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 /* Opens and returns a netdev for 'dpif_port', or a null pointer if the netdev
1519 * cannot be opened. On success, also fills in 'opp', in *HOST* byte order. */
1520 static struct netdev *
1521 ofport_open(const struct dpif_port *dpif_port, struct ofp_phy_port *opp)
1523 struct netdev_options netdev_options;
1524 enum netdev_flags flags;
1525 struct netdev *netdev;
1528 memset(&netdev_options, 0, sizeof netdev_options);
1529 netdev_options.name = dpif_port->name;
1530 netdev_options.type = dpif_port->type;
1531 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1533 error = netdev_open(&netdev_options, &netdev);
1535 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1536 "cannot be opened (%s)",
1537 dpif_port->name, dpif_port->port_no,
1538 dpif_port->name, strerror(error));
1542 netdev_get_flags(netdev, &flags);
1544 opp->port_no = odp_port_to_ofp_port(dpif_port->port_no);
1545 netdev_get_etheraddr(netdev, opp->hw_addr);
1546 ovs_strzcpy(opp->name, dpif_port->name, sizeof opp->name);
1547 opp->config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1548 opp->state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1549 netdev_get_features(netdev, &opp->curr, &opp->advertised,
1550 &opp->supported, &opp->peer);
1555 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1557 if (get_port(p, dpif_port->port_no)) {
1558 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1559 dpif_port->port_no);
1561 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1562 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1570 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1571 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1574 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1576 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1577 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1578 && a->state == b->state
1579 && !((a->config ^ b->config) & OFPPC_PORT_DOWN)
1580 && a->curr == b->curr
1581 && a->advertised == b->advertised
1582 && a->supported == b->supported
1583 && a->peer == b->peer);
1587 send_port_status(struct ofproto *p, const struct ofp_phy_port *opp,
1590 /* XXX Should limit the number of queued port status change messages. */
1591 struct ofconn *ofconn;
1592 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1593 struct ofp_port_status *ops;
1596 /* Primary controllers, even slaves, should always get port status
1597 updates. Otherwise obey ofconn_receives_async_msgs(). */
1598 if (ofconn->type != OFCONN_PRIMARY
1599 && !ofconn_receives_async_msgs(ofconn)) {
1603 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1604 ops->reason = reason;
1606 hton_ofp_phy_port(&ops->desc);
1607 queue_tx(b, ofconn, NULL);
1611 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1612 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1613 * one with the same name or port number). */
1615 ofport_install(struct ofproto *p,
1616 struct netdev *netdev, const struct ofp_phy_port *opp)
1618 const char *netdev_name = netdev_get_name(netdev);
1619 struct ofport *ofport;
1621 send_port_status(p, opp, OFPPR_ADD);
1623 /* Create ofport. */
1624 ofport = xmalloc(sizeof *ofport);
1625 ofport->netdev = netdev;
1627 ofport->odp_port = ofp_port_to_odp_port(opp->port_no);
1629 /* Add port to 'p'. */
1630 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1631 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1632 shash_add(&p->port_by_name, netdev_name, ofport);
1634 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1638 /* Removes 'ofport' from 'p' and destroys it. */
1640 ofport_remove(struct ofproto *p, struct ofport *ofport)
1642 send_port_status(p, &ofport->opp, OFPPR_DELETE);
1644 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1645 hmap_remove(&p->ports, &ofport->hmap_node);
1646 shash_delete(&p->port_by_name,
1647 shash_find(&p->port_by_name,
1648 netdev_get_name(ofport->netdev)));
1650 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1653 ofport_free(ofport);
1656 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1659 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1661 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1663 ofport_remove(ofproto, port);
1667 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1669 * Does not handle a name or port number change. The caller must implement
1670 * such a change as a delete followed by an add. */
1672 ofport_modified(struct ofproto *ofproto, struct ofport *port,
1673 struct netdev *netdev, struct ofp_phy_port *opp)
1675 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1676 port->opp.config = ((port->opp.config & ~OFPPC_PORT_DOWN)
1677 | (opp->config & OFPPC_PORT_DOWN));
1678 port->opp.state = opp->state;
1679 port->opp.curr = opp->curr;
1680 port->opp.advertised = opp->advertised;
1681 port->opp.supported = opp->supported;
1682 port->opp.peer = opp->peer;
1684 netdev_close(port->netdev);
1685 port->netdev = netdev;
1687 send_port_status(ofproto, &port->opp, OFPPR_MODIFY);
1691 ofport_free(struct ofport *ofport)
1694 netdev_close(ofport->netdev);
1699 static struct ofport *
1700 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1702 struct ofport *port;
1704 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1705 hash_int(odp_port, 0), &ofproto->ports) {
1706 if (port->odp_port == odp_port) {
1714 update_port(struct ofproto *ofproto, const char *name)
1716 struct dpif_port dpif_port;
1717 struct ofp_phy_port opp;
1718 struct netdev *netdev;
1719 struct ofport *port;
1721 COVERAGE_INC(ofproto_update_port);
1723 /* Fetch 'name''s location and properties from the datapath. */
1724 netdev = (!dpif_port_query_by_name(ofproto->dpif, name, &dpif_port)
1725 ? ofport_open(&dpif_port, &opp)
1728 port = get_port(ofproto, dpif_port.port_no);
1729 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1730 /* 'name' hasn't changed location. Any properties changed? */
1731 if (!ofport_equal(&port->opp, &opp)) {
1732 ofport_modified(ofproto, port, netdev, &opp);
1734 netdev_close(netdev);
1737 /* If 'port' is nonnull then its name differs from 'name' and thus
1738 * we should delete it. If we think there's a port named 'name'
1739 * then its port number must be wrong now so delete it too. */
1741 ofport_remove(ofproto, port);
1743 ofport_remove_with_name(ofproto, name);
1744 ofport_install(ofproto, netdev, &opp);
1747 /* Any port named 'name' is gone now. */
1748 ofport_remove_with_name(ofproto, name);
1750 dpif_port_destroy(&dpif_port);
1754 init_ports(struct ofproto *p)
1756 struct dpif_port_dump dump;
1757 struct dpif_port dpif_port;
1759 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1760 if (!ofport_conflicts(p, &dpif_port)) {
1761 struct ofp_phy_port opp;
1762 struct netdev *netdev;
1764 netdev = ofport_open(&dpif_port, &opp);
1766 ofport_install(p, netdev, &opp);
1774 static struct ofconn *
1775 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1777 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1778 ofconn->ofproto = p;
1779 list_push_back(&p->all_conns, &ofconn->node);
1780 ofconn->rconn = rconn;
1781 ofconn->type = type;
1782 ofconn->flow_format = NXFF_OPENFLOW10;
1783 ofconn->role = NX_ROLE_OTHER;
1784 ofconn->packet_in_counter = rconn_packet_counter_create ();
1785 ofconn->pktbuf = NULL;
1786 ofconn->miss_send_len = 0;
1787 ofconn->reply_counter = rconn_packet_counter_create ();
1792 ofconn_destroy(struct ofconn *ofconn)
1794 if (ofconn->type == OFCONN_PRIMARY) {
1795 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1798 list_remove(&ofconn->node);
1799 rconn_destroy(ofconn->rconn);
1800 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1801 rconn_packet_counter_destroy(ofconn->reply_counter);
1802 pktbuf_destroy(ofconn->pktbuf);
1807 ofconn_run(struct ofconn *ofconn)
1809 struct ofproto *p = ofconn->ofproto;
1813 for (i = 0; i < N_SCHEDULERS; i++) {
1814 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1817 rconn_run(ofconn->rconn);
1819 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1820 /* Limit the number of iterations to prevent other tasks from
1822 for (iteration = 0; iteration < 50; iteration++) {
1823 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1828 fail_open_maybe_recover(p->fail_open);
1830 handle_openflow(ofconn, of_msg);
1831 ofpbuf_delete(of_msg);
1835 if (!rconn_is_alive(ofconn->rconn)) {
1836 ofconn_destroy(ofconn);
1841 ofconn_wait(struct ofconn *ofconn)
1845 for (i = 0; i < N_SCHEDULERS; i++) {
1846 pinsched_wait(ofconn->schedulers[i]);
1848 rconn_run_wait(ofconn->rconn);
1849 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1850 rconn_recv_wait(ofconn->rconn);
1852 COVERAGE_INC(ofproto_ofconn_stuck);
1856 /* Returns true if 'ofconn' should receive asynchronous messages. */
1858 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1860 if (ofconn->type == OFCONN_PRIMARY) {
1861 /* Primary controllers always get asynchronous messages unless they
1862 * have configured themselves as "slaves". */
1863 return ofconn->role != NX_ROLE_SLAVE;
1865 /* Service connections don't get asynchronous messages unless they have
1866 * explicitly asked for them by setting a nonzero miss send length. */
1867 return ofconn->miss_send_len > 0;
1871 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1872 * and 'target', suitable for use in log messages for identifying the
1875 * The name is dynamically allocated. The caller should free it (with free())
1876 * when it is no longer needed. */
1878 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1880 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1884 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1888 for (i = 0; i < N_SCHEDULERS; i++) {
1889 struct pinsched **s = &ofconn->schedulers[i];
1893 *s = pinsched_create(rate, burst);
1895 pinsched_set_limits(*s, rate, burst);
1898 pinsched_destroy(*s);
1905 ofservice_reconfigure(struct ofservice *ofservice,
1906 const struct ofproto_controller *c)
1908 ofservice->probe_interval = c->probe_interval;
1909 ofservice->rate_limit = c->rate_limit;
1910 ofservice->burst_limit = c->burst_limit;
1913 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1914 * positive errno value. */
1916 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1918 struct ofservice *ofservice;
1919 struct pvconn *pvconn;
1922 error = pvconn_open(c->target, &pvconn);
1927 ofservice = xzalloc(sizeof *ofservice);
1928 hmap_insert(&ofproto->services, &ofservice->node,
1929 hash_string(c->target, 0));
1930 ofservice->pvconn = pvconn;
1932 ofservice_reconfigure(ofservice, c);
1938 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1940 hmap_remove(&ofproto->services, &ofservice->node);
1941 pvconn_close(ofservice->pvconn);
1945 /* Finds and returns the ofservice within 'ofproto' that has the given
1946 * 'target', or a null pointer if none exists. */
1947 static struct ofservice *
1948 ofservice_lookup(struct ofproto *ofproto, const char *target)
1950 struct ofservice *ofservice;
1952 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1953 &ofproto->services) {
1954 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1961 /* Returns true if 'rule' should be hidden from the controller.
1963 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1964 * (e.g. by in-band control) and are intentionally hidden from the
1967 rule_is_hidden(const struct rule *rule)
1969 return rule->cr.priority > UINT16_MAX;
1972 /* Creates and returns a new rule initialized as specified.
1974 * The caller is responsible for inserting the rule into the classifier (with
1975 * rule_insert()). */
1976 static struct rule *
1977 rule_create(const struct cls_rule *cls_rule,
1978 const union ofp_action *actions, size_t n_actions,
1979 uint16_t idle_timeout, uint16_t hard_timeout,
1980 ovs_be64 flow_cookie, bool send_flow_removed)
1982 struct rule *rule = xzalloc(sizeof *rule);
1983 rule->cr = *cls_rule;
1984 rule->idle_timeout = idle_timeout;
1985 rule->hard_timeout = hard_timeout;
1986 rule->flow_cookie = flow_cookie;
1987 rule->used = rule->created = time_msec();
1988 rule->send_flow_removed = send_flow_removed;
1989 list_init(&rule->facets);
1990 if (n_actions > 0) {
1991 rule->n_actions = n_actions;
1992 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1998 static struct rule *
1999 rule_from_cls_rule(const struct cls_rule *cls_rule)
2001 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2005 rule_free(struct rule *rule)
2007 free(rule->actions);
2011 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2012 * destroying any that no longer has a rule (which is probably all of them).
2014 * The caller must have already removed 'rule' from the classifier. */
2016 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2018 struct facet *facet, *next_facet;
2019 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2020 facet_revalidate(ofproto, facet);
2025 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2026 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2029 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2031 const union ofp_action *oa;
2032 struct actions_iterator i;
2034 if (out_port == htons(OFPP_NONE)) {
2037 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2038 oa = actions_next(&i)) {
2039 if (action_outputs_to_port(oa, out_port)) {
2046 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2047 * 'packet', which arrived on 'in_port'.
2049 * Takes ownership of 'packet'. */
2051 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2052 const struct nlattr *odp_actions, size_t actions_len,
2053 struct ofpbuf *packet)
2055 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2056 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2057 /* As an optimization, avoid a round-trip from userspace to kernel to
2058 * userspace. This also avoids possibly filling up kernel packet
2059 * buffers along the way. */
2060 struct dpif_upcall upcall;
2062 upcall.type = DPIF_UC_ACTION;
2063 upcall.packet = packet;
2066 upcall.userdata = nl_attr_get_u64(odp_actions);
2067 upcall.sample_pool = 0;
2068 upcall.actions = NULL;
2069 upcall.actions_len = 0;
2071 send_packet_in(ofproto, &upcall, flow, false);
2077 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2078 ofpbuf_delete(packet);
2083 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2084 * statistics appropriately. 'packet' must have at least sizeof(struct
2085 * ofp_packet_in) bytes of headroom.
2087 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2088 * applying flow_extract() to 'packet' would yield the same flow as
2091 * 'facet' must have accurately composed ODP actions; that is, it must not be
2092 * in need of revalidation.
2094 * Takes ownership of 'packet'. */
2096 facet_execute(struct ofproto *ofproto, struct facet *facet,
2097 struct ofpbuf *packet)
2099 struct dpif_flow_stats stats;
2101 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2103 flow_extract_stats(&facet->flow, packet, &stats);
2104 stats.used = time_msec();
2105 if (execute_odp_actions(ofproto, &facet->flow,
2106 facet->actions, facet->actions_len, packet)) {
2107 facet_update_stats(ofproto, facet, &stats);
2111 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2112 * statistics (or the statistics for one of its facets) appropriately.
2113 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2115 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2116 * with statistics for 'packet' either way.
2118 * Takes ownership of 'packet'. */
2120 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2121 struct ofpbuf *packet)
2123 struct action_xlate_ctx ctx;
2124 struct ofpbuf *odp_actions;
2125 struct facet *facet;
2129 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2131 flow_extract(packet, 0, in_port, &flow);
2133 /* First look for a related facet. If we find one, account it to that. */
2134 facet = facet_lookup_valid(ofproto, &flow);
2135 if (facet && facet->rule == rule) {
2136 facet_execute(ofproto, facet, packet);
2140 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2141 * create a new facet for it and use that. */
2142 if (rule_lookup(ofproto, &flow) == rule) {
2143 facet = facet_create(ofproto, rule, &flow, packet);
2144 facet_execute(ofproto, facet, packet);
2145 facet_install(ofproto, facet, true);
2149 /* We can't account anything to a facet. If we were to try, then that
2150 * facet would have a non-matching rule, busting our invariants. */
2151 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2152 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2153 size = packet->size;
2154 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2155 odp_actions->size, packet)) {
2156 rule->used = time_msec();
2157 rule->packet_count++;
2158 rule->byte_count += size;
2159 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2161 ofpbuf_delete(odp_actions);
2164 /* Inserts 'rule' into 'p''s flow table. */
2166 rule_insert(struct ofproto *p, struct rule *rule)
2168 struct rule *displaced_rule;
2170 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2171 if (displaced_rule) {
2172 rule_destroy(p, displaced_rule);
2174 p->need_revalidate = true;
2177 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2178 * 'flow' and an example 'packet' within that flow.
2180 * The caller must already have determined that no facet with an identical
2181 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2182 * 'ofproto''s classifier table. */
2183 static struct facet *
2184 facet_create(struct ofproto *ofproto, struct rule *rule,
2185 const struct flow *flow, const struct ofpbuf *packet)
2187 struct facet *facet;
2189 facet = xzalloc(sizeof *facet);
2190 facet->used = time_msec();
2191 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2192 list_push_back(&rule->facets, &facet->list_node);
2194 facet->flow = *flow;
2195 netflow_flow_init(&facet->nf_flow);
2196 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2198 facet_make_actions(ofproto, facet, packet);
2204 facet_free(struct facet *facet)
2206 free(facet->actions);
2210 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2212 * - Removes 'rule' from the classifier.
2214 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2215 * destroys them), via rule_destroy().
2218 rule_remove(struct ofproto *ofproto, struct rule *rule)
2220 COVERAGE_INC(ofproto_del_rule);
2221 ofproto->need_revalidate = true;
2222 classifier_remove(&ofproto->cls, &rule->cr);
2223 rule_destroy(ofproto, rule);
2226 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2228 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2229 * rule's statistics, via facet_uninstall().
2231 * - Removes 'facet' from its rule and from ofproto->facets.
2234 facet_remove(struct ofproto *ofproto, struct facet *facet)
2236 facet_uninstall(ofproto, facet);
2237 facet_flush_stats(ofproto, facet);
2238 hmap_remove(&ofproto->facets, &facet->hmap_node);
2239 list_remove(&facet->list_node);
2243 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2245 facet_make_actions(struct ofproto *p, struct facet *facet,
2246 const struct ofpbuf *packet)
2248 const struct rule *rule = facet->rule;
2249 struct ofpbuf *odp_actions;
2250 struct action_xlate_ctx ctx;
2252 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2253 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2254 facet->tags = ctx.tags;
2255 facet->may_install = ctx.may_set_up_flow;
2256 facet->nf_flow.output_iface = ctx.nf_output_iface;
2258 if (facet->actions_len != odp_actions->size
2259 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2260 free(facet->actions);
2261 facet->actions_len = odp_actions->size;
2262 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2265 ofpbuf_delete(odp_actions);
2269 facet_put__(struct ofproto *ofproto, struct facet *facet,
2270 const struct nlattr *actions, size_t actions_len,
2271 struct dpif_flow_stats *stats)
2273 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2274 enum dpif_flow_put_flags flags;
2277 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2279 flags |= DPIF_FP_ZERO_STATS;
2280 facet->dp_packet_count = 0;
2281 facet->dp_byte_count = 0;
2284 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2285 odp_flow_key_from_flow(&key, &facet->flow);
2286 assert(key.base == keybuf);
2288 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2289 actions, actions_len, stats);
2292 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2293 * 'zero_stats' is true, clears any existing statistics from the datapath for
2296 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2298 struct dpif_flow_stats stats;
2300 if (facet->may_install
2301 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2302 zero_stats ? &stats : NULL)) {
2303 facet->installed = true;
2307 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2308 * to the accounting hook function in the ofhooks structure. */
2310 facet_account(struct ofproto *ofproto,
2311 struct facet *facet, uint64_t extra_bytes)
2313 uint64_t total_bytes = facet->byte_count + extra_bytes;
2315 if (ofproto->ofhooks->account_flow_cb
2316 && total_bytes > facet->accounted_bytes)
2318 ofproto->ofhooks->account_flow_cb(
2319 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2320 total_bytes - facet->accounted_bytes, ofproto->aux);
2321 facet->accounted_bytes = total_bytes;
2325 /* If 'rule' is installed in the datapath, uninstalls it. */
2327 facet_uninstall(struct ofproto *p, struct facet *facet)
2329 if (facet->installed) {
2330 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2331 struct dpif_flow_stats stats;
2334 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2335 odp_flow_key_from_flow(&key, &facet->flow);
2336 assert(key.base == keybuf);
2338 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2339 facet_update_stats(p, facet, &stats);
2341 facet->installed = false;
2342 facet->dp_packet_count = 0;
2343 facet->dp_byte_count = 0;
2345 assert(facet->dp_packet_count == 0);
2346 assert(facet->dp_byte_count == 0);
2350 /* Returns true if the only action for 'facet' is to send to the controller.
2351 * (We don't report NetFlow expiration messages for such facets because they
2352 * are just part of the control logic for the network, not real traffic). */
2354 facet_is_controller_flow(struct facet *facet)
2357 && facet->rule->n_actions == 1
2358 && action_outputs_to_port(&facet->rule->actions[0],
2359 htons(OFPP_CONTROLLER)));
2362 /* Folds all of 'facet''s statistics into its rule. Also updates the
2363 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2364 * 'facet''s statistics in the datapath should have been zeroed and folded into
2365 * its packet and byte counts before this function is called. */
2367 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2369 assert(!facet->dp_byte_count);
2370 assert(!facet->dp_packet_count);
2372 facet_push_stats(ofproto, facet);
2373 facet_account(ofproto, facet, 0);
2375 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2376 struct ofexpired expired;
2377 expired.flow = facet->flow;
2378 expired.packet_count = facet->packet_count;
2379 expired.byte_count = facet->byte_count;
2380 expired.used = facet->used;
2381 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2384 facet->rule->packet_count += facet->packet_count;
2385 facet->rule->byte_count += facet->byte_count;
2387 /* Reset counters to prevent double counting if 'facet' ever gets
2389 facet->packet_count = 0;
2390 facet->byte_count = 0;
2391 facet->rs_packet_count = 0;
2392 facet->rs_byte_count = 0;
2393 facet->accounted_bytes = 0;
2395 netflow_flow_clear(&facet->nf_flow);
2398 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2399 * Returns it if found, otherwise a null pointer.
2401 * The returned facet might need revalidation; use facet_lookup_valid()
2402 * instead if that is important. */
2403 static struct facet *
2404 facet_find(struct ofproto *ofproto, const struct flow *flow)
2406 struct facet *facet;
2408 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2410 if (flow_equal(flow, &facet->flow)) {
2418 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2419 * Returns it if found, otherwise a null pointer.
2421 * The returned facet is guaranteed to be valid. */
2422 static struct facet *
2423 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2425 struct facet *facet = facet_find(ofproto, flow);
2427 /* The facet we found might not be valid, since we could be in need of
2428 * revalidation. If it is not valid, don't return it. */
2430 && ofproto->need_revalidate
2431 && !facet_revalidate(ofproto, facet)) {
2432 COVERAGE_INC(ofproto_invalidated);
2439 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2441 * - If the rule found is different from 'facet''s current rule, moves
2442 * 'facet' to the new rule and recompiles its actions.
2444 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2445 * where it is and recompiles its actions anyway.
2447 * - If there is none, destroys 'facet'.
2449 * Returns true if 'facet' still exists, false if it has been destroyed. */
2451 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2453 struct action_xlate_ctx ctx;
2454 struct ofpbuf *odp_actions;
2455 struct rule *new_rule;
2456 bool actions_changed;
2458 COVERAGE_INC(facet_revalidate);
2460 /* Determine the new rule. */
2461 new_rule = rule_lookup(ofproto, &facet->flow);
2463 /* No new rule, so delete the facet. */
2464 facet_remove(ofproto, facet);
2468 /* Calculate new ODP actions.
2470 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2471 * emit a NetFlow expiration and, if so, we need to have the old state
2472 * around to properly compose it. */
2473 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2474 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2475 actions_changed = (facet->actions_len != odp_actions->size
2476 || memcmp(facet->actions, odp_actions->data,
2477 facet->actions_len));
2479 /* If the ODP actions changed or the installability changed, then we need
2480 * to talk to the datapath. */
2481 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2482 if (ctx.may_set_up_flow) {
2483 struct dpif_flow_stats stats;
2485 facet_put__(ofproto, facet,
2486 odp_actions->data, odp_actions->size, &stats);
2487 facet_update_stats(ofproto, facet, &stats);
2489 facet_uninstall(ofproto, facet);
2492 /* The datapath flow is gone or has zeroed stats, so push stats out of
2493 * 'facet' into 'rule'. */
2494 facet_flush_stats(ofproto, facet);
2497 /* Update 'facet' now that we've taken care of all the old state. */
2498 facet->tags = ctx.tags;
2499 facet->nf_flow.output_iface = ctx.nf_output_iface;
2500 facet->may_install = ctx.may_set_up_flow;
2501 if (actions_changed) {
2502 free(facet->actions);
2503 facet->actions_len = odp_actions->size;
2504 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2506 if (facet->rule != new_rule) {
2507 COVERAGE_INC(facet_changed_rule);
2508 list_remove(&facet->list_node);
2509 list_push_back(&new_rule->facets, &facet->list_node);
2510 facet->rule = new_rule;
2511 facet->used = new_rule->created;
2512 facet->rs_used = facet->used;
2515 ofpbuf_delete(odp_actions);
2521 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2522 struct rconn_packet_counter *counter)
2524 update_openflow_length(msg);
2525 if (rconn_send(ofconn->rconn, msg, counter)) {
2531 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2534 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2536 COVERAGE_INC(ofproto_error);
2537 queue_tx(buf, ofconn, ofconn->reply_counter);
2542 hton_ofp_phy_port(struct ofp_phy_port *opp)
2544 opp->port_no = htons(opp->port_no);
2545 opp->config = htonl(opp->config);
2546 opp->state = htonl(opp->state);
2547 opp->curr = htonl(opp->curr);
2548 opp->advertised = htonl(opp->advertised);
2549 opp->supported = htonl(opp->supported);
2550 opp->peer = htonl(opp->peer);
2554 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2556 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2561 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2563 struct ofp_switch_features *osf;
2565 struct ofport *port;
2567 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2568 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2569 osf->n_buffers = htonl(pktbuf_capacity());
2571 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2572 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2573 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2574 (1u << OFPAT_SET_VLAN_VID) |
2575 (1u << OFPAT_SET_VLAN_PCP) |
2576 (1u << OFPAT_STRIP_VLAN) |
2577 (1u << OFPAT_SET_DL_SRC) |
2578 (1u << OFPAT_SET_DL_DST) |
2579 (1u << OFPAT_SET_NW_SRC) |
2580 (1u << OFPAT_SET_NW_DST) |
2581 (1u << OFPAT_SET_NW_TOS) |
2582 (1u << OFPAT_SET_TP_SRC) |
2583 (1u << OFPAT_SET_TP_DST) |
2584 (1u << OFPAT_ENQUEUE));
2586 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2587 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2590 queue_tx(buf, ofconn, ofconn->reply_counter);
2595 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2598 struct ofp_switch_config *osc;
2602 /* Figure out flags. */
2603 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2604 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2607 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2608 osc->flags = htons(flags);
2609 osc->miss_send_len = htons(ofconn->miss_send_len);
2610 queue_tx(buf, ofconn, ofconn->reply_counter);
2616 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2618 uint16_t flags = ntohs(osc->flags);
2620 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2621 switch (flags & OFPC_FRAG_MASK) {
2622 case OFPC_FRAG_NORMAL:
2623 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2625 case OFPC_FRAG_DROP:
2626 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2629 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2635 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2640 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2641 struct action_xlate_ctx *ctx);
2644 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2646 const struct ofport *ofport = get_port(ctx->ofproto, port);
2649 if (ofport->opp.config & OFPPC_NO_FWD) {
2650 /* Forwarding disabled on port. */
2655 * We don't have an ofport record for this port, but it doesn't hurt to
2656 * allow forwarding to it anyhow. Maybe such a port will appear later
2657 * and we're pre-populating the flow table.
2661 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2662 ctx->nf_output_iface = port;
2665 static struct rule *
2666 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2668 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2672 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2674 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2675 uint16_t old_in_port;
2678 /* Look up a flow with 'in_port' as the input port. Then restore the
2679 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2680 * have surprising behavior). */
2681 old_in_port = ctx->flow.in_port;
2682 ctx->flow.in_port = in_port;
2683 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2684 ctx->flow.in_port = old_in_port;
2686 if (ctx->resubmit_hook) {
2687 ctx->resubmit_hook(ctx, rule);
2692 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2696 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2698 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2699 MAX_RESUBMIT_RECURSION);
2704 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2705 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2707 struct ofport *ofport;
2709 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2710 uint16_t odp_port = ofport->odp_port;
2711 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2712 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2715 *nf_output_iface = NF_OUT_FLOOD;
2719 xlate_output_action__(struct action_xlate_ctx *ctx,
2720 uint16_t port, uint16_t max_len)
2723 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2725 ctx->nf_output_iface = NF_OUT_DROP;
2729 add_output_action(ctx, ctx->flow.in_port);
2732 xlate_table_action(ctx, ctx->flow.in_port);
2735 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2736 ctx->odp_actions, &ctx->tags,
2737 &ctx->nf_output_iface,
2738 ctx->ofproto->aux)) {
2739 COVERAGE_INC(ofproto_uninstallable);
2740 ctx->may_set_up_flow = false;
2744 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2745 &ctx->nf_output_iface, ctx->odp_actions);
2748 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2749 &ctx->nf_output_iface, ctx->odp_actions);
2751 case OFPP_CONTROLLER:
2752 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2755 add_output_action(ctx, ODPP_LOCAL);
2758 odp_port = ofp_port_to_odp_port(port);
2759 if (odp_port != ctx->flow.in_port) {
2760 add_output_action(ctx, odp_port);
2765 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2766 ctx->nf_output_iface = NF_OUT_FLOOD;
2767 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2768 ctx->nf_output_iface = prev_nf_output_iface;
2769 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2770 ctx->nf_output_iface != NF_OUT_FLOOD) {
2771 ctx->nf_output_iface = NF_OUT_MULTI;
2776 xlate_output_action(struct action_xlate_ctx *ctx,
2777 const struct ofp_action_output *oao)
2779 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2782 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2783 * optimization, because we're going to add another action that sets the
2784 * priority immediately after, or because there are no actions following the
2787 remove_pop_action(struct action_xlate_ctx *ctx)
2789 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2790 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2791 ctx->last_pop_priority = -1;
2796 add_pop_action(struct action_xlate_ctx *ctx)
2798 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2799 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2800 ctx->last_pop_priority = ctx->odp_actions->size;
2805 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2806 const struct ofp_action_enqueue *oae)
2808 uint16_t ofp_port, odp_port;
2812 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2815 /* Fall back to ordinary output action. */
2816 xlate_output_action__(ctx, ntohs(oae->port), 0);
2820 /* Figure out ODP output port. */
2821 ofp_port = ntohs(oae->port);
2822 if (ofp_port != OFPP_IN_PORT) {
2823 odp_port = ofp_port_to_odp_port(ofp_port);
2825 odp_port = ctx->flow.in_port;
2828 /* Add ODP actions. */
2829 remove_pop_action(ctx);
2830 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2831 add_output_action(ctx, odp_port);
2832 add_pop_action(ctx);
2834 /* Update NetFlow output port. */
2835 if (ctx->nf_output_iface == NF_OUT_DROP) {
2836 ctx->nf_output_iface = odp_port;
2837 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2838 ctx->nf_output_iface = NF_OUT_MULTI;
2843 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2844 const struct nx_action_set_queue *nasq)
2849 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2852 /* Couldn't translate queue to a priority, so ignore. A warning
2853 * has already been logged. */
2857 remove_pop_action(ctx);
2858 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2862 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2864 ovs_be16 tci = ctx->flow.vlan_tci;
2865 if (!(tci & htons(VLAN_CFI))) {
2866 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2868 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2869 tci & ~htons(VLAN_CFI));
2873 struct xlate_reg_state {
2879 save_reg_state(const struct action_xlate_ctx *ctx,
2880 struct xlate_reg_state *state)
2882 state->vlan_tci = ctx->flow.vlan_tci;
2883 state->tun_id = ctx->flow.tun_id;
2887 update_reg_state(struct action_xlate_ctx *ctx,
2888 const struct xlate_reg_state *state)
2890 if (ctx->flow.vlan_tci != state->vlan_tci) {
2891 xlate_set_dl_tci(ctx);
2893 if (ctx->flow.tun_id != state->tun_id) {
2894 nl_msg_put_be64(ctx->odp_actions,
2895 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2900 xlate_nicira_action(struct action_xlate_ctx *ctx,
2901 const struct nx_action_header *nah)
2903 const struct nx_action_resubmit *nar;
2904 const struct nx_action_set_tunnel *nast;
2905 const struct nx_action_set_queue *nasq;
2906 const struct nx_action_multipath *nam;
2907 enum nx_action_subtype subtype = ntohs(nah->subtype);
2908 struct xlate_reg_state state;
2911 assert(nah->vendor == htonl(NX_VENDOR_ID));
2913 case NXAST_RESUBMIT:
2914 nar = (const struct nx_action_resubmit *) nah;
2915 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2918 case NXAST_SET_TUNNEL:
2919 nast = (const struct nx_action_set_tunnel *) nah;
2920 tun_id = htonll(ntohl(nast->tun_id));
2921 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2922 ctx->flow.tun_id = tun_id;
2925 case NXAST_DROP_SPOOFED_ARP:
2926 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2927 nl_msg_put_flag(ctx->odp_actions,
2928 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2932 case NXAST_SET_QUEUE:
2933 nasq = (const struct nx_action_set_queue *) nah;
2934 xlate_set_queue_action(ctx, nasq);
2937 case NXAST_POP_QUEUE:
2938 add_pop_action(ctx);
2941 case NXAST_REG_MOVE:
2942 save_reg_state(ctx, &state);
2943 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2945 update_reg_state(ctx, &state);
2948 case NXAST_REG_LOAD:
2949 save_reg_state(ctx, &state);
2950 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2952 update_reg_state(ctx, &state);
2956 /* Nothing to do. */
2959 case NXAST_SET_TUNNEL64:
2960 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2961 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2962 ctx->flow.tun_id = tun_id;
2965 case NXAST_MULTIPATH:
2966 nam = (const struct nx_action_multipath *) nah;
2967 multipath_execute(nam, &ctx->flow);
2970 /* If you add a new action here that modifies flow data, don't forget to
2971 * update the flow key in ctx->flow at the same time. */
2973 case NXAST_SNAT__OBSOLETE:
2975 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2981 do_xlate_actions(const union ofp_action *in, size_t n_in,
2982 struct action_xlate_ctx *ctx)
2984 struct actions_iterator iter;
2985 const union ofp_action *ia;
2986 const struct ofport *port;
2988 port = get_port(ctx->ofproto, ctx->flow.in_port);
2989 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2990 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2991 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2992 /* Drop this flow. */
2996 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2997 enum ofp_action_type type = ntohs(ia->type);
2998 const struct ofp_action_dl_addr *oada;
3002 xlate_output_action(ctx, &ia->output);
3005 case OFPAT_SET_VLAN_VID:
3006 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3007 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3008 xlate_set_dl_tci(ctx);
3011 case OFPAT_SET_VLAN_PCP:
3012 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3013 ctx->flow.vlan_tci |= htons(
3014 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3015 xlate_set_dl_tci(ctx);
3018 case OFPAT_STRIP_VLAN:
3019 ctx->flow.vlan_tci = htons(0);
3020 xlate_set_dl_tci(ctx);
3023 case OFPAT_SET_DL_SRC:
3024 oada = ((struct ofp_action_dl_addr *) ia);
3025 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3026 oada->dl_addr, ETH_ADDR_LEN);
3027 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3030 case OFPAT_SET_DL_DST:
3031 oada = ((struct ofp_action_dl_addr *) ia);
3032 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3033 oada->dl_addr, ETH_ADDR_LEN);
3034 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3037 case OFPAT_SET_NW_SRC:
3038 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3039 ia->nw_addr.nw_addr);
3040 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3043 case OFPAT_SET_NW_DST:
3044 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3045 ia->nw_addr.nw_addr);
3046 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3049 case OFPAT_SET_NW_TOS:
3050 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3052 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3055 case OFPAT_SET_TP_SRC:
3056 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3057 ia->tp_port.tp_port);
3058 ctx->flow.tp_src = ia->tp_port.tp_port;
3061 case OFPAT_SET_TP_DST:
3062 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3063 ia->tp_port.tp_port);
3064 ctx->flow.tp_dst = ia->tp_port.tp_port;
3068 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3072 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3076 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3083 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3084 struct ofproto *ofproto, const struct flow *flow,
3085 const struct ofpbuf *packet)
3087 ctx->ofproto = ofproto;
3089 ctx->packet = packet;
3090 ctx->resubmit_hook = NULL;
3091 ctx->check_special = true;
3094 static struct ofpbuf *
3095 xlate_actions(struct action_xlate_ctx *ctx,
3096 const union ofp_action *in, size_t n_in)
3098 COVERAGE_INC(ofproto_ofp2odp);
3100 ctx->odp_actions = ofpbuf_new(512);
3102 ctx->may_set_up_flow = true;
3103 ctx->nf_output_iface = NF_OUT_DROP;
3105 ctx->last_pop_priority = -1;
3107 if (!ctx->check_special
3108 || !ctx->ofproto->ofhooks->special_cb
3109 || ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3110 ctx->ofproto->aux)) {
3111 do_xlate_actions(in, n_in, ctx);
3113 ctx->may_set_up_flow = false;
3116 remove_pop_action(ctx);
3118 /* Check with in-band control to see if we're allowed to set up this
3120 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3121 ctx->odp_actions->data, ctx->odp_actions->size)) {
3122 ctx->may_set_up_flow = false;
3125 return ctx->odp_actions;
3128 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3129 * error message code (composed with ofp_mkerr()) for the caller to propagate
3130 * upward. Otherwise, returns 0.
3132 * The log message mentions 'msg_type'. */
3134 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3136 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3137 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3138 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3141 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3148 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3150 struct ofproto *p = ofconn->ofproto;
3151 struct ofp_packet_out *opo;
3152 struct ofpbuf payload, *buffer;
3153 union ofp_action *ofp_actions;
3154 struct action_xlate_ctx ctx;
3155 struct ofpbuf *odp_actions;
3156 struct ofpbuf request;
3158 size_t n_ofp_actions;
3162 COVERAGE_INC(ofproto_packet_out);
3164 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3169 /* Get ofp_packet_out. */
3170 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3171 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3174 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3175 &ofp_actions, &n_ofp_actions);
3181 if (opo->buffer_id != htonl(UINT32_MAX)) {
3182 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3184 if (error || !buffer) {
3193 /* Extract flow, check actions. */
3194 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3196 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3202 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3203 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3204 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3205 ofpbuf_delete(odp_actions);
3208 ofpbuf_delete(buffer);
3213 update_port_config(struct ofproto *p, struct ofport *port,
3214 uint32_t config, uint32_t mask)
3216 mask &= config ^ port->opp.config;
3217 if (mask & OFPPC_PORT_DOWN) {
3218 if (config & OFPPC_PORT_DOWN) {
3219 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3221 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3224 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3225 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3226 if (mask & REVALIDATE_BITS) {
3227 COVERAGE_INC(ofproto_costly_flags);
3228 port->opp.config ^= mask & REVALIDATE_BITS;
3229 p->need_revalidate = true;
3231 #undef REVALIDATE_BITS
3232 if (mask & OFPPC_NO_PACKET_IN) {
3233 port->opp.config ^= OFPPC_NO_PACKET_IN;
3238 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3240 struct ofproto *p = ofconn->ofproto;
3241 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3242 struct ofport *port;
3245 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3250 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3252 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3253 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3254 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3256 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3257 if (opm->advertise) {
3258 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3264 static struct ofpbuf *
3265 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3267 struct ofp_stats_reply *osr;
3270 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3271 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3273 osr->flags = htons(0);
3277 static struct ofpbuf *
3278 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3280 const struct ofp_stats_request *osr
3281 = (const struct ofp_stats_request *) request;
3282 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3286 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3287 struct ofpbuf **msgp)
3289 struct ofpbuf *msg = *msgp;
3290 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3291 if (nbytes + msg->size > UINT16_MAX) {
3292 struct ofp_stats_reply *reply = msg->data;
3293 reply->flags = htons(OFPSF_REPLY_MORE);
3294 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3295 queue_tx(msg, ofconn, ofconn->reply_counter);
3297 return ofpbuf_put_uninit(*msgp, nbytes);
3300 static struct ofpbuf *
3301 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3303 struct nicira_stats_msg *nsm;
3306 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3307 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3308 nsm->type = htons(OFPST_VENDOR);
3309 nsm->flags = htons(0);
3310 nsm->vendor = htonl(NX_VENDOR_ID);
3311 nsm->subtype = subtype;
3315 static struct ofpbuf *
3316 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3318 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3322 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3323 struct ofpbuf **msgp)
3325 struct ofpbuf *msg = *msgp;
3326 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3327 if (nbytes + msg->size > UINT16_MAX) {
3328 struct nicira_stats_msg *reply = msg->data;
3329 reply->flags = htons(OFPSF_REPLY_MORE);
3330 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3331 queue_tx(msg, ofconn, ofconn->reply_counter);
3333 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3337 handle_desc_stats_request(struct ofconn *ofconn,
3338 const struct ofp_header *request)
3340 struct ofproto *p = ofconn->ofproto;
3341 struct ofp_desc_stats *ods;
3344 msg = start_ofp_stats_reply(request, sizeof *ods);
3345 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3346 memset(ods, 0, sizeof *ods);
3347 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3348 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3349 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3350 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3351 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3352 queue_tx(msg, ofconn, ofconn->reply_counter);
3358 handle_table_stats_request(struct ofconn *ofconn,
3359 const struct ofp_header *request)
3361 struct ofproto *p = ofconn->ofproto;
3362 struct ofp_table_stats *ots;
3365 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3367 /* Classifier table. */
3368 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3369 memset(ots, 0, sizeof *ots);
3370 strcpy(ots->name, "classifier");
3371 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3372 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3373 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3374 ots->active_count = htonl(classifier_count(&p->cls));
3375 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3376 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3378 queue_tx(msg, ofconn, ofconn->reply_counter);
3383 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3384 struct ofpbuf **msgp)
3386 struct netdev_stats stats;
3387 struct ofp_port_stats *ops;
3389 /* Intentionally ignore return value, since errors will set
3390 * 'stats' to all-1s, which is correct for OpenFlow, and
3391 * netdev_get_stats() will log errors. */
3392 netdev_get_stats(port->netdev, &stats);
3394 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3395 ops->port_no = htons(port->opp.port_no);
3396 memset(ops->pad, 0, sizeof ops->pad);
3397 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3398 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3399 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3400 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3401 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3402 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3403 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3404 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3405 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3406 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3407 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3408 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3412 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3414 struct ofproto *p = ofconn->ofproto;
3415 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3416 struct ofp_port_stats *ops;
3418 struct ofport *port;
3420 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3421 if (psr->port_no != htons(OFPP_NONE)) {
3422 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3424 append_port_stat(port, ofconn, &msg);
3427 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3428 append_port_stat(port, ofconn, &msg);
3432 queue_tx(msg, ofconn, ofconn->reply_counter);
3437 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3439 long long int msecs = time_msec() - start;
3440 *sec = htonl(msecs / 1000);
3441 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3445 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3446 ovs_be16 out_port, struct ofpbuf **replyp)
3448 struct ofp_flow_stats *ofs;
3449 uint64_t packet_count, byte_count;
3451 size_t act_len, len;
3453 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3457 act_len = sizeof *rule->actions * rule->n_actions;
3458 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3460 rule_get_stats(rule, &packet_count, &byte_count);
3462 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3463 ofs->length = htons(len);
3466 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3467 rule->flow_cookie, &cookie);
3468 put_32aligned_be64(&ofs->cookie, cookie);
3469 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3470 ofs->priority = htons(rule->cr.priority);
3471 ofs->idle_timeout = htons(rule->idle_timeout);
3472 ofs->hard_timeout = htons(rule->hard_timeout);
3473 memset(ofs->pad2, 0, sizeof ofs->pad2);
3474 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3475 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3476 if (rule->n_actions > 0) {
3477 memcpy(ofs->actions, rule->actions, act_len);
3482 is_valid_table(uint8_t table_id)
3484 if (table_id == 0 || table_id == 0xff) {
3487 /* It would probably be better to reply with an error but there doesn't
3488 * seem to be any appropriate value, so that might just be
3490 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3497 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3499 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3500 struct ofpbuf *reply;
3502 COVERAGE_INC(ofproto_flows_req);
3503 reply = start_ofp_stats_reply(oh, 1024);
3504 if (is_valid_table(fsr->table_id)) {
3505 struct cls_cursor cursor;
3506 struct cls_rule target;
3509 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3511 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3512 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3513 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3516 queue_tx(reply, ofconn, ofconn->reply_counter);
3522 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3523 ovs_be16 out_port, struct ofpbuf **replyp)
3525 struct nx_flow_stats *nfs;
3526 uint64_t packet_count, byte_count;
3527 size_t act_len, start_len;
3528 struct ofpbuf *reply;
3530 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3534 rule_get_stats(rule, &packet_count, &byte_count);
3536 act_len = sizeof *rule->actions * rule->n_actions;
3538 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3539 start_len = (*replyp)->size;
3542 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3545 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3546 nfs->cookie = rule->flow_cookie;
3547 nfs->priority = htons(rule->cr.priority);
3548 nfs->idle_timeout = htons(rule->idle_timeout);
3549 nfs->hard_timeout = htons(rule->hard_timeout);
3550 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3551 memset(nfs->pad2, 0, sizeof nfs->pad2);
3552 nfs->packet_count = htonll(packet_count);
3553 nfs->byte_count = htonll(byte_count);
3554 if (rule->n_actions > 0) {
3555 ofpbuf_put(reply, rule->actions, act_len);
3557 nfs->length = htons(reply->size - start_len);
3561 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3563 struct nx_flow_stats_request *nfsr;
3564 struct cls_rule target;
3565 struct ofpbuf *reply;
3569 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3571 /* Dissect the message. */
3572 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3573 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3578 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3581 COVERAGE_INC(ofproto_flows_req);
3582 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3583 if (is_valid_table(nfsr->table_id)) {
3584 struct cls_cursor cursor;
3587 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3588 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3589 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3592 queue_tx(reply, ofconn, ofconn->reply_counter);
3598 flow_stats_ds(struct rule *rule, struct ds *results)
3600 uint64_t packet_count, byte_count;
3601 size_t act_len = sizeof *rule->actions * rule->n_actions;
3603 rule_get_stats(rule, &packet_count, &byte_count);
3605 ds_put_format(results, "duration=%llds, ",
3606 (time_msec() - rule->created) / 1000);
3607 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3608 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3609 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3610 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3611 cls_rule_format(&rule->cr, results);
3612 ds_put_char(results, ',');
3614 ofp_print_actions(results, &rule->actions->header, act_len);
3616 ds_put_cstr(results, "drop");
3618 ds_put_cstr(results, "\n");
3621 /* Adds a pretty-printed description of all flows to 'results', including
3622 * hidden flows (e.g., set up by in-band control). */
3624 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3626 struct cls_cursor cursor;
3629 cls_cursor_init(&cursor, &p->cls, NULL);
3630 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3631 flow_stats_ds(rule, results);
3636 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3637 ovs_be16 out_port, uint8_t table_id,
3638 struct ofp_aggregate_stats_reply *oasr)
3640 uint64_t total_packets = 0;
3641 uint64_t total_bytes = 0;
3644 COVERAGE_INC(ofproto_agg_request);
3646 if (is_valid_table(table_id)) {
3647 struct cls_cursor cursor;
3650 cls_cursor_init(&cursor, &ofproto->cls, target);
3651 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3652 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3653 uint64_t packet_count;
3654 uint64_t byte_count;
3656 rule_get_stats(rule, &packet_count, &byte_count);
3658 total_packets += packet_count;
3659 total_bytes += byte_count;
3665 oasr->flow_count = htonl(n_flows);
3666 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3667 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3668 memset(oasr->pad, 0, sizeof oasr->pad);
3672 handle_aggregate_stats_request(struct ofconn *ofconn,
3673 const struct ofp_header *oh)
3675 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3676 struct ofp_aggregate_stats_reply *reply;
3677 struct cls_rule target;
3680 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3683 msg = start_ofp_stats_reply(oh, sizeof *reply);
3684 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3685 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3686 request->table_id, reply);
3687 queue_tx(msg, ofconn, ofconn->reply_counter);
3692 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3694 struct nx_aggregate_stats_request *request;
3695 struct ofp_aggregate_stats_reply *reply;
3696 struct cls_rule target;
3701 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3703 /* Dissect the message. */
3704 request = ofpbuf_pull(&b, sizeof *request);
3705 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3710 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3714 COVERAGE_INC(ofproto_flows_req);
3715 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3716 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3717 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3718 request->table_id, reply);
3719 queue_tx(buf, ofconn, ofconn->reply_counter);
3724 struct queue_stats_cbdata {
3725 struct ofconn *ofconn;
3726 struct ofport *ofport;
3731 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3732 const struct netdev_queue_stats *stats)
3734 struct ofp_queue_stats *reply;
3736 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3737 reply->port_no = htons(cbdata->ofport->opp.port_no);
3738 memset(reply->pad, 0, sizeof reply->pad);
3739 reply->queue_id = htonl(queue_id);
3740 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3741 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3742 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3746 handle_queue_stats_dump_cb(uint32_t queue_id,
3747 struct netdev_queue_stats *stats,
3750 struct queue_stats_cbdata *cbdata = cbdata_;
3752 put_queue_stats(cbdata, queue_id, stats);
3756 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3757 struct queue_stats_cbdata *cbdata)
3759 cbdata->ofport = port;
3760 if (queue_id == OFPQ_ALL) {
3761 netdev_dump_queue_stats(port->netdev,
3762 handle_queue_stats_dump_cb, cbdata);
3764 struct netdev_queue_stats stats;
3766 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3767 put_queue_stats(cbdata, queue_id, &stats);
3773 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3775 struct ofproto *ofproto = ofconn->ofproto;
3776 const struct ofp_queue_stats_request *qsr;
3777 struct queue_stats_cbdata cbdata;
3778 struct ofport *port;
3779 unsigned int port_no;
3782 qsr = ofputil_stats_body(oh);
3784 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3787 COVERAGE_INC(ofproto_queue_req);
3789 cbdata.ofconn = ofconn;
3790 cbdata.msg = start_ofp_stats_reply(oh, 128);
3792 port_no = ntohs(qsr->port_no);
3793 queue_id = ntohl(qsr->queue_id);
3794 if (port_no == OFPP_ALL) {
3795 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3796 handle_queue_stats_for_port(port, queue_id, &cbdata);
3798 } else if (port_no < ofproto->max_ports) {
3799 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3801 handle_queue_stats_for_port(port, queue_id, &cbdata);
3804 ofpbuf_delete(cbdata.msg);
3805 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3807 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3812 /* Updates 'facet''s used time. Caller is responsible for calling
3813 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3815 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3818 if (used > facet->used) {
3820 if (used > facet->rule->used) {
3821 facet->rule->used = used;
3823 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3827 /* Folds the statistics from 'stats' into the counters in 'facet'.
3829 * Because of the meaning of a facet's counters, it only makes sense to do this
3830 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3831 * packet that was sent by hand or if it represents statistics that have been
3832 * cleared out of the datapath. */
3834 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3835 const struct dpif_flow_stats *stats)
3837 if (stats->n_packets || stats->used > facet->used) {
3838 facet_update_time(ofproto, facet, stats->used);
3839 facet->packet_count += stats->n_packets;
3840 facet->byte_count += stats->n_bytes;
3841 facet_push_stats(ofproto, facet);
3842 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3847 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3849 uint64_t rs_packets, rs_bytes;
3851 assert(facet->packet_count >= facet->rs_packet_count);
3852 assert(facet->byte_count >= facet->rs_byte_count);
3853 assert(facet->used >= facet->rs_used);
3855 rs_packets = facet->packet_count - facet->rs_packet_count;
3856 rs_bytes = facet->byte_count - facet->rs_byte_count;
3858 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3859 facet->rs_packet_count = facet->packet_count;
3860 facet->rs_byte_count = facet->byte_count;
3861 facet->rs_used = facet->used;
3863 flow_push_stats(ofproto, facet->rule, &facet->flow,
3864 rs_packets, rs_bytes, facet->used);
3868 struct ofproto_push {
3869 struct action_xlate_ctx ctx;
3876 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3878 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3881 rule->packet_count += push->packets;
3882 rule->byte_count += push->bytes;
3883 rule->used = MAX(push->used, rule->used);
3887 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3888 * 'rule''s actions. */
3890 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3891 struct flow *flow, uint64_t packets, uint64_t bytes,
3894 struct ofproto_push push;
3896 push.packets = packets;
3900 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3901 push.ctx.resubmit_hook = push_resubmit;
3902 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3905 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3906 * in which no matching flow already exists in the flow table.
3908 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3909 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3910 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3912 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3915 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3917 struct ofproto *p = ofconn->ofproto;
3918 struct ofpbuf *packet;
3923 if (fm->flags & OFPFF_CHECK_OVERLAP
3924 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3925 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3929 if (fm->buffer_id != UINT32_MAX) {
3930 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3934 in_port = UINT16_MAX;
3937 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3938 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3939 fm->flags & OFPFF_SEND_FLOW_REM);
3940 rule_insert(p, rule);
3942 rule_execute(p, rule, in_port, packet);
3947 static struct rule *
3948 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3950 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3954 send_buffered_packet(struct ofconn *ofconn,
3955 struct rule *rule, uint32_t buffer_id)
3957 struct ofpbuf *packet;
3961 if (buffer_id == UINT32_MAX) {
3965 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3970 rule_execute(ofconn->ofproto, rule, in_port, packet);
3975 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3977 struct modify_flows_cbdata {
3978 struct ofproto *ofproto;
3979 const struct flow_mod *fm;
3983 static int modify_flow(struct ofproto *, const struct flow_mod *,
3986 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3987 * encoded by ofp_mkerr() on failure.
3989 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3992 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3994 struct ofproto *p = ofconn->ofproto;
3995 struct rule *match = NULL;
3996 struct cls_cursor cursor;
3999 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4000 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4001 if (!rule_is_hidden(rule)) {
4003 modify_flow(p, fm, rule);
4008 /* This credits the packet to whichever flow happened to match last.
4009 * That's weird. Maybe we should do a lookup for the flow that
4010 * actually matches the packet? Who knows. */
4011 send_buffered_packet(ofconn, match, fm->buffer_id);
4014 return add_flow(ofconn, fm);
4018 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4019 * code as encoded by ofp_mkerr() on failure.
4021 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4024 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4026 struct ofproto *p = ofconn->ofproto;
4027 struct rule *rule = find_flow_strict(p, fm);
4028 if (rule && !rule_is_hidden(rule)) {
4029 modify_flow(p, fm, rule);
4030 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4032 return add_flow(ofconn, fm);
4036 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4037 * been identified as a flow in 'p''s flow table to be modified, by changing
4038 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4039 * ofp_action[] structures). */
4041 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4043 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4045 rule->flow_cookie = fm->cookie;
4047 /* If the actions are the same, do nothing. */
4048 if (fm->n_actions == rule->n_actions
4050 || !memcmp(fm->actions, rule->actions, actions_len))) {
4054 /* Replace actions. */
4055 free(rule->actions);
4056 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4057 rule->n_actions = fm->n_actions;
4059 p->need_revalidate = true;
4064 /* OFPFC_DELETE implementation. */
4066 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4068 /* Implements OFPFC_DELETE. */
4070 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4072 struct rule *rule, *next_rule;
4073 struct cls_cursor cursor;
4075 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4076 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4077 delete_flow(p, rule, htons(fm->out_port));
4081 /* Implements OFPFC_DELETE_STRICT. */
4083 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4085 struct rule *rule = find_flow_strict(p, fm);
4087 delete_flow(p, rule, htons(fm->out_port));
4091 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4092 * been identified as a flow to delete from 'p''s flow table, by deleting the
4093 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4096 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4097 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4098 * specified 'out_port'. */
4100 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4102 if (rule_is_hidden(rule)) {
4106 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4110 rule_send_removed(p, rule, OFPRR_DELETE);
4111 rule_remove(p, rule);
4115 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4117 struct ofproto *p = ofconn->ofproto;
4121 error = reject_slave_controller(ofconn, "flow_mod");
4126 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4131 /* We do not support the emergency flow cache. It will hopefully get
4132 * dropped from OpenFlow in the near future. */
4133 if (fm.flags & OFPFF_EMERG) {
4134 /* There isn't a good fit for an error code, so just state that the
4135 * flow table is full. */
4136 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4139 error = validate_actions(fm.actions, fm.n_actions,
4140 &fm.cr.flow, p->max_ports);
4145 switch (fm.command) {
4147 return add_flow(ofconn, &fm);
4150 return modify_flows_loose(ofconn, &fm);
4152 case OFPFC_MODIFY_STRICT:
4153 return modify_flow_strict(ofconn, &fm);
4156 delete_flows_loose(p, &fm);
4159 case OFPFC_DELETE_STRICT:
4160 delete_flow_strict(p, &fm);
4164 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4169 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4171 const struct nxt_tun_id_cookie *msg
4172 = (const struct nxt_tun_id_cookie *) oh;
4174 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4179 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4181 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4182 struct nx_role_request *reply;
4186 if (ofconn->type != OFCONN_PRIMARY) {
4187 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4189 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4192 role = ntohl(nrr->role);
4193 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4194 && role != NX_ROLE_SLAVE) {
4195 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4197 /* There's no good error code for this. */
4198 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4201 if (role == NX_ROLE_MASTER) {
4202 struct ofconn *other;
4204 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4205 if (other->role == NX_ROLE_MASTER) {
4206 other->role = NX_ROLE_SLAVE;
4210 ofconn->role = role;
4212 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4213 reply->role = htonl(role);
4214 queue_tx(buf, ofconn, ofconn->reply_counter);
4220 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4222 const struct nxt_set_flow_format *msg
4223 = (const struct nxt_set_flow_format *) oh;
4226 format = ntohl(msg->format);
4227 if (format == NXFF_OPENFLOW10
4228 || format == NXFF_TUN_ID_FROM_COOKIE
4229 || format == NXFF_NXM) {
4230 ofconn->flow_format = format;
4233 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4238 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4240 struct ofp_header *ob;
4243 /* Currently, everything executes synchronously, so we can just
4244 * immediately send the barrier reply. */
4245 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4246 queue_tx(buf, ofconn, ofconn->reply_counter);
4251 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4253 const struct ofp_header *oh = msg->data;
4254 const struct ofputil_msg_type *type;
4257 error = ofputil_decode_msg_type(oh, &type);
4262 switch (ofputil_msg_type_code(type)) {
4263 /* OpenFlow requests. */
4264 case OFPUTIL_OFPT_ECHO_REQUEST:
4265 return handle_echo_request(ofconn, oh);
4267 case OFPUTIL_OFPT_FEATURES_REQUEST:
4268 return handle_features_request(ofconn, oh);
4270 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4271 return handle_get_config_request(ofconn, oh);
4273 case OFPUTIL_OFPT_SET_CONFIG:
4274 return handle_set_config(ofconn, msg->data);
4276 case OFPUTIL_OFPT_PACKET_OUT:
4277 return handle_packet_out(ofconn, oh);
4279 case OFPUTIL_OFPT_PORT_MOD:
4280 return handle_port_mod(ofconn, oh);
4282 case OFPUTIL_OFPT_FLOW_MOD:
4283 return handle_flow_mod(ofconn, oh);
4285 case OFPUTIL_OFPT_BARRIER_REQUEST:
4286 return handle_barrier_request(ofconn, oh);
4288 /* OpenFlow replies. */
4289 case OFPUTIL_OFPT_ECHO_REPLY:
4292 /* Nicira extension requests. */
4293 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4294 return handle_tun_id_from_cookie(ofconn, oh);
4296 case OFPUTIL_NXT_ROLE_REQUEST:
4297 return handle_role_request(ofconn, oh);
4299 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4300 return handle_nxt_set_flow_format(ofconn, oh);
4302 case OFPUTIL_NXT_FLOW_MOD:
4303 return handle_flow_mod(ofconn, oh);
4305 /* OpenFlow statistics requests. */
4306 case OFPUTIL_OFPST_DESC_REQUEST:
4307 return handle_desc_stats_request(ofconn, oh);
4309 case OFPUTIL_OFPST_FLOW_REQUEST:
4310 return handle_flow_stats_request(ofconn, oh);
4312 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4313 return handle_aggregate_stats_request(ofconn, oh);
4315 case OFPUTIL_OFPST_TABLE_REQUEST:
4316 return handle_table_stats_request(ofconn, oh);
4318 case OFPUTIL_OFPST_PORT_REQUEST:
4319 return handle_port_stats_request(ofconn, oh);
4321 case OFPUTIL_OFPST_QUEUE_REQUEST:
4322 return handle_queue_stats_request(ofconn, oh);
4324 /* Nicira extension statistics requests. */
4325 case OFPUTIL_NXST_FLOW_REQUEST:
4326 return handle_nxst_flow(ofconn, oh);
4328 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4329 return handle_nxst_aggregate(ofconn, oh);
4331 case OFPUTIL_INVALID:
4332 case OFPUTIL_OFPT_HELLO:
4333 case OFPUTIL_OFPT_ERROR:
4334 case OFPUTIL_OFPT_FEATURES_REPLY:
4335 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4336 case OFPUTIL_OFPT_PACKET_IN:
4337 case OFPUTIL_OFPT_FLOW_REMOVED:
4338 case OFPUTIL_OFPT_PORT_STATUS:
4339 case OFPUTIL_OFPT_BARRIER_REPLY:
4340 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4341 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4342 case OFPUTIL_OFPST_DESC_REPLY:
4343 case OFPUTIL_OFPST_FLOW_REPLY:
4344 case OFPUTIL_OFPST_QUEUE_REPLY:
4345 case OFPUTIL_OFPST_PORT_REPLY:
4346 case OFPUTIL_OFPST_TABLE_REPLY:
4347 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4348 case OFPUTIL_NXT_ROLE_REPLY:
4349 case OFPUTIL_NXT_FLOW_REMOVED:
4350 case OFPUTIL_NXST_FLOW_REPLY:
4351 case OFPUTIL_NXST_AGGREGATE_REPLY:
4353 if (VLOG_IS_WARN_ENABLED()) {
4354 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4355 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4358 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4359 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4361 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4367 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4369 int error = handle_openflow__(ofconn, ofp_msg);
4371 send_error_oh(ofconn, ofp_msg->data, error);
4373 COVERAGE_INC(ofproto_recv_openflow);
4377 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4379 struct facet *facet;
4382 /* Obtain in_port and tun_id, at least. */
4383 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4385 /* Set header pointers in 'flow'. */
4386 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4388 if (p->ofhooks->special_cb
4389 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4390 ofpbuf_delete(upcall->packet);
4394 /* Check with in-band control to see if this packet should be sent
4395 * to the local port regardless of the flow table. */
4396 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4397 struct ofpbuf odp_actions;
4399 ofpbuf_init(&odp_actions, 32);
4400 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, ODPP_LOCAL);
4401 dpif_execute(p->dpif, odp_actions.data, odp_actions.size,
4403 ofpbuf_uninit(&odp_actions);
4406 facet = facet_lookup_valid(p, &flow);
4408 struct rule *rule = rule_lookup(p, &flow);
4410 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4411 struct ofport *port = get_port(p, flow.in_port);
4413 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4414 COVERAGE_INC(ofproto_no_packet_in);
4415 /* XXX install 'drop' flow entry */
4416 ofpbuf_delete(upcall->packet);
4420 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4424 COVERAGE_INC(ofproto_packet_in);
4425 send_packet_in(p, upcall, &flow, false);
4429 facet = facet_create(p, rule, &flow, upcall->packet);
4430 } else if (!facet->may_install) {
4431 /* The facet is not installable, that is, we need to process every
4432 * packet, so process the current packet's actions into 'facet'. */
4433 facet_make_actions(p, facet, upcall->packet);
4436 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4438 * Extra-special case for fail-open mode.
4440 * We are in fail-open mode and the packet matched the fail-open rule,
4441 * but we are connected to a controller too. We should send the packet
4442 * up to the controller in the hope that it will try to set up a flow
4443 * and thereby allow us to exit fail-open.
4445 * See the top-level comment in fail-open.c for more information.
4447 send_packet_in(p, upcall, &flow, true);
4450 facet_execute(p, facet, upcall->packet);
4451 facet_install(p, facet, false);
4455 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4459 switch (upcall->type) {
4460 case DPIF_UC_ACTION:
4461 COVERAGE_INC(ofproto_ctlr_action);
4462 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4463 send_packet_in(p, upcall, &flow, false);
4466 case DPIF_UC_SAMPLE:
4468 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4469 ofproto_sflow_received(p->sflow, upcall, &flow);
4471 ofpbuf_delete(upcall->packet);
4475 handle_miss_upcall(p, upcall);
4478 case DPIF_N_UC_TYPES:
4480 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4485 /* Flow expiration. */
4487 static int ofproto_dp_max_idle(const struct ofproto *);
4488 static void ofproto_update_stats(struct ofproto *);
4489 static void rule_expire(struct ofproto *, struct rule *);
4490 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4492 /* This function is called periodically by ofproto_run(). Its job is to
4493 * collect updates for the flows that have been installed into the datapath,
4494 * most importantly when they last were used, and then use that information to
4495 * expire flows that have not been used recently.
4497 * Returns the number of milliseconds after which it should be called again. */
4499 ofproto_expire(struct ofproto *ofproto)
4501 struct rule *rule, *next_rule;
4502 struct cls_cursor cursor;
4505 /* Update stats for each flow in the datapath. */
4506 ofproto_update_stats(ofproto);
4508 /* Expire facets that have been idle too long. */
4509 dp_max_idle = ofproto_dp_max_idle(ofproto);
4510 ofproto_expire_facets(ofproto, dp_max_idle);
4512 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4513 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4514 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4515 rule_expire(ofproto, rule);
4518 /* Let the hook know that we're at a stable point: all outstanding data
4519 * in existing flows has been accounted to the account_cb. Thus, the
4520 * hook can now reasonably do operations that depend on having accurate
4521 * flow volume accounting (currently, that's just bond rebalancing). */
4522 if (ofproto->ofhooks->account_checkpoint_cb) {
4523 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4526 return MIN(dp_max_idle, 1000);
4529 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4531 * This function also pushes statistics updates to rules which each facet
4532 * resubmits into. Generally these statistics will be accurate. However, if a
4533 * facet changes the rule it resubmits into at some time in between
4534 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4535 * old rule will be incorrectly attributed to the new rule. This could be
4536 * avoided by calling ofproto_update_stats() whenever rules are created or
4537 * deleted. However, the performance impact of making so many calls to the
4538 * datapath do not justify the benefit of having perfectly accurate statistics.
4541 ofproto_update_stats(struct ofproto *p)
4543 const struct dpif_flow_stats *stats;
4544 struct dpif_flow_dump dump;
4545 const struct nlattr *key;
4548 dpif_flow_dump_start(&dump, p->dpif);
4549 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4550 struct facet *facet;
4553 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4557 odp_flow_key_format(key, key_len, &s);
4558 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4564 facet = facet_find(p, &flow);
4566 if (facet && facet->installed) {
4568 if (stats->n_packets >= facet->dp_packet_count) {
4569 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4571 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4574 if (stats->n_bytes >= facet->dp_byte_count) {
4575 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4577 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4580 facet->dp_packet_count = stats->n_packets;
4581 facet->dp_byte_count = stats->n_bytes;
4583 facet_update_time(p, facet, stats->used);
4584 facet_account(p, facet, stats->n_bytes);
4585 facet_push_stats(p, facet);
4587 /* There's a flow in the datapath that we know nothing about.
4589 COVERAGE_INC(ofproto_unexpected_rule);
4590 dpif_flow_del(p->dpif, key, key_len, NULL);
4593 dpif_flow_dump_done(&dump);
4596 /* Calculates and returns the number of milliseconds of idle time after which
4597 * facets should expire from the datapath and we should fold their statistics
4598 * into their parent rules in userspace. */
4600 ofproto_dp_max_idle(const struct ofproto *ofproto)
4603 * Idle time histogram.
4605 * Most of the time a switch has a relatively small number of facets. When
4606 * this is the case we might as well keep statistics for all of them in
4607 * userspace and to cache them in the kernel datapath for performance as
4610 * As the number of facets increases, the memory required to maintain
4611 * statistics about them in userspace and in the kernel becomes
4612 * significant. However, with a large number of facets it is likely that
4613 * only a few of them are "heavy hitters" that consume a large amount of
4614 * bandwidth. At this point, only heavy hitters are worth caching in the
4615 * kernel and maintaining in userspaces; other facets we can discard.
4617 * The technique used to compute the idle time is to build a histogram with
4618 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4619 * that is installed in the kernel gets dropped in the appropriate bucket.
4620 * After the histogram has been built, we compute the cutoff so that only
4621 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4622 * cached. At least the most-recently-used bucket of facets is kept, so
4623 * actually an arbitrary number of facets can be kept in any given
4624 * expiration run (though the next run will delete most of those unless
4625 * they receive additional data).
4627 * This requires a second pass through the facets, in addition to the pass
4628 * made by ofproto_update_stats(), because the former function never looks
4629 * at uninstallable facets.
4631 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4632 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4633 int buckets[N_BUCKETS] = { 0 };
4634 struct facet *facet;
4639 total = hmap_count(&ofproto->facets);
4640 if (total <= 1000) {
4641 return N_BUCKETS * BUCKET_WIDTH;
4644 /* Build histogram. */
4646 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4647 long long int idle = now - facet->used;
4648 int bucket = (idle <= 0 ? 0
4649 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4650 : (unsigned int) idle / BUCKET_WIDTH);
4654 /* Find the first bucket whose flows should be expired. */
4655 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4656 if (buckets[bucket]) {
4659 subtotal += buckets[bucket++];
4660 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4665 if (VLOG_IS_DBG_ENABLED()) {
4669 ds_put_cstr(&s, "keep");
4670 for (i = 0; i < N_BUCKETS; i++) {
4672 ds_put_cstr(&s, ", drop");
4675 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4678 VLOG_INFO("%s: %s (msec:count)",
4679 dpif_name(ofproto->dpif), ds_cstr(&s));
4683 return bucket * BUCKET_WIDTH;
4687 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4689 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4690 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4691 struct ofexpired expired;
4693 if (facet->installed) {
4694 struct dpif_flow_stats stats;
4696 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4698 facet_update_stats(ofproto, facet, &stats);
4701 expired.flow = facet->flow;
4702 expired.packet_count = facet->packet_count;
4703 expired.byte_count = facet->byte_count;
4704 expired.used = facet->used;
4705 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4710 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4712 long long int cutoff = time_msec() - dp_max_idle;
4713 struct facet *facet, *next_facet;
4715 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4716 facet_active_timeout(ofproto, facet);
4717 if (facet->used < cutoff) {
4718 facet_remove(ofproto, facet);
4723 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4724 * then delete it entirely. */
4726 rule_expire(struct ofproto *ofproto, struct rule *rule)
4728 struct facet *facet, *next_facet;
4732 /* Has 'rule' expired? */
4734 if (rule->hard_timeout
4735 && now > rule->created + rule->hard_timeout * 1000) {
4736 reason = OFPRR_HARD_TIMEOUT;
4737 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4738 && now >rule->used + rule->idle_timeout * 1000) {
4739 reason = OFPRR_IDLE_TIMEOUT;
4744 COVERAGE_INC(ofproto_expired);
4746 /* Update stats. (This is a no-op if the rule expired due to an idle
4747 * timeout, because that only happens when the rule has no facets left.) */
4748 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4749 facet_remove(ofproto, facet);
4752 /* Get rid of the rule. */
4753 if (!rule_is_hidden(rule)) {
4754 rule_send_removed(ofproto, rule, reason);
4756 rule_remove(ofproto, rule);
4759 static struct ofpbuf *
4760 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4763 struct ofp_flow_removed *ofr;
4766 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4767 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4768 rule->flow_cookie, &ofr->cookie);
4769 ofr->priority = htons(rule->cr.priority);
4770 ofr->reason = reason;
4771 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4772 ofr->idle_timeout = htons(rule->idle_timeout);
4773 ofr->packet_count = htonll(rule->packet_count);
4774 ofr->byte_count = htonll(rule->byte_count);
4779 static struct ofpbuf *
4780 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4782 struct nx_flow_removed *nfr;
4786 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4787 match_len = nx_put_match(buf, &rule->cr);
4790 nfr->cookie = rule->flow_cookie;
4791 nfr->priority = htons(rule->cr.priority);
4792 nfr->reason = reason;
4793 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4794 nfr->idle_timeout = htons(rule->idle_timeout);
4795 nfr->match_len = htons(match_len);
4796 nfr->packet_count = htonll(rule->packet_count);
4797 nfr->byte_count = htonll(rule->byte_count);
4803 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4805 struct ofconn *ofconn;
4807 if (!rule->send_flow_removed) {
4811 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4814 if (!rconn_is_connected(ofconn->rconn)
4815 || !ofconn_receives_async_msgs(ofconn)) {
4819 msg = (ofconn->flow_format == NXFF_NXM
4820 ? compose_nx_flow_removed(rule, reason)
4821 : compose_ofp_flow_removed(ofconn, rule, reason));
4823 /* Account flow expirations under ofconn->reply_counter, the counter
4824 * for replies to OpenFlow requests. That works because preventing
4825 * OpenFlow requests from being processed also prevents new flows from
4826 * being added (and expiring). (It also prevents processing OpenFlow
4827 * requests that would not add new flows, so it is imperfect.) */
4828 queue_tx(msg, ofconn, ofconn->reply_counter);
4832 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4833 * The returned statistics include statistics for all of 'rule''s facets. */
4835 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4838 struct facet *facet;
4840 /* Start from historical data for 'rule' itself that are no longer tracked
4841 * in facets. This counts, for example, facets that have expired. */
4842 p = rule->packet_count;
4843 b = rule->byte_count;
4845 /* Add any statistics that are tracked by facets. This includes
4846 * statistical data recently updated by ofproto_update_stats() as well as
4847 * stats for packets that were executed "by hand" via dpif_execute(). */
4848 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4849 p += facet->packet_count;
4850 b += facet->byte_count;
4857 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4859 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4861 struct ofconn *ofconn = ofconn_;
4863 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4864 ofconn->packet_in_counter, 100);
4867 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4868 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4869 * scheduler for sending.
4871 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4872 * Otherwise, ownership is transferred to this function. */
4874 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4875 const struct flow *flow, bool clone)
4877 enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
4878 struct ofproto *ofproto = ofconn->ofproto;
4879 struct ofp_packet_in *opi;
4880 int total_len, send_len;
4881 struct ofpbuf *packet;
4885 /* Get OpenFlow buffer_id. */
4886 if (upcall->type == DPIF_UC_ACTION) {
4887 buffer_id = UINT32_MAX;
4888 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4889 buffer_id = pktbuf_get_null();
4890 } else if (!ofconn->pktbuf) {
4891 buffer_id = UINT32_MAX;
4893 buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
4896 /* Figure out how much of the packet to send. */
4897 total_len = send_len = upcall->packet->size;
4898 if (buffer_id != UINT32_MAX) {
4899 send_len = MIN(send_len, ofconn->miss_send_len);
4901 if (upcall->type == DPIF_UC_ACTION) {
4902 send_len = MIN(send_len, upcall->userdata);
4905 /* Copy or steal buffer for OFPT_PACKET_IN. */
4907 packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
4908 send_len, OPI_SIZE);
4910 packet = upcall->packet;
4911 packet->size = send_len;
4914 /* Add OFPT_PACKET_IN. */
4915 opi = ofpbuf_push_zeros(packet, OPI_SIZE);
4916 opi->header.version = OFP_VERSION;
4917 opi->header.type = OFPT_PACKET_IN;
4918 opi->total_len = htons(total_len);
4919 opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
4920 opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4921 opi->buffer_id = htonl(buffer_id);
4922 update_openflow_length(packet);
4924 /* Hand over to packet scheduler. It might immediately call into
4925 * do_send_packet_in() or it might buffer it for a while (until a later
4926 * call to pinsched_run()). */
4927 idx = upcall->type == DPIF_UC_MISS ? 0 : 1;
4928 pinsched_send(ofconn->schedulers[idx], flow->in_port,
4929 packet, do_send_packet_in, ofconn);
4932 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4933 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4934 * their individual configurations.
4936 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4937 * Otherwise, ownership is transferred to this function. */
4939 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4940 const struct flow *flow, bool clone)
4942 struct ofconn *ofconn, *prev;
4945 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4946 if (ofconn_receives_async_msgs(ofconn)) {
4948 schedule_packet_in(prev, upcall, flow, true);
4954 schedule_packet_in(prev, upcall, flow, clone);
4955 } else if (!clone) {
4956 ofpbuf_delete(upcall->packet);
4961 pick_datapath_id(const struct ofproto *ofproto)
4963 const struct ofport *port;
4965 port = get_port(ofproto, ODPP_LOCAL);
4967 uint8_t ea[ETH_ADDR_LEN];
4970 error = netdev_get_etheraddr(port->netdev, ea);
4972 return eth_addr_to_uint64(ea);
4974 VLOG_WARN("could not get MAC address for %s (%s)",
4975 netdev_get_name(port->netdev), strerror(error));
4977 return ofproto->fallback_dpid;
4981 pick_fallback_dpid(void)
4983 uint8_t ea[ETH_ADDR_LEN];
4984 eth_addr_nicira_random(ea);
4985 return eth_addr_to_uint64(ea);
4989 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4990 void *aux OVS_UNUSED)
4992 const struct shash_node *node;
4996 SHASH_FOR_EACH (node, &all_ofprotos) {
4997 ds_put_format(&results, "%s\n", node->name);
4999 unixctl_command_reply(conn, 200, ds_cstr(&results));
5000 ds_destroy(&results);
5003 struct ofproto_trace {
5004 struct action_xlate_ctx ctx;
5010 trace_format_rule(struct ds *result, int level, const struct rule *rule)
5012 ds_put_char_multiple(result, '\t', level);
5014 ds_put_cstr(result, "No match\n");
5018 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5019 ntohll(rule->flow_cookie));
5020 cls_rule_format(&rule->cr, result);
5021 ds_put_char(result, '\n');
5023 ds_put_char_multiple(result, '\t', level);
5024 ds_put_cstr(result, "OpenFlow ");
5025 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5026 rule->n_actions * sizeof *rule->actions);
5027 ds_put_char(result, '\n');
5031 trace_format_flow(struct ds *result, int level, const char *title,
5032 struct ofproto_trace *trace)
5034 ds_put_char_multiple(result, '\t', level);
5035 ds_put_format(result, "%s: ", title);
5036 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5037 ds_put_cstr(result, "unchanged");
5039 flow_format(result, &trace->ctx.flow);
5040 trace->flow = trace->ctx.flow;
5042 ds_put_char(result, '\n');
5046 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5048 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5049 struct ds *result = trace->result;
5051 ds_put_char(result, '\n');
5052 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5053 trace_format_rule(result, ctx->recurse + 1, rule);
5057 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5058 void *aux OVS_UNUSED)
5060 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5061 char *args = xstrdup(args_);
5062 char *save_ptr = NULL;
5063 struct ofproto *ofproto;
5064 struct ofpbuf packet;
5072 ofpbuf_init(&packet, strlen(args) / 2);
5075 dpname = strtok_r(args, " ", &save_ptr);
5076 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5077 in_port_s = strtok_r(NULL, " ", &save_ptr);
5078 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5079 if (!dpname || !in_port_s || !packet_s) {
5080 unixctl_command_reply(conn, 501, "Bad command syntax");
5084 ofproto = shash_find_data(&all_ofprotos, dpname);
5086 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5091 tun_id = htonll(strtoull(tun_id_s, NULL, 10));
5092 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5094 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5095 packet_s += strspn(packet_s, " ");
5096 if (*packet_s != '\0') {
5097 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5100 if (packet.size < ETH_HEADER_LEN) {
5101 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5105 ds_put_cstr(&result, "Packet: ");
5106 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5107 ds_put_cstr(&result, s);
5110 flow_extract(&packet, tun_id, in_port, &flow);
5111 ds_put_cstr(&result, "Flow: ");
5112 flow_format(&result, &flow);
5113 ds_put_char(&result, '\n');
5115 rule = rule_lookup(ofproto, &flow);
5116 trace_format_rule(&result, 0, rule);
5118 struct ofproto_trace trace;
5119 struct ofpbuf *odp_actions;
5121 trace.result = &result;
5123 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5124 trace.ctx.resubmit_hook = trace_resubmit;
5125 odp_actions = xlate_actions(&trace.ctx,
5126 rule->actions, rule->n_actions);
5128 ds_put_char(&result, '\n');
5129 trace_format_flow(&result, 0, "Final flow", &trace);
5130 ds_put_cstr(&result, "Datapath actions: ");
5131 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5132 ofpbuf_delete(odp_actions);
5135 unixctl_command_reply(conn, 200, ds_cstr(&result));
5138 ds_destroy(&result);
5139 ofpbuf_uninit(&packet);
5144 ofproto_unixctl_init(void)
5146 static bool registered;
5152 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5153 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5157 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5158 struct ofpbuf *odp_actions, tag_type *tags,
5159 uint16_t *nf_output_iface, void *ofproto_)
5161 struct ofproto *ofproto = ofproto_;
5164 /* Drop frames for reserved multicast addresses. */
5165 if (eth_addr_is_reserved(flow->dl_dst)) {
5169 /* Learn source MAC (but don't try to learn from revalidation). */
5170 if (packet != NULL) {
5171 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5173 GRAT_ARP_LOCK_NONE);
5175 /* The log messages here could actually be useful in debugging,
5176 * so keep the rate limit relatively high. */
5177 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5178 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5179 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5180 ofproto_revalidate(ofproto, rev_tag);
5184 /* Determine output port. */
5185 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5188 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5189 nf_output_iface, odp_actions);
5190 } else if (out_port != flow->in_port) {
5191 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5192 *nf_output_iface = out_port;
5200 static const struct ofhooks default_ofhooks = {
5201 default_normal_ofhook_cb,