle_openflow__(ofconn, ofp_msg);
4354 send_error_oh(ofconn, ofp_msg->data, error);
4356 COVERAGE_INC(ofproto_recv_openflow);
4360 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4362 struct facet *facet;
4365 /* Obtain in_port and tun_id, at least. */
4366 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4368 /* Set header pointers in 'flow'. */
4369 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4371 if (p->ofhooks->special_cb
4372 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4373 ofpbuf_delete(upcall->packet);
4377 /* Check with in-band control to see if this packet should be sent
4378 * to the local port regardless of the flow table. */
4379 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4380 struct ofpbuf odp_actions;
4382 ofpbuf_init(&odp_actions, 32);
4383 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, ODPP_LOCAL);
4384 dpif_execute(p->dpif, odp_actions.data, odp_actions.size,
4386 ofpbuf_uninit(&odp_actions);
4389 facet = facet_lookup_valid(p, &flow);
4391 struct rule *rule = rule_lookup(p, &flow);
4393 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4394 struct ofport *port = get_port(p, flow.in_port);
4396 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4397 COVERAGE_INC(ofproto_no_packet_in);
4398 /* XXX install 'drop' flow entry */
4399 ofpbuf_delete(upcall->packet);
4403 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4407 COVERAGE_INC(ofproto_packet_in);
4408 send_packet_in(p, upcall, &flow, false);
4412 facet = facet_create(p, rule, &flow, upcall->packet);
4413 } else if (!facet->may_install) {
4414 /* The facet is not installable, that is, we need to process every
4415 * packet, so process the current packet's actions into 'facet'. */
4416 facet_make_actions(p, facet, upcall->packet);
4419 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4421 * Extra-special case for fail-open mode.
4423 * We are in fail-open mode and the packet matched the fail-open rule,
4424 * but we are connected to a controller too. We should send the packet
4425 * up to the controller in the hope that it will try to set up a flow
4426 * and thereby allow us to exit fail-open.
4428 * See the top-level comment in fail-open.c for more information.
4430 send_packet_in(p, upcall, &flow, true);
4433 facet_execute(p, facet, upcall->packet);
4434 facet_install(p, facet, false);
4438 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4442 switch (upcall->type) {
4443 case DPIF_UC_ACTION:
4444 COVERAGE_INC(ofproto_ctlr_action);
4445 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4446 send_packet_in(p, upcall, &flow, false);
4449 case DPIF_UC_SAMPLE:
4451 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4452 ofproto_sflow_received(p->sflow, upcall, &flow);
4454 ofpbuf_delete(upcall->packet);
4458 handle_miss_upcall(p, upcall);
4461 case DPIF_N_UC_TYPES:
4463 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4468 /* Flow expiration. */
4470 static int ofproto_dp_max_idle(const struct ofproto *);
4471 static void ofproto_update_stats(struct ofproto *);
4472 static void rule_expire(struct ofproto *, struct rule *);
4473 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4475 /* This function is called periodically by ofproto_run(). Its job is to
4476 * collect updates for the flows that have been installed into the datapath,
4477 * most importantly when they last were used, and then use that information to
4478 * expire flows that have not been used recently.
4480 * Returns the number of milliseconds after which it should be called again. */
4482 ofproto_expire(struct ofproto *ofproto)
4484 struct rule *rule, *next_rule;
4485 struct cls_cursor cursor;
4488 /* Update stats for each flow in the datapath. */
4489 ofproto_update_stats(ofproto);
4491 /* Expire facets that have been idle too long. */
4492 dp_max_idle = ofproto_dp_max_idle(ofproto);
4493 ofproto_expire_facets(ofproto, dp_max_idle);
4495 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4496 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4497 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4498 rule_expire(ofproto, rule);
4501 /* Let the hook know that we're at a stable point: all outstanding data
4502 * in existing flows has been accounted to the account_cb. Thus, the
4503 * hook can now reasonably do operations that depend on having accurate
4504 * flow volume accounting (currently, that's just bond rebalancing). */
4505 if (ofproto->ofhooks->account_checkpoint_cb) {
4506 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4509 return MIN(dp_max_idle, 1000);
4512 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4514 * This function also pushes statistics updates to rules which each facet
4515 * resubmits into. Generally these statistics will be accurate. However, if a
4516 * facet changes the rule it resubmits into at some time in between
4517 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4518 * old rule will be incorrectly attributed to the new rule. This could be
4519 * avoided by calling ofproto_update_stats() whenever rules are created or
4520 * deleted. However, the performance impact of making so many calls to the
4521 * datapath do not justify the benefit of having perfectly accurate statistics.
4524 ofproto_update_stats(struct ofproto *p)
4526 const struct dpif_flow_stats *stats;
4527 struct dpif_flow_dump dump;
4528 const struct nlattr *key;
4531 dpif_flow_dump_start(&dump, p->dpif);
4532 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4533 struct facet *facet;
4536 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4540 odp_flow_key_format(key, key_len, &s);
4541 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4547 facet = facet_find(p, &flow);
4549 if (facet && facet->installed) {
4551 if (stats->n_packets >= facet->dp_packet_count) {
4552 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4554 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4557 if (stats->n_bytes >= facet->dp_byte_count) {
4558 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4560 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4563 facet->dp_packet_count = stats->n_packets;
4564 facet->dp_byte_count = stats->n_bytes;
4566 facet_update_time(p, facet, stats->used);
4567 facet_account(p, facet, stats->n_bytes);
4568 facet_push_stats(p, facet);
4570 /* There's a flow in the datapath that we know nothing about.
4572 COVERAGE_INC(ofproto_unexpected_rule);
4573 dpif_flow_del(p->dpif, key, key_len, NULL);
4576 dpif_flow_dump_done(&dump);
4579 /* Calculates and returns the number of milliseconds of idle time after which
4580 * facets should expire from the datapath and we should fold their statistics
4581 * into their parent rules in userspace. */
4583 ofproto_dp_max_idle(const struct ofproto *ofproto)
4586 * Idle time histogram.
4588 * Most of the time a switch has a relatively small number of facets. When
4589 * this is the case we might as well keep statistics for all of them in
4590 * userspace and to cache them in the kernel datapath for performance as
4593 * As the number of facets increases, the memory required to maintain
4594 * statistics about them in userspace and in the kernel becomes
4595 * significant. However, with a large number of facets it is likely that
4596 * only a few of them are "heavy hitters" that consume a large amount of
4597 * bandwidth. At this point, only heavy hitters are worth caching in the
4598 * kernel and maintaining in userspaces; other facets we can discard.
4600 * The technique used to compute the idle time is to build a histogram with
4601 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4602 * that is installed in the kernel gets dropped in the appropriate bucket.
4603 * After the histogram has been built, we compute the cutoff so that only
4604 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4605 * cached. At least the most-recently-used bucket of facets is kept, so
4606 * actually an arbitrary number of facets can be kept in any given
4607 * expiration run (though the next run will delete most of those unless
4608 * they receive additional data).
4610 * This requires a second pass through the facets, in addition to the pass
4611 * made by ofproto_update_stats(), because the former function never looks
4612 * at uninstallable facets.
4614 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4615 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4616 int buckets[N_BUCKETS] = { 0 };
4617 struct facet *facet;
4622 total = hmap_count(&ofproto->facets);
4623 if (total <= 1000) {
4624 return N_BUCKETS * BUCKET_WIDTH;
4627 /* Build histogram. */
4629 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4630 long long int idle = now - facet->used;
4631 int bucket = (idle <= 0 ? 0
4632 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4633 : (unsigned int) idle / BUCKET_WIDTH);
4637 /* Find the first bucket whose flows should be expired. */
4638 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4639 if (buckets[bucket]) {
4642 subtotal += buckets[bucket++];
4643 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4648 if (VLOG_IS_DBG_ENABLED()) {
4652 ds_put_cstr(&s, "keep");
4653 for (i = 0; i < N_BUCKETS; i++) {
4655 ds_put_cstr(&s, ", drop");
4658 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4661 VLOG_INFO("%s: %s (msec:count)",
4662 dpif_name(ofproto->dpif), ds_cstr(&s));
4666 return bucket * BUCKET_WIDTH;
4670 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4672 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4673 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4674 struct ofexpired expired;
4676 if (facet->installed) {
4677 struct dpif_flow_stats stats;
4679 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4681 facet_update_stats(ofproto, facet, &stats);
4684 expired.flow = facet->flow;
4685 expired.packet_count = facet->packet_count;
4686 expired.byte_count = facet->byte_count;
4687 expired.used = facet->used;
4688 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4693 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4695 long long int cutoff = time_msec() - dp_max_idle;
4696 struct facet *facet, *next_facet;
4698 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4699 facet_active_timeout(ofproto, facet);
4700 if (facet->used < cutoff) {
4701 facet_remove(ofproto, facet);
4706 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4707 * then delete it entirely. */
4709 rule_expire(struct ofproto *ofproto, struct rule *rule)
4711 struct facet *facet, *next_facet;
4715 /* Has 'rule' expired? */
4717 if (rule->hard_timeout
4718 && now > rule->created + rule->hard_timeout * 1000) {
4719 reason = OFPRR_HARD_TIMEOUT;
4720 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4721 && now >rule->used + rule->idle_timeout * 1000) {
4722 reason = OFPRR_IDLE_TIMEOUT;
4727 COVERAGE_INC(ofproto_expired);
4729 /* Update stats. (This is a no-op if the rule expired due to an idle
4730 * timeout, because that only happens when the rule has no facets left.) */
4731 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4732 facet_remove(ofproto, facet);
4735 /* Get rid of the rule. */
4736 if (!rule_is_hidden(rule)) {
4737 rule_send_removed(ofproto, rule, reason);
4739 rule_remove(ofproto, rule);
4742 static struct ofpbuf *
4743 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4746 struct ofp_flow_removed *ofr;
4749 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4750 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4751 rule->flow_cookie, &ofr->cookie);
4752 ofr->priority = htons(rule->cr.priority);
4753 ofr->reason = reason;
4754 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4755 ofr->idle_timeout = htons(rule->idle_timeout);
4756 ofr->packet_count = htonll(rule->packet_count);
4757 ofr->byte_count = htonll(rule->byte_count);
4762 static struct ofpbuf *
4763 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4765 struct nx_flow_removed *nfr;
4769 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4770 match_len = nx_put_match(buf, &rule->cr);
4773 nfr->cookie = rule->flow_cookie;
4774 nfr->priority = htons(rule->cr.priority);
4775 nfr->reason = reason;
4776 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4777 nfr->idle_timeout = htons(rule->idle_timeout);
4778 nfr->match_len = htons(match_len);
4779 nfr->packet_count = htonll(rule->packet_count);
4780 nfr->byte_count = htonll(rule->byte_count);
4786 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4788 struct ofconn *ofconn;
4790 if (!rule->send_flow_removed) {
4794 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4797 if (!rconn_is_connected(ofconn->rconn)
4798 || !ofconn_receives_async_msgs(ofconn)) {
4802 msg = (ofconn->flow_format == NXFF_NXM
4803 ? compose_nx_flow_removed(rule, reason)
4804 : compose_ofp_flow_removed(ofconn, rule, reason));
4806 /* Account flow expirations under ofconn->reply_counter, the counter
4807 * for replies to OpenFlow requests. That works because preventing
4808 * OpenFlow requests from being processed also prevents new flows from
4809 * being added (and expiring). (It also prevents processing OpenFlow
4810 * requests that would not add new flows, so it is imperfect.) */
4811 queue_tx(msg, ofconn, ofconn->reply_counter);
4815 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4816 * The returned statistics include statistics for all of 'rule''s facets. */
4818 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4821 struct facet *facet;
4823 /* Start from historical data for 'rule' itself that are no longer tracked
4824 * in facets. This counts, for example, facets that have expired. */
4825 p = rule->packet_count;
4826 b = rule->byte_count;
4828 /* Add any statistics that are tracked by facets. This includes
4829 * statistical data recently updated by ofproto_update_stats() as well as
4830 * stats for packets that were executed "by hand" via dpif_execute(). */
4831 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4832 p += facet->packet_count;
4833 b += facet->byte_count;
4840 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4842 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4844 struct ofconn *ofconn = ofconn_;
4846 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4847 ofconn->packet_in_counter, 100);
4850 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4851 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4852 * scheduler for sending.
4854 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4855 * Otherwise, ownership is transferred to this function. */
4857 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4858 const struct flow *flow, bool clone)
4860 enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
4861 struct ofproto *ofproto = ofconn->ofproto;
4862 struct ofp_packet_in *opi;
4863 int total_len, send_len;
4864 struct ofpbuf *packet;
4868 /* Get OpenFlow buffer_id. */
4869 if (upcall->type == DPIF_UC_ACTION) {
4870 buffer_id = UINT32_MAX;
4871 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4872 buffer_id = pktbuf_get_null();
4873 } else if (!ofconn->pktbuf) {
4874 buffer_id = UINT32_MAX;
4876 buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
4879 /* Figure out how much of the packet to send. */
4880 total_len = send_len = upcall->packet->size;
4881 if (buffer_id != UINT32_MAX) {
4882 send_len = MIN(send_len, ofconn->miss_send_len);
4884 if (upcall->type == DPIF_UC_ACTION) {
4885 send_len = MIN(send_len, upcall->userdata);
4888 /* Copy or steal buffer for OFPT_PACKET_IN. */
4890 packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
4891 send_len, OPI_SIZE);
4893 packet = upcall->packet;
4894 packet->size = send_len;
4897 /* Add OFPT_PACKET_IN. */
4898 opi = ofpbuf_push_zeros(packet, OPI_SIZE);
4899 opi->header.version = OFP_VERSION;
4900 opi->header.type = OFPT_PACKET_IN;
4901 opi->total_len = htons(total_len);
4902 opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
4903 opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4904 opi->buffer_id = htonl(buffer_id);
4905 update_openflow_length(packet);
4907 /* Hand over to packet scheduler. It might immediately call into
4908 * do_send_packet_in() or it might buffer it for a while (until a later
4909 * call to pinsched_run()). */
4910 idx = upcall->type == DPIF_UC_MISS ? 0 : 1;
4911 pinsched_send(ofconn->schedulers[idx], flow->in_port,
4912 packet, do_send_packet_in, ofconn);
4915 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4916 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4917 * their individual configurations.
4919 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4920 * Otherwise, ownership is transferred to this function. */
4922 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4923 const struct flow *flow, bool clone)
4925 struct ofconn *ofconn, *prev;
4928 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4929 if (ofconn_receives_async_msgs(ofconn)) {
4931 schedule_packet_in(prev, upcall, flow, true);
4937 schedule_packet_in(prev, upcall, flow, clone);
4938 } else if (!clone) {
4939 ofpbuf_delete(upcall->packet);
4944 pick_datapath_id(const struct ofproto *ofproto)
4946 const struct ofport *port;
4948 port = get_port(ofproto, ODPP_LOCAL);
4950 uint8_t ea[ETH_ADDR_LEN];
4953 error = netdev_get_etheraddr(port->netdev, ea);
4955 return eth_addr_to_uint64(ea);
4957 VLOG_WARN("could not get MAC address for %s (%s)",
4958 netdev_get_name(port->netdev), strerror(error));
4960 return ofproto->fallback_dpid;
4964 pick_fallback_dpid(void)
4966 uint8_t ea[ETH_ADDR_LEN];
4967 eth_addr_nicira_random(ea);
4968 return eth_addr_to_uint64(ea);
4972 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4973 void *aux OVS_UNUSED)
4975 const struct shash_node *node;
4979 SHASH_FOR_EACH (node, &all_ofprotos) {
4980 ds_put_format(&results, "%s\n", node->name);
4982 unixctl_command_reply(conn, 200, ds_cstr(&results));
4983 ds_destroy(&results);
4986 struct ofproto_trace {
4987 struct action_xlate_ctx ctx;
4993 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4995 ds_put_char_multiple(result, '\t', level);
4997 ds_put_cstr(result, "No match\n");
5001 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5002 ntohll(rule->flow_cookie));
5003 cls_rule_format(&rule->cr, result);
5004 ds_put_char(result, '\n');
5006 ds_put_char_multiple(result, '\t', level);
5007 ds_put_cstr(result, "OpenFlow ");
5008 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5009 rule->n_actions * sizeof *rule->actions);
5010 ds_put_char(result, '\n');
5014 trace_format_flow(struct ds *result, int level, const char *title,
5015 struct ofproto_trace *trace)
5017 ds_put_char_multiple(result, '\t', level);
5018 ds_put_format(result, "%s: ", title);
5019 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5020 ds_put_cstr(result, "unchanged");
5022 flow_format(result, &trace->ctx.flow);
5023 trace->flow = trace->ctx.flow;
5025 ds_put_char(result, '\n');
5029 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5031 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5032 struct ds *result = trace->result;
5034 ds_put_char(result, '\n');
5035 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5036 trace_format_rule(result, ctx->recurse + 1, rule);
5040 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5041 void *aux OVS_UNUSED)
5043 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5044 char *args = xstrdup(args_);
5045 char *save_ptr = NULL;
5046 struct ofproto *ofproto;
5047 struct ofpbuf packet;
5055 ofpbuf_init(&packet, strlen(args) / 2);
5058 dpname = strtok_r(args, " ", &save_ptr);
5059 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5060 in_port_s = strtok_r(NULL, " ", &save_ptr);
5061 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5062 if (!dpname || !in_port_s || !packet_s) {
5063 unixctl_command_reply(conn, 501, "Bad command syntax");
5067 ofproto = shash_find_data(&all_ofprotos, dpname);
5069 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5074 tun_id = htonll(strtoull(tun_id_s, NULL, 10));
5075 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5077 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5078 packet_s += strspn(packet_s, " ");
5079 if (*packet_s != '\0') {
5080 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5083 if (packet.size < ETH_HEADER_LEN) {
5084 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5088 ds_put_cstr(&result, "Packet: ");
5089 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5090 ds_put_cstr(&result, s);
5093 flow_extract(&packet, tun_id, in_port, &flow);
5094 ds_put_cstr(&result, "Flow: ");
5095 flow_format(&result, &flow);
5096 ds_put_char(&result, '\n');
5098 rule = rule_lookup(ofproto, &flow);
5099 trace_format_rule(&result, 0, rule);
5101 struct ofproto_trace trace;
5102 struct ofpbuf *odp_actions;
5104 trace.result = &result;
5106 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5107 trace.ctx.resubmit_hook = trace_resubmit;
5108 odp_actions = xlate_actions(&trace.ctx,
5109 rule->actions, rule->n_actions);
5111 ds_put_char(&result, '\n');
5112 trace_format_flow(&result, 0, "Final flow", &trace);
5113 ds_put_cstr(&result, "Datapath actions: ");
5114 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5115 ofpbuf_delete(odp_actions);
5118 unixctl_command_reply(conn, 200, ds_cstr(&result));
5121 ds_destroy(&result);
5122 ofpbuf_uninit(&packet);
5127 ofproto_unixctl_init(void)
5129 static bool registered;
5135 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5136 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5140 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5141 struct ofpbuf *odp_actions, tag_type *tags,
5142 uint16_t *nf_output_iface, void *ofproto_)
5144 struct ofproto *ofproto = ofproto_;
5147 /* Drop frames for reserved multicast addresses. */
5148 if (eth_addr_is_reserved(flow->dl_dst)) {
5152 /* Learn source MAC (but don't try to learn from revalidation). */
5153 if (packet != NULL) {
5154 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5156 GRAT_ARP_LOCK_NONE);
5158 /* The log messages here could actually be useful in debugging,
5159 * so keep the rate limit relatively high. */
5160 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5161 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5162 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5163 ofproto_revalidate(ofproto, rev_tag);
5167 /* Determine output port. */
5168 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5171 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5172 nf_output_iface, odp_actions);
5173 } else if (out_port != flow->in_port) {
5174 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5175 *nf_output_iface = out_port;
5183 static const struct ofhooks default_ofhooks = {
5184 default_normal_ofhook_cb,