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"
29 #include "classifier.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
38 #include "mac-learning.h"
39 #include "multipath.h"
45 #include "ofp-print.h"
47 #include "ofproto-sflow.h"
49 #include "openflow/nicira-ext.h"
50 #include "openflow/openflow.h"
51 #include "openvswitch/datapath-protocol.h"
55 #include "poll-loop.h"
58 #include "stream-ssl.h"
62 #include "unaligned.h"
67 VLOG_DEFINE_THIS_MODULE(ofproto);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(odp_overflow);
72 COVERAGE_DEFINE(ofproto_agg_request);
73 COVERAGE_DEFINE(ofproto_costly_flags);
74 COVERAGE_DEFINE(ofproto_ctlr_action);
75 COVERAGE_DEFINE(ofproto_del_rule);
76 COVERAGE_DEFINE(ofproto_error);
77 COVERAGE_DEFINE(ofproto_expiration);
78 COVERAGE_DEFINE(ofproto_expired);
79 COVERAGE_DEFINE(ofproto_flows_req);
80 COVERAGE_DEFINE(ofproto_flush);
81 COVERAGE_DEFINE(ofproto_invalidated);
82 COVERAGE_DEFINE(ofproto_no_packet_in);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
94 * flow translation. */
95 #define MAX_RESUBMIT_RECURSION 16
100 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
101 struct netdev *netdev;
102 struct ofp_phy_port opp; /* In host byte order. */
104 struct cfm *cfm; /* Connectivity Fault Management, if any. */
107 static void ofport_free(struct ofport *);
108 static void ofport_run(struct ofproto *, struct ofport *);
109 static void ofport_wait(struct ofport *);
111 struct action_xlate_ctx {
112 /* action_xlate_ctx_init() initializes these members. */
115 struct ofproto *ofproto;
117 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
118 * this flow when actions change header fields. */
121 /* The packet corresponding to 'flow', or a null pointer if we are
122 * revalidating without a packet to refer to. */
123 const struct ofpbuf *packet;
125 /* If nonnull, called just before executing a resubmit action.
127 * This is normally null so the client has to set it manually after
128 * calling action_xlate_ctx_init(). */
129 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
131 /* If true, the speciality of 'flow' should be checked before executing
132 * its actions. If special_cb returns false on 'flow' rendered
133 * uninstallable and no actions will be executed. */
136 /* xlate_actions() initializes and uses these members. The client might want
137 * to look at them after it returns. */
139 struct ofpbuf *odp_actions; /* Datapath actions. */
140 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
141 bool may_set_up_flow; /* True ordinarily; false if the actions must
142 * be reassessed for every packet. */
143 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
145 /* xlate_actions() initializes and uses these members, but the client has no
146 * reason to look at them. */
148 int recurse; /* Recursion level, via xlate_table_action. */
149 int last_pop_priority; /* Offset in 'odp_actions' just past most
150 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
153 static void action_xlate_ctx_init(struct action_xlate_ctx *,
154 struct ofproto *, const struct flow *,
155 const struct ofpbuf *);
156 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
157 const union ofp_action *in, size_t n_in);
159 /* An OpenFlow flow. */
161 long long int used; /* Time last used; time created if not used. */
162 long long int created; /* Creation time. */
166 * - Do include packets and bytes from facets that have been deleted or
167 * whose own statistics have been folded into the rule.
169 * - Do include packets and bytes sent "by hand" that were accounted to
170 * the rule without any facet being involved (this is a rare corner
171 * case in rule_execute()).
173 * - Do not include packet or bytes that can be obtained from any facet's
174 * packet_count or byte_count member or that can be obtained from the
175 * datapath by, e.g., dpif_flow_get() for any facet.
177 uint64_t packet_count; /* Number of packets received. */
178 uint64_t byte_count; /* Number of bytes received. */
180 ovs_be64 flow_cookie; /* Controller-issued identifier. */
182 struct cls_rule cr; /* In owning ofproto's classifier. */
183 uint16_t idle_timeout; /* In seconds from time of last use. */
184 uint16_t hard_timeout; /* In seconds from time of creation. */
185 bool send_flow_removed; /* Send a flow removed message? */
186 int n_actions; /* Number of elements in actions[]. */
187 union ofp_action *actions; /* OpenFlow actions. */
188 struct list facets; /* List of "struct facet"s. */
191 static struct rule *rule_from_cls_rule(const struct cls_rule *);
192 static bool rule_is_hidden(const struct rule *);
194 static struct rule *rule_create(const struct cls_rule *,
195 const union ofp_action *, size_t n_actions,
196 uint16_t idle_timeout, uint16_t hard_timeout,
197 ovs_be64 flow_cookie, bool send_flow_removed);
198 static void rule_destroy(struct ofproto *, struct rule *);
199 static void rule_free(struct rule *);
201 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
202 static void rule_insert(struct ofproto *, struct rule *);
203 static void rule_remove(struct ofproto *, struct rule *);
205 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
206 static void rule_get_stats(const struct rule *, uint64_t *packets,
209 /* An exact-match instantiation of an OpenFlow flow. */
211 long long int used; /* Time last used; time created if not used. */
215 * - Do include packets and bytes sent "by hand", e.g. with
218 * - Do include packets and bytes that were obtained from the datapath
219 * when a flow was deleted (e.g. dpif_flow_del()) or when its
220 * statistics were reset (e.g. dpif_flow_put() with
221 * DPIF_FP_ZERO_STATS).
223 * - Do not include any packets or bytes that can currently be obtained
224 * from the datapath by, e.g., dpif_flow_get().
226 uint64_t packet_count; /* Number of packets received. */
227 uint64_t byte_count; /* Number of bytes received. */
229 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
230 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
232 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
233 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
234 long long int rs_used; /* Used time pushed to resubmit children. */
236 /* Number of bytes passed to account_cb. This may include bytes that can
237 * currently obtained from the datapath (thus, it can be greater than
239 uint64_t accounted_bytes;
241 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
242 struct list list_node; /* In owning rule's 'facets' list. */
243 struct rule *rule; /* Owning rule. */
244 struct flow flow; /* Exact-match flow. */
245 bool installed; /* Installed in datapath? */
246 bool may_install; /* True ordinarily; false if actions must
247 * be reassessed for every packet. */
248 size_t actions_len; /* Number of bytes in actions[]. */
249 struct nlattr *actions; /* Datapath actions. */
250 tag_type tags; /* Tags (set only by hooks). */
251 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
254 static struct facet *facet_create(struct ofproto *, struct rule *,
256 const struct ofpbuf *packet);
257 static void facet_remove(struct ofproto *, struct facet *);
258 static void facet_free(struct facet *);
260 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
261 static bool facet_revalidate(struct ofproto *, struct facet *);
263 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
264 static void facet_uninstall(struct ofproto *, struct facet *);
265 static void facet_flush_stats(struct ofproto *, struct facet *);
267 static void facet_make_actions(struct ofproto *, struct facet *,
268 const struct ofpbuf *packet);
269 static void facet_update_stats(struct ofproto *, struct facet *,
270 const struct dpif_flow_stats *);
271 static void facet_push_stats(struct ofproto *, struct facet *);
273 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
274 const struct flow *, bool clone);
278 uint64_t datapath_id; /* Datapath ID. */
279 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
280 char *mfr_desc; /* Manufacturer. */
281 char *hw_desc; /* Hardware. */
282 char *sw_desc; /* Software version. */
283 char *serial_desc; /* Serial number. */
284 char *dp_desc; /* Datapath description. */
288 struct netdev_monitor *netdev_monitor;
289 struct hmap ports; /* Contains "struct ofport"s. */
290 struct shash port_by_name;
294 struct netflow *netflow;
295 struct ofproto_sflow *sflow;
298 struct classifier cls;
299 long long int next_expiration;
303 bool need_revalidate;
304 struct tag_set revalidate_set;
306 /* OpenFlow connections. */
307 struct connmgr *connmgr;
309 /* Hooks for ovs-vswitchd. */
310 const struct ofhooks *ofhooks;
313 /* Used by default ofhooks. */
314 struct mac_learning *ml;
317 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
318 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
320 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
322 static const struct ofhooks default_ofhooks;
324 static uint64_t pick_datapath_id(const struct ofproto *);
325 static uint64_t pick_fallback_dpid(void);
327 static int ofproto_expire(struct ofproto *);
328 static void flow_push_stats(struct ofproto *, const struct rule *,
329 struct flow *, uint64_t packets, uint64_t bytes,
332 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
334 static void handle_openflow(struct ofconn *, struct ofpbuf *);
336 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
337 static void update_port(struct ofproto *, const char *devname);
338 static int init_ports(struct ofproto *);
339 static void reinit_ports(struct ofproto *);
341 static void ofproto_unixctl_init(void);
344 ofproto_create(const char *datapath, const char *datapath_type,
345 const struct ofhooks *ofhooks, void *aux,
346 struct ofproto **ofprotop)
348 char local_name[IF_NAMESIZE];
355 ofproto_unixctl_init();
357 /* Connect to datapath and start listening for messages. */
358 error = dpif_open(datapath, datapath_type, &dpif);
360 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
363 error = dpif_recv_set_mask(dpif,
364 ((1u << DPIF_UC_MISS) |
365 (1u << DPIF_UC_ACTION) |
366 (1u << DPIF_UC_SAMPLE)));
368 VLOG_ERR("failed to listen on datapath %s: %s",
369 datapath, strerror(error));
373 dpif_flow_flush(dpif);
374 dpif_recv_purge(dpif);
376 error = dpif_port_get_name(dpif, ODPP_LOCAL,
377 local_name, sizeof local_name);
379 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
380 datapath, strerror(error));
384 /* Initialize settings. */
385 p = xzalloc(sizeof *p);
386 p->fallback_dpid = pick_fallback_dpid();
387 p->datapath_id = p->fallback_dpid;
388 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
389 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
390 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
391 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
392 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
394 /* Initialize datapath. */
396 p->netdev_monitor = netdev_monitor_create();
397 hmap_init(&p->ports);
398 shash_init(&p->port_by_name);
399 p->max_ports = dpif_get_max_ports(dpif);
401 /* Initialize submodules. */
405 /* Initialize flow table. */
406 classifier_init(&p->cls);
407 p->next_expiration = time_msec() + 1000;
409 /* Initialize facet table. */
410 hmap_init(&p->facets);
411 p->need_revalidate = false;
412 tag_set_init(&p->revalidate_set);
414 /* Initialize hooks. */
416 p->ofhooks = ofhooks;
420 p->ofhooks = &default_ofhooks;
422 p->ml = mac_learning_create();
425 /* Pick final datapath ID. */
426 p->datapath_id = pick_datapath_id(p);
427 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
429 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
431 /* Initialize OpenFlow connections. */
432 p->connmgr = connmgr_create(p, datapath, local_name);
439 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
441 uint64_t old_dpid = p->datapath_id;
442 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
443 if (p->datapath_id != old_dpid) {
444 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
446 /* Force all active connections to reconnect, since there is no way to
447 * notify a controller that the datapath ID has changed. */
448 ofproto_reconnect_controllers(p);
453 ofproto_set_controllers(struct ofproto *p,
454 const struct ofproto_controller *controllers,
455 size_t n_controllers)
457 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
461 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
463 connmgr_set_fail_mode(p->connmgr, fail_mode);
466 /* Drops the connections between 'ofproto' and all of its controllers, forcing
467 * them to reconnect. */
469 ofproto_reconnect_controllers(struct ofproto *ofproto)
471 connmgr_reconnect(ofproto->connmgr);
474 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
475 * in-band control should guarantee access, in the same way that in-band
476 * control guarantees access to OpenFlow controllers. */
478 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
479 const struct sockaddr_in *extras, size_t n)
481 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
484 /* Sets the OpenFlow queue used by flows set up by in-band control on
485 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
486 * flows will use the default queue. */
488 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
490 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
494 ofproto_set_desc(struct ofproto *p,
495 const char *mfr_desc, const char *hw_desc,
496 const char *sw_desc, const char *serial_desc,
499 struct ofp_desc_stats *ods;
502 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
503 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
504 sizeof ods->mfr_desc);
507 p->mfr_desc = xstrdup(mfr_desc);
510 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
511 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
512 sizeof ods->hw_desc);
515 p->hw_desc = xstrdup(hw_desc);
518 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
519 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
520 sizeof ods->sw_desc);
523 p->sw_desc = xstrdup(sw_desc);
526 if (strlen(serial_desc) >= sizeof ods->serial_num) {
527 VLOG_WARN("truncating serial_desc, must be less than %zu "
529 sizeof ods->serial_num);
531 free(p->serial_desc);
532 p->serial_desc = xstrdup(serial_desc);
535 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
536 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
537 sizeof ods->dp_desc);
540 p->dp_desc = xstrdup(dp_desc);
545 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
547 return connmgr_set_snoops(ofproto->connmgr, snoops);
551 ofproto_set_netflow(struct ofproto *ofproto,
552 const struct netflow_options *nf_options)
554 if (nf_options && nf_options->collectors.n) {
555 if (!ofproto->netflow) {
556 ofproto->netflow = netflow_create();
558 return netflow_set_options(ofproto->netflow, nf_options);
560 netflow_destroy(ofproto->netflow);
561 ofproto->netflow = NULL;
567 ofproto_set_sflow(struct ofproto *ofproto,
568 const struct ofproto_sflow_options *oso)
570 struct ofproto_sflow *os = ofproto->sflow;
573 struct ofport *ofport;
575 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
576 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
577 ofproto_sflow_add_port(os, ofport->odp_port,
578 netdev_get_name(ofport->netdev));
581 ofproto_sflow_set_options(os, oso);
583 ofproto_sflow_destroy(os);
584 ofproto->sflow = NULL;
588 /* Connectivity Fault Management configuration. */
590 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
592 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
594 struct ofport *ofport = get_port(ofproto, port_no);
595 if (ofport && ofport->cfm){
596 cfm_destroy(ofport->cfm);
601 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
602 * basic configuration from the configuration members in 'cfm', and the set of
603 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
604 * Ignores the statistics members of 'cfm'.
606 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
608 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
609 const struct cfm *cfm,
610 const uint16_t *remote_mps, size_t n_remote_mps)
612 struct ofport *ofport;
614 ofport = get_port(ofproto, port_no);
616 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
617 dpif_name(ofproto->dpif), port_no);
622 ofport->cfm = cfm_create();
625 ofport->cfm->mpid = cfm->mpid;
626 ofport->cfm->interval = cfm->interval;
627 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
629 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
631 if (!cfm_configure(ofport->cfm)) {
632 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
633 dpif_name(ofproto->dpif), port_no,
634 netdev_get_name(ofport->netdev));
635 cfm_destroy(ofport->cfm);
640 /* Returns the connectivity fault management object associated with 'port_no'
641 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
642 * 'port_no' or if that port does not have CFM configured. The caller must not
643 * modify or destroy the returned object. */
645 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
647 struct ofport *ofport = get_port(ofproto, port_no);
648 return ofport ? ofport->cfm : NULL;
652 ofproto_get_datapath_id(const struct ofproto *ofproto)
654 return ofproto->datapath_id;
658 ofproto_has_primary_controller(const struct ofproto *ofproto)
660 return connmgr_has_controllers(ofproto->connmgr);
663 enum ofproto_fail_mode
664 ofproto_get_fail_mode(const struct ofproto *p)
666 return connmgr_get_fail_mode(p->connmgr);
670 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
672 connmgr_get_snoops(ofproto->connmgr, snoops);
676 ofproto_destroy(struct ofproto *p)
678 struct ofport *ofport, *next_ofport;
684 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
686 ofproto_flush_flows(p);
687 connmgr_destroy(p->connmgr);
688 classifier_destroy(&p->cls);
689 hmap_destroy(&p->facets);
692 netdev_monitor_destroy(p->netdev_monitor);
693 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
694 hmap_remove(&p->ports, &ofport->hmap_node);
697 shash_destroy(&p->port_by_name);
699 netflow_destroy(p->netflow);
700 ofproto_sflow_destroy(p->sflow);
702 mac_learning_destroy(p->ml);
707 free(p->serial_desc);
710 hmap_destroy(&p->ports);
716 ofproto_run(struct ofproto *p)
718 int error = ofproto_run1(p);
720 error = ofproto_run2(p, false);
726 process_port_change(struct ofproto *ofproto, int error, char *devname)
728 if (error == ENOBUFS) {
729 reinit_ports(ofproto);
731 update_port(ofproto, devname);
737 ofproto_run1(struct ofproto *p)
739 struct ofport *ofport;
744 if (shash_is_empty(&p->port_by_name)) {
748 for (i = 0; i < 50; i++) {
749 struct dpif_upcall packet;
751 error = dpif_recv(p->dpif, &packet);
753 if (error == ENODEV) {
754 /* Someone destroyed the datapath behind our back. The caller
755 * better destroy us and give up, because we're just going to
756 * spin from here on out. */
757 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
758 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
765 handle_upcall(p, &packet);
768 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
769 process_port_change(p, error, devname);
771 while ((error = netdev_monitor_poll(p->netdev_monitor,
772 &devname)) != EAGAIN) {
773 process_port_change(p, error, devname);
776 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
777 ofport_run(p, ofport);
780 connmgr_run(p->connmgr, handle_openflow);
782 if (time_msec() >= p->next_expiration) {
783 int delay = ofproto_expire(p);
784 p->next_expiration = time_msec() + delay;
785 COVERAGE_INC(ofproto_expiration);
789 netflow_run(p->netflow);
792 ofproto_sflow_run(p->sflow);
799 ofproto_run2(struct ofproto *p, bool revalidate_all)
801 /* Figure out what we need to revalidate now, if anything. */
802 struct tag_set revalidate_set = p->revalidate_set;
803 if (p->need_revalidate) {
804 revalidate_all = true;
807 /* Clear the revalidation flags. */
808 tag_set_init(&p->revalidate_set);
809 p->need_revalidate = false;
811 /* Now revalidate if there's anything to do. */
812 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
813 struct facet *facet, *next;
815 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
817 || tag_set_intersects(&revalidate_set, facet->tags)) {
818 facet_revalidate(p, facet);
827 ofproto_wait(struct ofproto *p)
829 struct ofport *ofport;
831 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
834 dpif_recv_wait(p->dpif);
835 dpif_port_poll_wait(p->dpif);
836 netdev_monitor_poll_wait(p->netdev_monitor);
838 ofproto_sflow_wait(p->sflow);
840 if (!tag_set_is_empty(&p->revalidate_set)) {
841 poll_immediate_wake();
843 if (p->need_revalidate) {
844 /* Shouldn't happen, but if it does just go around again. */
845 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
846 poll_immediate_wake();
847 } else if (p->next_expiration != LLONG_MAX) {
848 poll_timer_wait_until(p->next_expiration);
850 connmgr_wait(p->connmgr);
854 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
856 tag_set_add(&ofproto->revalidate_set, tag);
860 ofproto_get_revalidate_set(struct ofproto *ofproto)
862 return &ofproto->revalidate_set;
866 ofproto_is_alive(const struct ofproto *p)
868 return connmgr_has_controllers(p->connmgr);
872 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
875 connmgr_get_controller_info(ofproto->connmgr, info);
879 ofproto_free_ofproto_controller_info(struct shash *info)
881 struct shash_node *node;
883 SHASH_FOR_EACH (node, info) {
884 struct ofproto_controller_info *cinfo = node->data;
885 while (cinfo->pairs.n) {
886 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
893 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
895 * This is almost the same as calling dpif_port_del() directly on the
896 * datapath, but it also makes 'ofproto' close its open netdev for the port
897 * (if any). This makes it possible to create a new netdev of a different
898 * type under the same name, which otherwise the netdev library would refuse
899 * to do because of the conflict. (The netdev would eventually get closed on
900 * the next trip through ofproto_run(), but this interface is more direct.)
902 * Returns 0 if successful, otherwise a positive errno. */
904 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
906 struct ofport *ofport = get_port(ofproto, odp_port);
907 const char *name = ofport ? ofport->opp.name : "<unknown>";
910 error = dpif_port_del(ofproto->dpif, odp_port);
912 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
913 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
915 /* 'name' is ofport->opp.name and update_port() is going to destroy
916 * 'ofport'. Just in case update_port() refers to 'name' after it
917 * destroys 'ofport', make a copy of it around the update_port()
919 char *devname = xstrdup(name);
920 update_port(ofproto, devname);
926 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
927 * true if 'odp_port' exists and should be included, false otherwise. */
929 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
931 struct ofport *ofport = get_port(ofproto, odp_port);
932 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
935 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
936 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
937 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
939 * Returns 0 if successful, otherwise a positive errno value. */
941 ofproto_send_packet(struct ofproto *ofproto,
942 uint32_t port_no, uint16_t vlan_tci,
943 const struct ofpbuf *packet)
945 struct ofpbuf odp_actions;
948 ofpbuf_init(&odp_actions, 32);
950 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
951 ntohs(vlan_tci & ~VLAN_CFI));
953 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
954 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
956 ofpbuf_uninit(&odp_actions);
959 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
960 dpif_name(ofproto->dpif), port_no, strerror(error));
965 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
966 * performs the 'n_actions' actions in 'actions'. The new flow will not
969 * If cls_rule->priority is in the range of priorities supported by OpenFlow
970 * (0...65535, inclusive) then the flow will be visible to OpenFlow
971 * controllers; otherwise, it will be hidden.
973 * The caller retains ownership of 'cls_rule' and 'actions'. */
975 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
976 const union ofp_action *actions, size_t n_actions)
979 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
980 rule_insert(p, rule);
984 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
988 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
991 rule_remove(ofproto, rule);
996 ofproto_flush_flows(struct ofproto *ofproto)
998 struct facet *facet, *next_facet;
999 struct rule *rule, *next_rule;
1000 struct cls_cursor cursor;
1002 COVERAGE_INC(ofproto_flush);
1004 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1005 /* Mark the facet as not installed so that facet_remove() doesn't
1006 * bother trying to uninstall it. There is no point in uninstalling it
1007 * individually since we are about to blow away all the facets with
1008 * dpif_flow_flush(). */
1009 facet->installed = false;
1010 facet->dp_packet_count = 0;
1011 facet->dp_byte_count = 0;
1012 facet_remove(ofproto, facet);
1015 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1016 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1017 rule_remove(ofproto, rule);
1020 dpif_flow_flush(ofproto->dpif);
1021 connmgr_flushed(ofproto->connmgr);
1025 reinit_ports(struct ofproto *p)
1027 struct dpif_port_dump dump;
1028 struct shash_node *node;
1029 struct shash devnames;
1030 struct ofport *ofport;
1031 struct dpif_port dpif_port;
1033 COVERAGE_INC(ofproto_reinit_ports);
1035 shash_init(&devnames);
1036 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1037 shash_add_once (&devnames, ofport->opp.name, NULL);
1039 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1040 shash_add_once (&devnames, dpif_port.name, NULL);
1043 SHASH_FOR_EACH (node, &devnames) {
1044 update_port(p, node->name);
1046 shash_destroy(&devnames);
1049 static struct ofport *
1050 make_ofport(const struct dpif_port *dpif_port)
1052 struct netdev_options netdev_options;
1053 enum netdev_flags flags;
1054 struct ofport *ofport;
1055 struct netdev *netdev;
1058 memset(&netdev_options, 0, sizeof netdev_options);
1059 netdev_options.name = dpif_port->name;
1060 netdev_options.type = dpif_port->type;
1061 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1063 error = netdev_open(&netdev_options, &netdev);
1065 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1066 "cannot be opened (%s)",
1067 dpif_port->name, dpif_port->port_no,
1068 dpif_port->name, strerror(error));
1072 ofport = xzalloc(sizeof *ofport);
1073 ofport->netdev = netdev;
1074 ofport->odp_port = dpif_port->port_no;
1075 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1076 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1077 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1079 netdev_get_flags(netdev, &flags);
1080 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1082 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1084 netdev_get_features(netdev,
1085 &ofport->opp.curr, &ofport->opp.advertised,
1086 &ofport->opp.supported, &ofport->opp.peer);
1091 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1093 if (get_port(p, dpif_port->port_no)) {
1094 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1095 dpif_port->port_no);
1097 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1098 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1107 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1109 const struct ofp_phy_port *a = &a_->opp;
1110 const struct ofp_phy_port *b = &b_->opp;
1112 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1113 return (a->port_no == b->port_no
1114 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1115 && !strcmp(a->name, b->name)
1116 && a->state == b->state
1117 && a->config == b->config
1118 && a->curr == b->curr
1119 && a->advertised == b->advertised
1120 && a->supported == b->supported
1121 && a->peer == b->peer);
1125 ofport_install(struct ofproto *p, struct ofport *ofport)
1127 const char *netdev_name = ofport->opp.name;
1129 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1130 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1131 shash_add(&p->port_by_name, netdev_name, ofport);
1133 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1138 ofport_remove(struct ofproto *p, struct ofport *ofport)
1140 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1141 hmap_remove(&p->ports, &ofport->hmap_node);
1142 shash_delete(&p->port_by_name,
1143 shash_find(&p->port_by_name, ofport->opp.name));
1145 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1150 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1153 cfm_run(ofport->cfm);
1155 if (cfm_should_send_ccm(ofport->cfm)) {
1156 struct ofpbuf packet;
1159 ofpbuf_init(&packet, 0);
1160 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1161 ETH_TYPE_CFM, sizeof *ccm);
1162 cfm_compose_ccm(ofport->cfm, ccm);
1163 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1164 ofpbuf_uninit(&packet);
1170 ofport_wait(struct ofport *ofport)
1173 cfm_wait(ofport->cfm);
1178 ofport_free(struct ofport *ofport)
1181 cfm_destroy(ofport->cfm);
1182 netdev_close(ofport->netdev);
1187 static struct ofport *
1188 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1190 struct ofport *port;
1192 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1193 hash_int(odp_port, 0), &ofproto->ports) {
1194 if (port->odp_port == odp_port) {
1202 update_port(struct ofproto *p, const char *devname)
1204 struct dpif_port dpif_port;
1205 struct ofport *old_ofport;
1206 struct ofport *new_ofport;
1209 COVERAGE_INC(ofproto_update_port);
1211 /* Query the datapath for port information. */
1212 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1214 /* Find the old ofport. */
1215 old_ofport = shash_find_data(&p->port_by_name, devname);
1218 /* There's no port named 'devname' but there might be a port with
1219 * the same port number. This could happen if a port is deleted
1220 * and then a new one added in its place very quickly, or if a port
1221 * is renamed. In the former case we want to send an OFPPR_DELETE
1222 * and an OFPPR_ADD, and in the latter case we want to send a
1223 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1224 * the old port's ifindex against the new port, or perhaps less
1225 * reliably but more portably by comparing the old port's MAC
1226 * against the new port's MAC. However, this code isn't that smart
1227 * and always sends an OFPPR_MODIFY (XXX). */
1228 old_ofport = get_port(p, dpif_port.port_no);
1230 } else if (error != ENOENT && error != ENODEV) {
1231 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1232 "%s", strerror(error));
1236 /* Create a new ofport. */
1237 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1239 /* Eliminate a few pathological cases. */
1240 if (!old_ofport && !new_ofport) {
1242 } else if (old_ofport && new_ofport) {
1243 /* Most of the 'config' bits are OpenFlow soft state, but
1244 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1245 * OpenFlow bits from old_ofport. (make_ofport() only sets
1246 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1247 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1249 if (ofport_equal(old_ofport, new_ofport)) {
1250 /* False alarm--no change. */
1251 ofport_free(new_ofport);
1256 /* Now deal with the normal cases. */
1258 ofport_remove(p, old_ofport);
1261 ofport_install(p, new_ofport);
1263 connmgr_send_port_status(p->connmgr,
1264 new_ofport ? &new_ofport->opp : &old_ofport->opp,
1265 (!old_ofport ? OFPPR_ADD
1266 : !new_ofport ? OFPPR_DELETE
1268 ofport_free(old_ofport);
1271 dpif_port_destroy(&dpif_port);
1275 init_ports(struct ofproto *p)
1277 struct dpif_port_dump dump;
1278 struct dpif_port dpif_port;
1280 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1281 if (!ofport_conflicts(p, &dpif_port)) {
1282 struct ofport *ofport = make_ofport(&dpif_port);
1284 ofport_install(p, ofport);
1292 /* Returns true if 'rule' should be hidden from the controller.
1294 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1295 * (e.g. by in-band control) and are intentionally hidden from the
1298 rule_is_hidden(const struct rule *rule)
1300 return rule->cr.priority > UINT16_MAX;
1303 /* Creates and returns a new rule initialized as specified.
1305 * The caller is responsible for inserting the rule into the classifier (with
1306 * rule_insert()). */
1307 static struct rule *
1308 rule_create(const struct cls_rule *cls_rule,
1309 const union ofp_action *actions, size_t n_actions,
1310 uint16_t idle_timeout, uint16_t hard_timeout,
1311 ovs_be64 flow_cookie, bool send_flow_removed)
1313 struct rule *rule = xzalloc(sizeof *rule);
1314 rule->cr = *cls_rule;
1315 rule->idle_timeout = idle_timeout;
1316 rule->hard_timeout = hard_timeout;
1317 rule->flow_cookie = flow_cookie;
1318 rule->used = rule->created = time_msec();
1319 rule->send_flow_removed = send_flow_removed;
1320 list_init(&rule->facets);
1321 if (n_actions > 0) {
1322 rule->n_actions = n_actions;
1323 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1329 static struct rule *
1330 rule_from_cls_rule(const struct cls_rule *cls_rule)
1332 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1336 rule_free(struct rule *rule)
1338 free(rule->actions);
1342 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1343 * destroying any that no longer has a rule (which is probably all of them).
1345 * The caller must have already removed 'rule' from the classifier. */
1347 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1349 struct facet *facet, *next_facet;
1350 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1351 facet_revalidate(ofproto, facet);
1356 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1357 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1360 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1362 const union ofp_action *oa;
1363 struct actions_iterator i;
1365 if (out_port == htons(OFPP_NONE)) {
1368 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1369 oa = actions_next(&i)) {
1370 if (action_outputs_to_port(oa, out_port)) {
1377 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1378 * 'packet', which arrived on 'in_port'.
1380 * Takes ownership of 'packet'. */
1382 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1383 const struct nlattr *odp_actions, size_t actions_len,
1384 struct ofpbuf *packet)
1386 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1387 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1388 /* As an optimization, avoid a round-trip from userspace to kernel to
1389 * userspace. This also avoids possibly filling up kernel packet
1390 * buffers along the way. */
1391 struct dpif_upcall upcall;
1393 upcall.type = DPIF_UC_ACTION;
1394 upcall.packet = packet;
1397 upcall.userdata = nl_attr_get_u64(odp_actions);
1398 upcall.sample_pool = 0;
1399 upcall.actions = NULL;
1400 upcall.actions_len = 0;
1402 send_packet_in(ofproto, &upcall, flow, false);
1408 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1409 ofpbuf_delete(packet);
1414 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1415 * statistics appropriately. 'packet' must have at least sizeof(struct
1416 * ofp_packet_in) bytes of headroom.
1418 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1419 * applying flow_extract() to 'packet' would yield the same flow as
1422 * 'facet' must have accurately composed ODP actions; that is, it must not be
1423 * in need of revalidation.
1425 * Takes ownership of 'packet'. */
1427 facet_execute(struct ofproto *ofproto, struct facet *facet,
1428 struct ofpbuf *packet)
1430 struct dpif_flow_stats stats;
1432 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1434 flow_extract_stats(&facet->flow, packet, &stats);
1435 stats.used = time_msec();
1436 if (execute_odp_actions(ofproto, &facet->flow,
1437 facet->actions, facet->actions_len, packet)) {
1438 facet_update_stats(ofproto, facet, &stats);
1442 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1443 * statistics (or the statistics for one of its facets) appropriately.
1444 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1446 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1447 * with statistics for 'packet' either way.
1449 * Takes ownership of 'packet'. */
1451 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1452 struct ofpbuf *packet)
1454 struct action_xlate_ctx ctx;
1455 struct ofpbuf *odp_actions;
1456 struct facet *facet;
1460 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1462 flow_extract(packet, 0, in_port, &flow);
1464 /* First look for a related facet. If we find one, account it to that. */
1465 facet = facet_lookup_valid(ofproto, &flow);
1466 if (facet && facet->rule == rule) {
1467 facet_execute(ofproto, facet, packet);
1471 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1472 * create a new facet for it and use that. */
1473 if (rule_lookup(ofproto, &flow) == rule) {
1474 facet = facet_create(ofproto, rule, &flow, packet);
1475 facet_execute(ofproto, facet, packet);
1476 facet_install(ofproto, facet, true);
1480 /* We can't account anything to a facet. If we were to try, then that
1481 * facet would have a non-matching rule, busting our invariants. */
1482 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1483 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1484 size = packet->size;
1485 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1486 odp_actions->size, packet)) {
1487 rule->used = time_msec();
1488 rule->packet_count++;
1489 rule->byte_count += size;
1490 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1492 ofpbuf_delete(odp_actions);
1495 /* Inserts 'rule' into 'p''s flow table. */
1497 rule_insert(struct ofproto *p, struct rule *rule)
1499 struct rule *displaced_rule;
1501 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1502 if (displaced_rule) {
1503 rule_destroy(p, displaced_rule);
1505 p->need_revalidate = true;
1508 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1509 * 'flow' and an example 'packet' within that flow.
1511 * The caller must already have determined that no facet with an identical
1512 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1513 * 'ofproto''s classifier table. */
1514 static struct facet *
1515 facet_create(struct ofproto *ofproto, struct rule *rule,
1516 const struct flow *flow, const struct ofpbuf *packet)
1518 struct facet *facet;
1520 facet = xzalloc(sizeof *facet);
1521 facet->used = time_msec();
1522 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1523 list_push_back(&rule->facets, &facet->list_node);
1525 facet->flow = *flow;
1526 netflow_flow_init(&facet->nf_flow);
1527 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1529 facet_make_actions(ofproto, facet, packet);
1535 facet_free(struct facet *facet)
1537 free(facet->actions);
1541 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1543 * - Removes 'rule' from the classifier.
1545 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1546 * destroys them), via rule_destroy().
1549 rule_remove(struct ofproto *ofproto, struct rule *rule)
1551 COVERAGE_INC(ofproto_del_rule);
1552 ofproto->need_revalidate = true;
1553 classifier_remove(&ofproto->cls, &rule->cr);
1554 rule_destroy(ofproto, rule);
1557 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1559 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1560 * rule's statistics, via facet_uninstall().
1562 * - Removes 'facet' from its rule and from ofproto->facets.
1565 facet_remove(struct ofproto *ofproto, struct facet *facet)
1567 facet_uninstall(ofproto, facet);
1568 facet_flush_stats(ofproto, facet);
1569 hmap_remove(&ofproto->facets, &facet->hmap_node);
1570 list_remove(&facet->list_node);
1574 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1576 facet_make_actions(struct ofproto *p, struct facet *facet,
1577 const struct ofpbuf *packet)
1579 const struct rule *rule = facet->rule;
1580 struct ofpbuf *odp_actions;
1581 struct action_xlate_ctx ctx;
1583 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1584 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1585 facet->tags = ctx.tags;
1586 facet->may_install = ctx.may_set_up_flow;
1587 facet->nf_flow.output_iface = ctx.nf_output_iface;
1589 if (facet->actions_len != odp_actions->size
1590 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1591 free(facet->actions);
1592 facet->actions_len = odp_actions->size;
1593 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1596 ofpbuf_delete(odp_actions);
1600 facet_put__(struct ofproto *ofproto, struct facet *facet,
1601 const struct nlattr *actions, size_t actions_len,
1602 struct dpif_flow_stats *stats)
1604 struct odputil_keybuf keybuf;
1605 enum dpif_flow_put_flags flags;
1608 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1610 flags |= DPIF_FP_ZERO_STATS;
1611 facet->dp_packet_count = 0;
1612 facet->dp_byte_count = 0;
1615 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1616 odp_flow_key_from_flow(&key, &facet->flow);
1618 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1619 actions, actions_len, stats);
1622 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1623 * 'zero_stats' is true, clears any existing statistics from the datapath for
1626 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1628 struct dpif_flow_stats stats;
1630 if (facet->may_install
1631 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1632 zero_stats ? &stats : NULL)) {
1633 facet->installed = true;
1637 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1638 * to the accounting hook function in the ofhooks structure. */
1640 facet_account(struct ofproto *ofproto,
1641 struct facet *facet, uint64_t extra_bytes)
1643 uint64_t total_bytes = facet->byte_count + extra_bytes;
1645 if (ofproto->ofhooks->account_flow_cb
1646 && total_bytes > facet->accounted_bytes)
1648 ofproto->ofhooks->account_flow_cb(
1649 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1650 total_bytes - facet->accounted_bytes, ofproto->aux);
1651 facet->accounted_bytes = total_bytes;
1655 /* If 'rule' is installed in the datapath, uninstalls it. */
1657 facet_uninstall(struct ofproto *p, struct facet *facet)
1659 if (facet->installed) {
1660 struct odputil_keybuf keybuf;
1661 struct dpif_flow_stats stats;
1664 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1665 odp_flow_key_from_flow(&key, &facet->flow);
1667 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1668 facet_update_stats(p, facet, &stats);
1670 facet->installed = false;
1671 facet->dp_packet_count = 0;
1672 facet->dp_byte_count = 0;
1674 assert(facet->dp_packet_count == 0);
1675 assert(facet->dp_byte_count == 0);
1679 /* Returns true if the only action for 'facet' is to send to the controller.
1680 * (We don't report NetFlow expiration messages for such facets because they
1681 * are just part of the control logic for the network, not real traffic). */
1683 facet_is_controller_flow(struct facet *facet)
1686 && facet->rule->n_actions == 1
1687 && action_outputs_to_port(&facet->rule->actions[0],
1688 htons(OFPP_CONTROLLER)));
1691 /* Folds all of 'facet''s statistics into its rule. Also updates the
1692 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1693 * 'facet''s statistics in the datapath should have been zeroed and folded into
1694 * its packet and byte counts before this function is called. */
1696 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1698 assert(!facet->dp_byte_count);
1699 assert(!facet->dp_packet_count);
1701 facet_push_stats(ofproto, facet);
1702 facet_account(ofproto, facet, 0);
1704 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1705 struct ofexpired expired;
1706 expired.flow = facet->flow;
1707 expired.packet_count = facet->packet_count;
1708 expired.byte_count = facet->byte_count;
1709 expired.used = facet->used;
1710 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1713 facet->rule->packet_count += facet->packet_count;
1714 facet->rule->byte_count += facet->byte_count;
1716 /* Reset counters to prevent double counting if 'facet' ever gets
1718 facet->packet_count = 0;
1719 facet->byte_count = 0;
1720 facet->rs_packet_count = 0;
1721 facet->rs_byte_count = 0;
1722 facet->accounted_bytes = 0;
1724 netflow_flow_clear(&facet->nf_flow);
1727 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1728 * Returns it if found, otherwise a null pointer.
1730 * The returned facet might need revalidation; use facet_lookup_valid()
1731 * instead if that is important. */
1732 static struct facet *
1733 facet_find(struct ofproto *ofproto, const struct flow *flow)
1735 struct facet *facet;
1737 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1739 if (flow_equal(flow, &facet->flow)) {
1747 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1748 * Returns it if found, otherwise a null pointer.
1750 * The returned facet is guaranteed to be valid. */
1751 static struct facet *
1752 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1754 struct facet *facet = facet_find(ofproto, flow);
1756 /* The facet we found might not be valid, since we could be in need of
1757 * revalidation. If it is not valid, don't return it. */
1759 && ofproto->need_revalidate
1760 && !facet_revalidate(ofproto, facet)) {
1761 COVERAGE_INC(ofproto_invalidated);
1768 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1770 * - If the rule found is different from 'facet''s current rule, moves
1771 * 'facet' to the new rule and recompiles its actions.
1773 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1774 * where it is and recompiles its actions anyway.
1776 * - If there is none, destroys 'facet'.
1778 * Returns true if 'facet' still exists, false if it has been destroyed. */
1780 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1782 struct action_xlate_ctx ctx;
1783 struct ofpbuf *odp_actions;
1784 struct rule *new_rule;
1785 bool actions_changed;
1787 COVERAGE_INC(facet_revalidate);
1789 /* Determine the new rule. */
1790 new_rule = rule_lookup(ofproto, &facet->flow);
1792 /* No new rule, so delete the facet. */
1793 facet_remove(ofproto, facet);
1797 /* Calculate new ODP actions.
1799 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1800 * emit a NetFlow expiration and, if so, we need to have the old state
1801 * around to properly compose it. */
1802 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1803 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1804 actions_changed = (facet->actions_len != odp_actions->size
1805 || memcmp(facet->actions, odp_actions->data,
1806 facet->actions_len));
1808 /* If the ODP actions changed or the installability changed, then we need
1809 * to talk to the datapath. */
1810 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
1811 if (ctx.may_set_up_flow) {
1812 struct dpif_flow_stats stats;
1814 facet_put__(ofproto, facet,
1815 odp_actions->data, odp_actions->size, &stats);
1816 facet_update_stats(ofproto, facet, &stats);
1818 facet_uninstall(ofproto, facet);
1821 /* The datapath flow is gone or has zeroed stats, so push stats out of
1822 * 'facet' into 'rule'. */
1823 facet_flush_stats(ofproto, facet);
1826 /* Update 'facet' now that we've taken care of all the old state. */
1827 facet->tags = ctx.tags;
1828 facet->nf_flow.output_iface = ctx.nf_output_iface;
1829 facet->may_install = ctx.may_set_up_flow;
1830 if (actions_changed) {
1831 free(facet->actions);
1832 facet->actions_len = odp_actions->size;
1833 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1835 if (facet->rule != new_rule) {
1836 COVERAGE_INC(facet_changed_rule);
1837 list_remove(&facet->list_node);
1838 list_push_back(&new_rule->facets, &facet->list_node);
1839 facet->rule = new_rule;
1840 facet->used = new_rule->created;
1841 facet->rs_used = facet->used;
1844 ofpbuf_delete(odp_actions);
1850 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1853 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
1855 COVERAGE_INC(ofproto_error);
1856 ofconn_send_reply(ofconn, buf);
1861 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1863 ofconn_send_reply(ofconn, make_echo_reply(oh));
1868 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1870 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1871 struct ofp_switch_features *osf;
1873 struct ofport *port;
1875 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1876 osf->datapath_id = htonll(ofproto->datapath_id);
1877 osf->n_buffers = htonl(pktbuf_capacity());
1879 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1880 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1881 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1882 (1u << OFPAT_SET_VLAN_VID) |
1883 (1u << OFPAT_SET_VLAN_PCP) |
1884 (1u << OFPAT_STRIP_VLAN) |
1885 (1u << OFPAT_SET_DL_SRC) |
1886 (1u << OFPAT_SET_DL_DST) |
1887 (1u << OFPAT_SET_NW_SRC) |
1888 (1u << OFPAT_SET_NW_DST) |
1889 (1u << OFPAT_SET_NW_TOS) |
1890 (1u << OFPAT_SET_TP_SRC) |
1891 (1u << OFPAT_SET_TP_DST) |
1892 (1u << OFPAT_ENQUEUE));
1894 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1895 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1898 ofconn_send_reply(ofconn, buf);
1903 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1905 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1907 struct ofp_switch_config *osc;
1911 /* Figure out flags. */
1912 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
1913 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1916 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1917 osc->flags = htons(flags);
1918 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1919 ofconn_send_reply(ofconn, buf);
1925 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1927 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1928 uint16_t flags = ntohs(osc->flags);
1930 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1931 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1932 switch (flags & OFPC_FRAG_MASK) {
1933 case OFPC_FRAG_NORMAL:
1934 dpif_set_drop_frags(ofproto->dpif, false);
1936 case OFPC_FRAG_DROP:
1937 dpif_set_drop_frags(ofproto->dpif, true);
1940 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1946 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1951 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1952 struct action_xlate_ctx *ctx);
1955 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1957 const struct ofport *ofport = get_port(ctx->ofproto, port);
1960 if (ofport->opp.config & OFPPC_NO_FWD) {
1961 /* Forwarding disabled on port. */
1966 * We don't have an ofport record for this port, but it doesn't hurt to
1967 * allow forwarding to it anyhow. Maybe such a port will appear later
1968 * and we're pre-populating the flow table.
1972 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
1973 ctx->nf_output_iface = port;
1976 static struct rule *
1977 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
1979 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1983 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1985 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1986 uint16_t old_in_port;
1989 /* Look up a flow with 'in_port' as the input port. Then restore the
1990 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1991 * have surprising behavior). */
1992 old_in_port = ctx->flow.in_port;
1993 ctx->flow.in_port = in_port;
1994 rule = rule_lookup(ctx->ofproto, &ctx->flow);
1995 ctx->flow.in_port = old_in_port;
1997 if (ctx->resubmit_hook) {
1998 ctx->resubmit_hook(ctx, rule);
2003 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2007 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2009 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2010 MAX_RESUBMIT_RECURSION);
2015 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2016 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2018 struct ofport *ofport;
2020 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2021 uint16_t odp_port = ofport->odp_port;
2022 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2023 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2026 *nf_output_iface = NF_OUT_FLOOD;
2030 xlate_output_action__(struct action_xlate_ctx *ctx,
2031 uint16_t port, uint16_t max_len)
2034 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2036 ctx->nf_output_iface = NF_OUT_DROP;
2040 add_output_action(ctx, ctx->flow.in_port);
2043 xlate_table_action(ctx, ctx->flow.in_port);
2046 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2047 ctx->odp_actions, &ctx->tags,
2048 &ctx->nf_output_iface,
2049 ctx->ofproto->aux)) {
2050 COVERAGE_INC(ofproto_uninstallable);
2051 ctx->may_set_up_flow = false;
2055 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2056 &ctx->nf_output_iface, ctx->odp_actions);
2059 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2060 &ctx->nf_output_iface, ctx->odp_actions);
2062 case OFPP_CONTROLLER:
2063 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2066 add_output_action(ctx, ODPP_LOCAL);
2069 odp_port = ofp_port_to_odp_port(port);
2070 if (odp_port != ctx->flow.in_port) {
2071 add_output_action(ctx, odp_port);
2076 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2077 ctx->nf_output_iface = NF_OUT_FLOOD;
2078 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2079 ctx->nf_output_iface = prev_nf_output_iface;
2080 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2081 ctx->nf_output_iface != NF_OUT_FLOOD) {
2082 ctx->nf_output_iface = NF_OUT_MULTI;
2087 xlate_output_action(struct action_xlate_ctx *ctx,
2088 const struct ofp_action_output *oao)
2090 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2093 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2094 * optimization, because we're going to add another action that sets the
2095 * priority immediately after, or because there are no actions following the
2098 remove_pop_action(struct action_xlate_ctx *ctx)
2100 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2101 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2102 ctx->last_pop_priority = -1;
2107 add_pop_action(struct action_xlate_ctx *ctx)
2109 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2110 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2111 ctx->last_pop_priority = ctx->odp_actions->size;
2116 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2117 const struct ofp_action_enqueue *oae)
2119 uint16_t ofp_port, odp_port;
2123 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2126 /* Fall back to ordinary output action. */
2127 xlate_output_action__(ctx, ntohs(oae->port), 0);
2131 /* Figure out ODP output port. */
2132 ofp_port = ntohs(oae->port);
2133 if (ofp_port != OFPP_IN_PORT) {
2134 odp_port = ofp_port_to_odp_port(ofp_port);
2136 odp_port = ctx->flow.in_port;
2139 /* Add ODP actions. */
2140 remove_pop_action(ctx);
2141 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2142 add_output_action(ctx, odp_port);
2143 add_pop_action(ctx);
2145 /* Update NetFlow output port. */
2146 if (ctx->nf_output_iface == NF_OUT_DROP) {
2147 ctx->nf_output_iface = odp_port;
2148 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2149 ctx->nf_output_iface = NF_OUT_MULTI;
2154 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2155 const struct nx_action_set_queue *nasq)
2160 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2163 /* Couldn't translate queue to a priority, so ignore. A warning
2164 * has already been logged. */
2168 remove_pop_action(ctx);
2169 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2173 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2175 ovs_be16 tci = ctx->flow.vlan_tci;
2176 if (!(tci & htons(VLAN_CFI))) {
2177 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2179 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2180 tci & ~htons(VLAN_CFI));
2184 struct xlate_reg_state {
2190 save_reg_state(const struct action_xlate_ctx *ctx,
2191 struct xlate_reg_state *state)
2193 state->vlan_tci = ctx->flow.vlan_tci;
2194 state->tun_id = ctx->flow.tun_id;
2198 update_reg_state(struct action_xlate_ctx *ctx,
2199 const struct xlate_reg_state *state)
2201 if (ctx->flow.vlan_tci != state->vlan_tci) {
2202 xlate_set_dl_tci(ctx);
2204 if (ctx->flow.tun_id != state->tun_id) {
2205 nl_msg_put_be64(ctx->odp_actions,
2206 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2211 xlate_nicira_action(struct action_xlate_ctx *ctx,
2212 const struct nx_action_header *nah)
2214 const struct nx_action_resubmit *nar;
2215 const struct nx_action_set_tunnel *nast;
2216 const struct nx_action_set_queue *nasq;
2217 const struct nx_action_multipath *nam;
2218 enum nx_action_subtype subtype = ntohs(nah->subtype);
2219 struct xlate_reg_state state;
2222 assert(nah->vendor == htonl(NX_VENDOR_ID));
2224 case NXAST_RESUBMIT:
2225 nar = (const struct nx_action_resubmit *) nah;
2226 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2229 case NXAST_SET_TUNNEL:
2230 nast = (const struct nx_action_set_tunnel *) nah;
2231 tun_id = htonll(ntohl(nast->tun_id));
2232 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2233 ctx->flow.tun_id = tun_id;
2236 case NXAST_DROP_SPOOFED_ARP:
2237 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2238 nl_msg_put_flag(ctx->odp_actions,
2239 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2243 case NXAST_SET_QUEUE:
2244 nasq = (const struct nx_action_set_queue *) nah;
2245 xlate_set_queue_action(ctx, nasq);
2248 case NXAST_POP_QUEUE:
2249 add_pop_action(ctx);
2252 case NXAST_REG_MOVE:
2253 save_reg_state(ctx, &state);
2254 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2256 update_reg_state(ctx, &state);
2259 case NXAST_REG_LOAD:
2260 save_reg_state(ctx, &state);
2261 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2263 update_reg_state(ctx, &state);
2267 /* Nothing to do. */
2270 case NXAST_SET_TUNNEL64:
2271 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2272 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2273 ctx->flow.tun_id = tun_id;
2276 case NXAST_MULTIPATH:
2277 nam = (const struct nx_action_multipath *) nah;
2278 multipath_execute(nam, &ctx->flow);
2281 /* If you add a new action here that modifies flow data, don't forget to
2282 * update the flow key in ctx->flow at the same time. */
2284 case NXAST_SNAT__OBSOLETE:
2286 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2292 do_xlate_actions(const union ofp_action *in, size_t n_in,
2293 struct action_xlate_ctx *ctx)
2295 struct actions_iterator iter;
2296 const union ofp_action *ia;
2297 const struct ofport *port;
2299 port = get_port(ctx->ofproto, ctx->flow.in_port);
2300 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2301 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2302 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2303 /* Drop this flow. */
2307 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2308 enum ofp_action_type type = ntohs(ia->type);
2309 const struct ofp_action_dl_addr *oada;
2313 xlate_output_action(ctx, &ia->output);
2316 case OFPAT_SET_VLAN_VID:
2317 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2318 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2319 xlate_set_dl_tci(ctx);
2322 case OFPAT_SET_VLAN_PCP:
2323 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2324 ctx->flow.vlan_tci |= htons(
2325 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2326 xlate_set_dl_tci(ctx);
2329 case OFPAT_STRIP_VLAN:
2330 ctx->flow.vlan_tci = htons(0);
2331 xlate_set_dl_tci(ctx);
2334 case OFPAT_SET_DL_SRC:
2335 oada = ((struct ofp_action_dl_addr *) ia);
2336 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2337 oada->dl_addr, ETH_ADDR_LEN);
2338 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2341 case OFPAT_SET_DL_DST:
2342 oada = ((struct ofp_action_dl_addr *) ia);
2343 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2344 oada->dl_addr, ETH_ADDR_LEN);
2345 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2348 case OFPAT_SET_NW_SRC:
2349 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2350 ia->nw_addr.nw_addr);
2351 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2354 case OFPAT_SET_NW_DST:
2355 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2356 ia->nw_addr.nw_addr);
2357 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2360 case OFPAT_SET_NW_TOS:
2361 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2363 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2366 case OFPAT_SET_TP_SRC:
2367 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2368 ia->tp_port.tp_port);
2369 ctx->flow.tp_src = ia->tp_port.tp_port;
2372 case OFPAT_SET_TP_DST:
2373 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2374 ia->tp_port.tp_port);
2375 ctx->flow.tp_dst = ia->tp_port.tp_port;
2379 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2383 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2387 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2394 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2395 struct ofproto *ofproto, const struct flow *flow,
2396 const struct ofpbuf *packet)
2398 ctx->ofproto = ofproto;
2400 ctx->packet = packet;
2401 ctx->resubmit_hook = NULL;
2402 ctx->check_special = true;
2406 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2407 const struct ofpbuf *packet)
2409 struct ofport *ofport;
2411 ofport = get_port(ofproto, flow->in_port);
2412 if (ofport && ofport->cfm) {
2413 cfm_process_heartbeat(ofport->cfm, packet);
2417 static struct ofpbuf *
2418 xlate_actions(struct action_xlate_ctx *ctx,
2419 const union ofp_action *in, size_t n_in)
2421 COVERAGE_INC(ofproto_ofp2odp);
2423 ctx->odp_actions = ofpbuf_new(512);
2425 ctx->may_set_up_flow = true;
2426 ctx->nf_output_iface = NF_OUT_DROP;
2428 ctx->last_pop_priority = -1;
2430 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2432 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2434 ctx->may_set_up_flow = false;
2435 } else if (ctx->check_special
2436 && ctx->ofproto->ofhooks->special_cb
2437 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2438 ctx->ofproto->aux)) {
2439 ctx->may_set_up_flow = false;
2441 do_xlate_actions(in, n_in, ctx);
2444 remove_pop_action(ctx);
2446 /* Check with in-band control to see if we're allowed to set up this
2448 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2449 ctx->odp_actions->data,
2450 ctx->odp_actions->size)) {
2451 ctx->may_set_up_flow = false;
2454 return ctx->odp_actions;
2457 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2458 * error message code (composed with ofp_mkerr()) for the caller to propagate
2459 * upward. Otherwise, returns 0.
2461 * The log message mentions 'msg_type'. */
2463 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2465 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2466 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2467 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2468 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2471 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2478 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2480 struct ofproto *p = ofconn_get_ofproto(ofconn);
2481 struct ofp_packet_out *opo;
2482 struct ofpbuf payload, *buffer;
2483 union ofp_action *ofp_actions;
2484 struct action_xlate_ctx ctx;
2485 struct ofpbuf *odp_actions;
2486 struct ofpbuf request;
2488 size_t n_ofp_actions;
2492 COVERAGE_INC(ofproto_packet_out);
2494 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2499 /* Get ofp_packet_out. */
2500 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2501 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2504 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2505 &ofp_actions, &n_ofp_actions);
2511 if (opo->buffer_id != htonl(UINT32_MAX)) {
2512 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2514 if (error || !buffer) {
2523 /* Extract flow, check actions. */
2524 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2526 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2532 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2533 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2534 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2535 ofpbuf_delete(odp_actions);
2538 ofpbuf_delete(buffer);
2543 update_port_config(struct ofproto *p, struct ofport *port,
2544 uint32_t config, uint32_t mask)
2546 mask &= config ^ port->opp.config;
2547 if (mask & OFPPC_PORT_DOWN) {
2548 if (config & OFPPC_PORT_DOWN) {
2549 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2551 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2554 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2555 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2556 if (mask & REVALIDATE_BITS) {
2557 COVERAGE_INC(ofproto_costly_flags);
2558 port->opp.config ^= mask & REVALIDATE_BITS;
2559 p->need_revalidate = true;
2561 #undef REVALIDATE_BITS
2562 if (mask & OFPPC_NO_PACKET_IN) {
2563 port->opp.config ^= OFPPC_NO_PACKET_IN;
2568 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2570 struct ofproto *p = ofconn_get_ofproto(ofconn);
2571 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2572 struct ofport *port;
2575 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2580 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2582 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2583 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2584 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2586 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2587 if (opm->advertise) {
2588 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2594 static struct ofpbuf *
2595 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2597 struct ofp_stats_reply *osr;
2600 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2601 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2603 osr->flags = htons(0);
2607 static struct ofpbuf *
2608 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2610 const struct ofp_stats_request *osr
2611 = (const struct ofp_stats_request *) request;
2612 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2616 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2617 struct ofpbuf **msgp)
2619 struct ofpbuf *msg = *msgp;
2620 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2621 if (nbytes + msg->size > UINT16_MAX) {
2622 struct ofp_stats_reply *reply = msg->data;
2623 reply->flags = htons(OFPSF_REPLY_MORE);
2624 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2625 ofconn_send_reply(ofconn, msg);
2627 return ofpbuf_put_uninit(*msgp, nbytes);
2630 static struct ofpbuf *
2631 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2633 struct nicira_stats_msg *nsm;
2636 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2637 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2638 nsm->type = htons(OFPST_VENDOR);
2639 nsm->flags = htons(0);
2640 nsm->vendor = htonl(NX_VENDOR_ID);
2641 nsm->subtype = subtype;
2645 static struct ofpbuf *
2646 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2648 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2652 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2653 struct ofpbuf **msgp)
2655 struct ofpbuf *msg = *msgp;
2656 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2657 if (nbytes + msg->size > UINT16_MAX) {
2658 struct nicira_stats_msg *reply = msg->data;
2659 reply->flags = htons(OFPSF_REPLY_MORE);
2660 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2661 ofconn_send_reply(ofconn, msg);
2663 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2667 handle_desc_stats_request(struct ofconn *ofconn,
2668 const struct ofp_header *request)
2670 struct ofproto *p = ofconn_get_ofproto(ofconn);
2671 struct ofp_desc_stats *ods;
2674 msg = start_ofp_stats_reply(request, sizeof *ods);
2675 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2676 memset(ods, 0, sizeof *ods);
2677 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2678 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2679 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2680 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2681 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2682 ofconn_send_reply(ofconn, msg);
2688 handle_table_stats_request(struct ofconn *ofconn,
2689 const struct ofp_header *request)
2691 struct ofproto *p = ofconn_get_ofproto(ofconn);
2692 struct ofp_table_stats *ots;
2695 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2697 /* Classifier table. */
2698 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2699 memset(ots, 0, sizeof *ots);
2700 strcpy(ots->name, "classifier");
2701 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2702 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2703 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2704 ots->active_count = htonl(classifier_count(&p->cls));
2705 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2706 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2708 ofconn_send_reply(ofconn, msg);
2713 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2714 struct ofpbuf **msgp)
2716 struct netdev_stats stats;
2717 struct ofp_port_stats *ops;
2719 /* Intentionally ignore return value, since errors will set
2720 * 'stats' to all-1s, which is correct for OpenFlow, and
2721 * netdev_get_stats() will log errors. */
2722 netdev_get_stats(port->netdev, &stats);
2724 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2725 ops->port_no = htons(port->opp.port_no);
2726 memset(ops->pad, 0, sizeof ops->pad);
2727 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2728 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2729 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2730 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2731 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2732 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2733 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2734 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2735 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2736 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2737 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2738 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2742 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2744 struct ofproto *p = ofconn_get_ofproto(ofconn);
2745 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2746 struct ofp_port_stats *ops;
2748 struct ofport *port;
2750 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2751 if (psr->port_no != htons(OFPP_NONE)) {
2752 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2754 append_port_stat(port, ofconn, &msg);
2757 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2758 append_port_stat(port, ofconn, &msg);
2762 ofconn_send_reply(ofconn, msg);
2767 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2769 long long int msecs = time_msec() - start;
2770 *sec = msecs / 1000;
2771 *nsec = (msecs % 1000) * (1000 * 1000);
2775 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2779 calc_flow_duration__(start, &sec, &nsec);
2780 *sec_be = htonl(sec);
2781 *nsec_be = htonl(nsec);
2785 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2786 ovs_be16 out_port, struct ofpbuf **replyp)
2788 struct ofp_flow_stats *ofs;
2789 uint64_t packet_count, byte_count;
2791 size_t act_len, len;
2793 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2797 act_len = sizeof *rule->actions * rule->n_actions;
2798 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2800 rule_get_stats(rule, &packet_count, &byte_count);
2802 ofs = append_ofp_stats_reply(len, ofconn, replyp);
2803 ofs->length = htons(len);
2806 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
2807 &ofs->match, rule->flow_cookie, &cookie);
2808 put_32aligned_be64(&ofs->cookie, cookie);
2809 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
2810 ofs->priority = htons(rule->cr.priority);
2811 ofs->idle_timeout = htons(rule->idle_timeout);
2812 ofs->hard_timeout = htons(rule->hard_timeout);
2813 memset(ofs->pad2, 0, sizeof ofs->pad2);
2814 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
2815 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
2816 if (rule->n_actions > 0) {
2817 memcpy(ofs->actions, rule->actions, act_len);
2822 is_valid_table(uint8_t table_id)
2824 if (table_id == 0 || table_id == 0xff) {
2827 /* It would probably be better to reply with an error but there doesn't
2828 * seem to be any appropriate value, so that might just be
2830 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
2837 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2839 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
2840 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2841 struct ofpbuf *reply;
2843 COVERAGE_INC(ofproto_flows_req);
2844 reply = start_ofp_stats_reply(oh, 1024);
2845 if (is_valid_table(fsr->table_id)) {
2846 struct cls_cursor cursor;
2847 struct cls_rule target;
2850 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
2852 cls_cursor_init(&cursor, &ofproto->cls, &target);
2853 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2854 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
2857 ofconn_send_reply(ofconn, reply);
2863 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
2864 ovs_be16 out_port, struct ofpbuf **replyp)
2866 struct nx_flow_stats *nfs;
2867 uint64_t packet_count, byte_count;
2868 size_t act_len, start_len;
2869 struct ofpbuf *reply;
2871 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2875 rule_get_stats(rule, &packet_count, &byte_count);
2877 act_len = sizeof *rule->actions * rule->n_actions;
2879 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
2880 start_len = (*replyp)->size;
2883 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
2886 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
2887 nfs->cookie = rule->flow_cookie;
2888 nfs->priority = htons(rule->cr.priority);
2889 nfs->idle_timeout = htons(rule->idle_timeout);
2890 nfs->hard_timeout = htons(rule->hard_timeout);
2891 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
2892 memset(nfs->pad2, 0, sizeof nfs->pad2);
2893 nfs->packet_count = htonll(packet_count);
2894 nfs->byte_count = htonll(byte_count);
2895 if (rule->n_actions > 0) {
2896 ofpbuf_put(reply, rule->actions, act_len);
2898 nfs->length = htons(reply->size - start_len);
2902 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
2904 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2905 struct nx_flow_stats_request *nfsr;
2906 struct cls_rule target;
2907 struct ofpbuf *reply;
2911 ofpbuf_use_const(&b, oh, ntohs(oh->length));
2913 /* Dissect the message. */
2914 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
2915 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
2920 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2923 COVERAGE_INC(ofproto_flows_req);
2924 reply = start_nxstats_reply(&nfsr->nsm, 1024);
2925 if (is_valid_table(nfsr->table_id)) {
2926 struct cls_cursor cursor;
2929 cls_cursor_init(&cursor, &ofproto->cls, &target);
2930 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2931 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
2934 ofconn_send_reply(ofconn, reply);
2940 flow_stats_ds(struct rule *rule, struct ds *results)
2942 uint64_t packet_count, byte_count;
2943 size_t act_len = sizeof *rule->actions * rule->n_actions;
2945 rule_get_stats(rule, &packet_count, &byte_count);
2947 ds_put_format(results, "duration=%llds, ",
2948 (time_msec() - rule->created) / 1000);
2949 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
2950 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2951 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2952 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2953 cls_rule_format(&rule->cr, results);
2954 ds_put_char(results, ',');
2956 ofp_print_actions(results, &rule->actions->header, act_len);
2958 ds_put_cstr(results, "drop");
2960 ds_put_cstr(results, "\n");
2963 /* Adds a pretty-printed description of all flows to 'results', including
2964 * hidden flows (e.g., set up by in-band control). */
2966 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2968 struct cls_cursor cursor;
2971 cls_cursor_init(&cursor, &p->cls, NULL);
2972 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2973 flow_stats_ds(rule, results);
2978 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
2979 ovs_be16 out_port, uint8_t table_id,
2980 struct ofp_aggregate_stats_reply *oasr)
2982 uint64_t total_packets = 0;
2983 uint64_t total_bytes = 0;
2986 COVERAGE_INC(ofproto_agg_request);
2988 if (is_valid_table(table_id)) {
2989 struct cls_cursor cursor;
2992 cls_cursor_init(&cursor, &ofproto->cls, target);
2993 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2994 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
2995 uint64_t packet_count;
2996 uint64_t byte_count;
2998 rule_get_stats(rule, &packet_count, &byte_count);
3000 total_packets += packet_count;
3001 total_bytes += byte_count;
3007 oasr->flow_count = htonl(n_flows);
3008 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3009 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3010 memset(oasr->pad, 0, sizeof oasr->pad);
3014 handle_aggregate_stats_request(struct ofconn *ofconn,
3015 const struct ofp_header *oh)
3017 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3018 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3019 struct ofp_aggregate_stats_reply *reply;
3020 struct cls_rule target;
3023 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3026 msg = start_ofp_stats_reply(oh, sizeof *reply);
3027 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3028 query_aggregate_stats(ofproto, &target, request->out_port,
3029 request->table_id, reply);
3030 ofconn_send_reply(ofconn, msg);
3035 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3037 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3038 struct nx_aggregate_stats_request *request;
3039 struct ofp_aggregate_stats_reply *reply;
3040 struct cls_rule target;
3045 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3047 /* Dissect the message. */
3048 request = ofpbuf_pull(&b, sizeof *request);
3049 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3054 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3058 COVERAGE_INC(ofproto_flows_req);
3059 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3060 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3061 query_aggregate_stats(ofproto, &target, request->out_port,
3062 request->table_id, reply);
3063 ofconn_send_reply(ofconn, buf);
3068 struct queue_stats_cbdata {
3069 struct ofconn *ofconn;
3070 struct ofport *ofport;
3075 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3076 const struct netdev_queue_stats *stats)
3078 struct ofp_queue_stats *reply;
3080 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3081 reply->port_no = htons(cbdata->ofport->opp.port_no);
3082 memset(reply->pad, 0, sizeof reply->pad);
3083 reply->queue_id = htonl(queue_id);
3084 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3085 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3086 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3090 handle_queue_stats_dump_cb(uint32_t queue_id,
3091 struct netdev_queue_stats *stats,
3094 struct queue_stats_cbdata *cbdata = cbdata_;
3096 put_queue_stats(cbdata, queue_id, stats);
3100 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3101 struct queue_stats_cbdata *cbdata)
3103 cbdata->ofport = port;
3104 if (queue_id == OFPQ_ALL) {
3105 netdev_dump_queue_stats(port->netdev,
3106 handle_queue_stats_dump_cb, cbdata);
3108 struct netdev_queue_stats stats;
3110 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3111 put_queue_stats(cbdata, queue_id, &stats);
3117 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3119 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3120 const struct ofp_queue_stats_request *qsr;
3121 struct queue_stats_cbdata cbdata;
3122 struct ofport *port;
3123 unsigned int port_no;
3126 qsr = ofputil_stats_body(oh);
3128 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3131 COVERAGE_INC(ofproto_queue_req);
3133 cbdata.ofconn = ofconn;
3134 cbdata.msg = start_ofp_stats_reply(oh, 128);
3136 port_no = ntohs(qsr->port_no);
3137 queue_id = ntohl(qsr->queue_id);
3138 if (port_no == OFPP_ALL) {
3139 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3140 handle_queue_stats_for_port(port, queue_id, &cbdata);
3142 } else if (port_no < ofproto->max_ports) {
3143 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3145 handle_queue_stats_for_port(port, queue_id, &cbdata);
3148 ofpbuf_delete(cbdata.msg);
3149 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3151 ofconn_send_reply(ofconn, cbdata.msg);
3156 /* Updates 'facet''s used time. Caller is responsible for calling
3157 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3159 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3162 if (used > facet->used) {
3164 if (used > facet->rule->used) {
3165 facet->rule->used = used;
3167 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3171 /* Folds the statistics from 'stats' into the counters in 'facet'.
3173 * Because of the meaning of a facet's counters, it only makes sense to do this
3174 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3175 * packet that was sent by hand or if it represents statistics that have been
3176 * cleared out of the datapath. */
3178 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3179 const struct dpif_flow_stats *stats)
3181 if (stats->n_packets || stats->used > facet->used) {
3182 facet_update_time(ofproto, facet, stats->used);
3183 facet->packet_count += stats->n_packets;
3184 facet->byte_count += stats->n_bytes;
3185 facet_push_stats(ofproto, facet);
3186 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3191 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3193 uint64_t rs_packets, rs_bytes;
3195 assert(facet->packet_count >= facet->rs_packet_count);
3196 assert(facet->byte_count >= facet->rs_byte_count);
3197 assert(facet->used >= facet->rs_used);
3199 rs_packets = facet->packet_count - facet->rs_packet_count;
3200 rs_bytes = facet->byte_count - facet->rs_byte_count;
3202 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3203 facet->rs_packet_count = facet->packet_count;
3204 facet->rs_byte_count = facet->byte_count;
3205 facet->rs_used = facet->used;
3207 flow_push_stats(ofproto, facet->rule, &facet->flow,
3208 rs_packets, rs_bytes, facet->used);
3212 struct ofproto_push {
3213 struct action_xlate_ctx ctx;
3220 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3222 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3225 rule->packet_count += push->packets;
3226 rule->byte_count += push->bytes;
3227 rule->used = MAX(push->used, rule->used);
3231 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3232 * 'rule''s actions. */
3234 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3235 struct flow *flow, uint64_t packets, uint64_t bytes,
3238 struct ofproto_push push;
3240 push.packets = packets;
3244 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3245 push.ctx.resubmit_hook = push_resubmit;
3246 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3249 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3250 * in which no matching flow already exists in the flow table.
3252 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3253 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3254 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3256 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3259 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3261 struct ofproto *p = ofconn_get_ofproto(ofconn);
3262 struct ofpbuf *packet;
3267 if (fm->flags & OFPFF_CHECK_OVERLAP
3268 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3269 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3273 if (fm->buffer_id != UINT32_MAX) {
3274 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3278 in_port = UINT16_MAX;
3281 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3282 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3283 fm->flags & OFPFF_SEND_FLOW_REM);
3284 rule_insert(p, rule);
3286 rule_execute(p, rule, in_port, packet);
3291 static struct rule *
3292 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3294 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3298 send_buffered_packet(struct ofconn *ofconn,
3299 struct rule *rule, uint32_t buffer_id)
3301 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3302 struct ofpbuf *packet;
3306 if (buffer_id == UINT32_MAX) {
3310 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3315 rule_execute(ofproto, rule, in_port, packet);
3320 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3322 struct modify_flows_cbdata {
3323 struct ofproto *ofproto;
3324 const struct flow_mod *fm;
3328 static int modify_flow(struct ofproto *, const struct flow_mod *,
3331 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3332 * encoded by ofp_mkerr() on failure.
3334 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3337 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3339 struct ofproto *p = ofconn_get_ofproto(ofconn);
3340 struct rule *match = NULL;
3341 struct cls_cursor cursor;
3344 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3345 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3346 if (!rule_is_hidden(rule)) {
3348 modify_flow(p, fm, rule);
3353 /* This credits the packet to whichever flow happened to match last.
3354 * That's weird. Maybe we should do a lookup for the flow that
3355 * actually matches the packet? Who knows. */
3356 send_buffered_packet(ofconn, match, fm->buffer_id);
3359 return add_flow(ofconn, fm);
3363 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3364 * code as encoded by ofp_mkerr() on failure.
3366 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3369 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3371 struct ofproto *p = ofconn_get_ofproto(ofconn);
3372 struct rule *rule = find_flow_strict(p, fm);
3373 if (rule && !rule_is_hidden(rule)) {
3374 modify_flow(p, fm, rule);
3375 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3377 return add_flow(ofconn, fm);
3381 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3382 * been identified as a flow in 'p''s flow table to be modified, by changing
3383 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3384 * ofp_action[] structures). */
3386 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3388 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3390 rule->flow_cookie = fm->cookie;
3392 /* If the actions are the same, do nothing. */
3393 if (fm->n_actions == rule->n_actions
3395 || !memcmp(fm->actions, rule->actions, actions_len))) {
3399 /* Replace actions. */
3400 free(rule->actions);
3401 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3402 rule->n_actions = fm->n_actions;
3404 p->need_revalidate = true;
3409 /* OFPFC_DELETE implementation. */
3411 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3413 /* Implements OFPFC_DELETE. */
3415 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3417 struct rule *rule, *next_rule;
3418 struct cls_cursor cursor;
3420 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3421 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3422 delete_flow(p, rule, htons(fm->out_port));
3426 /* Implements OFPFC_DELETE_STRICT. */
3428 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3430 struct rule *rule = find_flow_strict(p, fm);
3432 delete_flow(p, rule, htons(fm->out_port));
3436 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3437 * been identified as a flow to delete from 'p''s flow table, by deleting the
3438 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3441 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3442 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3443 * specified 'out_port'. */
3445 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3447 if (rule_is_hidden(rule)) {
3451 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3455 rule_send_removed(p, rule, OFPRR_DELETE);
3456 rule_remove(p, rule);
3460 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3462 struct ofproto *p = ofconn_get_ofproto(ofconn);
3466 error = reject_slave_controller(ofconn, "flow_mod");
3471 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3476 /* We do not support the emergency flow cache. It will hopefully get
3477 * dropped from OpenFlow in the near future. */
3478 if (fm.flags & OFPFF_EMERG) {
3479 /* There isn't a good fit for an error code, so just state that the
3480 * flow table is full. */
3481 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3484 error = validate_actions(fm.actions, fm.n_actions,
3485 &fm.cr.flow, p->max_ports);
3490 switch (fm.command) {
3492 return add_flow(ofconn, &fm);
3495 return modify_flows_loose(ofconn, &fm);
3497 case OFPFC_MODIFY_STRICT:
3498 return modify_flow_strict(ofconn, &fm);
3501 delete_flows_loose(p, &fm);
3504 case OFPFC_DELETE_STRICT:
3505 delete_flow_strict(p, &fm);
3509 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3514 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3516 const struct nxt_tun_id_cookie *msg
3517 = (const struct nxt_tun_id_cookie *) oh;
3518 enum nx_flow_format flow_format;
3520 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3521 ofconn_set_flow_format(ofconn, flow_format);
3527 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3529 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3530 struct nx_role_request *reply;
3534 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3535 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3536 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3539 role = ntohl(nrr->role);
3540 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3541 && role != NX_ROLE_SLAVE) {
3542 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3544 /* There's no good error code for this. */
3545 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3548 ofconn_set_role(ofconn, role);
3550 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3551 reply->role = htonl(role);
3552 ofconn_send_reply(ofconn, buf);
3558 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3560 const struct nxt_set_flow_format *msg
3561 = (const struct nxt_set_flow_format *) oh;
3564 format = ntohl(msg->format);
3565 if (format == NXFF_OPENFLOW10
3566 || format == NXFF_TUN_ID_FROM_COOKIE
3567 || format == NXFF_NXM) {
3568 ofconn_set_flow_format(ofconn, format);
3571 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3576 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3578 struct ofp_header *ob;
3581 /* Currently, everything executes synchronously, so we can just
3582 * immediately send the barrier reply. */
3583 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3584 ofconn_send_reply(ofconn, buf);
3589 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3591 const struct ofp_header *oh = msg->data;
3592 const struct ofputil_msg_type *type;
3595 error = ofputil_decode_msg_type(oh, &type);
3600 switch (ofputil_msg_type_code(type)) {
3601 /* OpenFlow requests. */
3602 case OFPUTIL_OFPT_ECHO_REQUEST:
3603 return handle_echo_request(ofconn, oh);
3605 case OFPUTIL_OFPT_FEATURES_REQUEST:
3606 return handle_features_request(ofconn, oh);
3608 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3609 return handle_get_config_request(ofconn, oh);
3611 case OFPUTIL_OFPT_SET_CONFIG:
3612 return handle_set_config(ofconn, msg->data);
3614 case OFPUTIL_OFPT_PACKET_OUT:
3615 return handle_packet_out(ofconn, oh);
3617 case OFPUTIL_OFPT_PORT_MOD:
3618 return handle_port_mod(ofconn, oh);
3620 case OFPUTIL_OFPT_FLOW_MOD:
3621 return handle_flow_mod(ofconn, oh);
3623 case OFPUTIL_OFPT_BARRIER_REQUEST:
3624 return handle_barrier_request(ofconn, oh);
3626 /* OpenFlow replies. */
3627 case OFPUTIL_OFPT_ECHO_REPLY:
3630 /* Nicira extension requests. */
3631 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3632 return handle_tun_id_from_cookie(ofconn, oh);
3634 case OFPUTIL_NXT_ROLE_REQUEST:
3635 return handle_role_request(ofconn, oh);
3637 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3638 return handle_nxt_set_flow_format(ofconn, oh);
3640 case OFPUTIL_NXT_FLOW_MOD:
3641 return handle_flow_mod(ofconn, oh);
3643 /* OpenFlow statistics requests. */
3644 case OFPUTIL_OFPST_DESC_REQUEST:
3645 return handle_desc_stats_request(ofconn, oh);
3647 case OFPUTIL_OFPST_FLOW_REQUEST:
3648 return handle_flow_stats_request(ofconn, oh);
3650 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3651 return handle_aggregate_stats_request(ofconn, oh);
3653 case OFPUTIL_OFPST_TABLE_REQUEST:
3654 return handle_table_stats_request(ofconn, oh);
3656 case OFPUTIL_OFPST_PORT_REQUEST:
3657 return handle_port_stats_request(ofconn, oh);
3659 case OFPUTIL_OFPST_QUEUE_REQUEST:
3660 return handle_queue_stats_request(ofconn, oh);
3662 /* Nicira extension statistics requests. */
3663 case OFPUTIL_NXST_FLOW_REQUEST:
3664 return handle_nxst_flow(ofconn, oh);
3666 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3667 return handle_nxst_aggregate(ofconn, oh);
3669 case OFPUTIL_INVALID:
3670 case OFPUTIL_OFPT_HELLO:
3671 case OFPUTIL_OFPT_ERROR:
3672 case OFPUTIL_OFPT_FEATURES_REPLY:
3673 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3674 case OFPUTIL_OFPT_PACKET_IN:
3675 case OFPUTIL_OFPT_FLOW_REMOVED:
3676 case OFPUTIL_OFPT_PORT_STATUS:
3677 case OFPUTIL_OFPT_BARRIER_REPLY:
3678 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3679 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3680 case OFPUTIL_OFPST_DESC_REPLY:
3681 case OFPUTIL_OFPST_FLOW_REPLY:
3682 case OFPUTIL_OFPST_QUEUE_REPLY:
3683 case OFPUTIL_OFPST_PORT_REPLY:
3684 case OFPUTIL_OFPST_TABLE_REPLY:
3685 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3686 case OFPUTIL_NXT_ROLE_REPLY:
3687 case OFPUTIL_NXT_FLOW_REMOVED:
3688 case OFPUTIL_NXST_FLOW_REPLY:
3689 case OFPUTIL_NXST_AGGREGATE_REPLY:
3691 if (VLOG_IS_WARN_ENABLED()) {
3692 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3693 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3696 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3697 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3699 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3705 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3707 int error = handle_openflow__(ofconn, ofp_msg);
3709 send_error_oh(ofconn, ofp_msg->data, error);
3711 COVERAGE_INC(ofproto_recv_openflow);
3715 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3717 struct facet *facet;
3720 /* Obtain in_port and tun_id, at least. */
3721 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3723 /* Set header pointers in 'flow'. */
3724 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3726 if (cfm_should_process_flow(&flow)) {
3727 ofproto_process_cfm(p, &flow, upcall->packet);
3728 ofpbuf_delete(upcall->packet);
3730 } else if (p->ofhooks->special_cb
3731 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3732 ofpbuf_delete(upcall->packet);
3736 /* Check with in-band control to see if this packet should be sent
3737 * to the local port regardless of the flow table. */
3738 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3739 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3742 facet = facet_lookup_valid(p, &flow);
3744 struct rule *rule = rule_lookup(p, &flow);
3746 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3747 struct ofport *port = get_port(p, flow.in_port);
3749 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3750 COVERAGE_INC(ofproto_no_packet_in);
3751 /* XXX install 'drop' flow entry */
3752 ofpbuf_delete(upcall->packet);
3756 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3760 COVERAGE_INC(ofproto_packet_in);
3761 send_packet_in(p, upcall, &flow, false);
3765 facet = facet_create(p, rule, &flow, upcall->packet);
3766 } else if (!facet->may_install) {
3767 /* The facet is not installable, that is, we need to process every
3768 * packet, so process the current packet's actions into 'facet'. */
3769 facet_make_actions(p, facet, upcall->packet);
3772 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3774 * Extra-special case for fail-open mode.
3776 * We are in fail-open mode and the packet matched the fail-open rule,
3777 * but we are connected to a controller too. We should send the packet
3778 * up to the controller in the hope that it will try to set up a flow
3779 * and thereby allow us to exit fail-open.
3781 * See the top-level comment in fail-open.c for more information.
3783 send_packet_in(p, upcall, &flow, true);
3786 facet_execute(p, facet, upcall->packet);
3787 facet_install(p, facet, false);
3791 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3795 switch (upcall->type) {
3796 case DPIF_UC_ACTION:
3797 COVERAGE_INC(ofproto_ctlr_action);
3798 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3799 send_packet_in(p, upcall, &flow, false);
3802 case DPIF_UC_SAMPLE:
3804 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3805 ofproto_sflow_received(p->sflow, upcall, &flow);
3807 ofpbuf_delete(upcall->packet);
3811 handle_miss_upcall(p, upcall);
3814 case DPIF_N_UC_TYPES:
3816 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3821 /* Flow expiration. */
3823 static int ofproto_dp_max_idle(const struct ofproto *);
3824 static void ofproto_update_stats(struct ofproto *);
3825 static void rule_expire(struct ofproto *, struct rule *);
3826 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
3828 /* This function is called periodically by ofproto_run(). Its job is to
3829 * collect updates for the flows that have been installed into the datapath,
3830 * most importantly when they last were used, and then use that information to
3831 * expire flows that have not been used recently.
3833 * Returns the number of milliseconds after which it should be called again. */
3835 ofproto_expire(struct ofproto *ofproto)
3837 struct rule *rule, *next_rule;
3838 struct cls_cursor cursor;
3841 /* Update stats for each flow in the datapath. */
3842 ofproto_update_stats(ofproto);
3844 /* Expire facets that have been idle too long. */
3845 dp_max_idle = ofproto_dp_max_idle(ofproto);
3846 ofproto_expire_facets(ofproto, dp_max_idle);
3848 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3849 cls_cursor_init(&cursor, &ofproto->cls, NULL);
3850 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3851 rule_expire(ofproto, rule);
3854 /* Let the hook know that we're at a stable point: all outstanding data
3855 * in existing flows has been accounted to the account_cb. Thus, the
3856 * hook can now reasonably do operations that depend on having accurate
3857 * flow volume accounting (currently, that's just bond rebalancing). */
3858 if (ofproto->ofhooks->account_checkpoint_cb) {
3859 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
3862 return MIN(dp_max_idle, 1000);
3865 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3867 * This function also pushes statistics updates to rules which each facet
3868 * resubmits into. Generally these statistics will be accurate. However, if a
3869 * facet changes the rule it resubmits into at some time in between
3870 * ofproto_update_stats() runs, it is possible that statistics accrued to the
3871 * old rule will be incorrectly attributed to the new rule. This could be
3872 * avoided by calling ofproto_update_stats() whenever rules are created or
3873 * deleted. However, the performance impact of making so many calls to the
3874 * datapath do not justify the benefit of having perfectly accurate statistics.
3877 ofproto_update_stats(struct ofproto *p)
3879 const struct dpif_flow_stats *stats;
3880 struct dpif_flow_dump dump;
3881 const struct nlattr *key;
3884 dpif_flow_dump_start(&dump, p->dpif);
3885 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3886 struct facet *facet;
3889 if (odp_flow_key_to_flow(key, key_len, &flow)) {
3893 odp_flow_key_format(key, key_len, &s);
3894 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
3900 facet = facet_find(p, &flow);
3902 if (facet && facet->installed) {
3904 if (stats->n_packets >= facet->dp_packet_count) {
3905 facet->packet_count += stats->n_packets - facet->dp_packet_count;
3907 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3910 if (stats->n_bytes >= facet->dp_byte_count) {
3911 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
3913 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3916 facet->dp_packet_count = stats->n_packets;
3917 facet->dp_byte_count = stats->n_bytes;
3919 facet_update_time(p, facet, stats->used);
3920 facet_account(p, facet, stats->n_bytes);
3921 facet_push_stats(p, facet);
3923 /* There's a flow in the datapath that we know nothing about.
3925 COVERAGE_INC(ofproto_unexpected_rule);
3926 dpif_flow_del(p->dpif, key, key_len, NULL);
3929 dpif_flow_dump_done(&dump);
3932 /* Calculates and returns the number of milliseconds of idle time after which
3933 * facets should expire from the datapath and we should fold their statistics
3934 * into their parent rules in userspace. */
3936 ofproto_dp_max_idle(const struct ofproto *ofproto)
3939 * Idle time histogram.
3941 * Most of the time a switch has a relatively small number of facets. When
3942 * this is the case we might as well keep statistics for all of them in
3943 * userspace and to cache them in the kernel datapath for performance as
3946 * As the number of facets increases, the memory required to maintain
3947 * statistics about them in userspace and in the kernel becomes
3948 * significant. However, with a large number of facets it is likely that
3949 * only a few of them are "heavy hitters" that consume a large amount of
3950 * bandwidth. At this point, only heavy hitters are worth caching in the
3951 * kernel and maintaining in userspaces; other facets we can discard.
3953 * The technique used to compute the idle time is to build a histogram with
3954 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
3955 * that is installed in the kernel gets dropped in the appropriate bucket.
3956 * After the histogram has been built, we compute the cutoff so that only
3957 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
3958 * cached. At least the most-recently-used bucket of facets is kept, so
3959 * actually an arbitrary number of facets can be kept in any given
3960 * expiration run (though the next run will delete most of those unless
3961 * they receive additional data).
3963 * This requires a second pass through the facets, in addition to the pass
3964 * made by ofproto_update_stats(), because the former function never looks
3965 * at uninstallable facets.
3967 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3968 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3969 int buckets[N_BUCKETS] = { 0 };
3970 struct facet *facet;
3975 total = hmap_count(&ofproto->facets);
3976 if (total <= 1000) {
3977 return N_BUCKETS * BUCKET_WIDTH;
3980 /* Build histogram. */
3982 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
3983 long long int idle = now - facet->used;
3984 int bucket = (idle <= 0 ? 0
3985 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3986 : (unsigned int) idle / BUCKET_WIDTH);
3990 /* Find the first bucket whose flows should be expired. */
3991 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
3992 if (buckets[bucket]) {
3995 subtotal += buckets[bucket++];
3996 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4001 if (VLOG_IS_DBG_ENABLED()) {
4005 ds_put_cstr(&s, "keep");
4006 for (i = 0; i < N_BUCKETS; i++) {
4008 ds_put_cstr(&s, ", drop");
4011 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4014 VLOG_INFO("%s: %s (msec:count)",
4015 dpif_name(ofproto->dpif), ds_cstr(&s));
4019 return bucket * BUCKET_WIDTH;
4023 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4025 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4026 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4027 struct ofexpired expired;
4029 if (facet->installed) {
4030 struct dpif_flow_stats stats;
4032 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4034 facet_update_stats(ofproto, facet, &stats);
4037 expired.flow = facet->flow;
4038 expired.packet_count = facet->packet_count;
4039 expired.byte_count = facet->byte_count;
4040 expired.used = facet->used;
4041 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4046 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4048 long long int cutoff = time_msec() - dp_max_idle;
4049 struct facet *facet, *next_facet;
4051 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4052 facet_active_timeout(ofproto, facet);
4053 if (facet->used < cutoff) {
4054 facet_remove(ofproto, facet);
4059 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4060 * then delete it entirely. */
4062 rule_expire(struct ofproto *ofproto, struct rule *rule)
4064 struct facet *facet, *next_facet;
4068 /* Has 'rule' expired? */
4070 if (rule->hard_timeout
4071 && now > rule->created + rule->hard_timeout * 1000) {
4072 reason = OFPRR_HARD_TIMEOUT;
4073 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4074 && now >rule->used + rule->idle_timeout * 1000) {
4075 reason = OFPRR_IDLE_TIMEOUT;
4080 COVERAGE_INC(ofproto_expired);
4082 /* Update stats. (This is a no-op if the rule expired due to an idle
4083 * timeout, because that only happens when the rule has no facets left.) */
4084 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4085 facet_remove(ofproto, facet);
4088 /* Get rid of the rule. */
4089 if (!rule_is_hidden(rule)) {
4090 rule_send_removed(ofproto, rule, reason);
4092 rule_remove(ofproto, rule);
4096 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4098 struct ofputil_flow_removed fr;
4100 if (!rule->send_flow_removed) {
4105 fr.cookie = rule->flow_cookie;
4107 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4108 fr.idle_timeout = rule->idle_timeout;
4109 fr.packet_count = rule->packet_count;
4110 fr.byte_count = rule->byte_count;
4112 connmgr_send_flow_removed(p->connmgr, &fr);
4115 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4116 * The returned statistics include statistics for all of 'rule''s facets. */
4118 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4121 struct facet *facet;
4123 /* Start from historical data for 'rule' itself that are no longer tracked
4124 * in facets. This counts, for example, facets that have expired. */
4125 p = rule->packet_count;
4126 b = rule->byte_count;
4128 /* Add any statistics that are tracked by facets. This includes
4129 * statistical data recently updated by ofproto_update_stats() as well as
4130 * stats for packets that were executed "by hand" via dpif_execute(). */
4131 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4132 p += facet->packet_count;
4133 b += facet->byte_count;
4140 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4141 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4142 * their individual configurations.
4144 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4145 * Otherwise, ownership is transferred to this function. */
4147 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4148 const struct flow *flow, bool clone)
4150 struct ofputil_packet_in pin;
4152 pin.packet = upcall->packet;
4153 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4154 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4155 pin.buffer_id = 0; /* not yet known */
4156 pin.send_len = upcall->userdata;
4157 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4158 clone ? NULL : upcall->packet);
4162 pick_datapath_id(const struct ofproto *ofproto)
4164 const struct ofport *port;
4166 port = get_port(ofproto, ODPP_LOCAL);
4168 uint8_t ea[ETH_ADDR_LEN];
4171 error = netdev_get_etheraddr(port->netdev, ea);
4173 return eth_addr_to_uint64(ea);
4175 VLOG_WARN("could not get MAC address for %s (%s)",
4176 netdev_get_name(port->netdev), strerror(error));
4178 return ofproto->fallback_dpid;
4182 pick_fallback_dpid(void)
4184 uint8_t ea[ETH_ADDR_LEN];
4185 eth_addr_nicira_random(ea);
4186 return eth_addr_to_uint64(ea);
4190 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4191 void *aux OVS_UNUSED)
4193 const struct shash_node *node;
4197 SHASH_FOR_EACH (node, &all_ofprotos) {
4198 ds_put_format(&results, "%s\n", node->name);
4200 unixctl_command_reply(conn, 200, ds_cstr(&results));
4201 ds_destroy(&results);
4204 struct ofproto_trace {
4205 struct action_xlate_ctx ctx;
4211 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4213 ds_put_char_multiple(result, '\t', level);
4215 ds_put_cstr(result, "No match\n");
4219 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4220 ntohll(rule->flow_cookie));
4221 cls_rule_format(&rule->cr, result);
4222 ds_put_char(result, '\n');
4224 ds_put_char_multiple(result, '\t', level);
4225 ds_put_cstr(result, "OpenFlow ");
4226 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4227 rule->n_actions * sizeof *rule->actions);
4228 ds_put_char(result, '\n');
4232 trace_format_flow(struct ds *result, int level, const char *title,
4233 struct ofproto_trace *trace)
4235 ds_put_char_multiple(result, '\t', level);
4236 ds_put_format(result, "%s: ", title);
4237 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4238 ds_put_cstr(result, "unchanged");
4240 flow_format(result, &trace->ctx.flow);
4241 trace->flow = trace->ctx.flow;
4243 ds_put_char(result, '\n');
4247 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4249 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4250 struct ds *result = trace->result;
4252 ds_put_char(result, '\n');
4253 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4254 trace_format_rule(result, ctx->recurse + 1, rule);
4258 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4259 void *aux OVS_UNUSED)
4261 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4262 char *args = xstrdup(args_);
4263 char *save_ptr = NULL;
4264 struct ofproto *ofproto;
4265 struct ofpbuf packet;
4273 ofpbuf_init(&packet, strlen(args) / 2);
4276 dpname = strtok_r(args, " ", &save_ptr);
4277 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4278 in_port_s = strtok_r(NULL, " ", &save_ptr);
4279 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4280 if (!dpname || !in_port_s || !packet_s) {
4281 unixctl_command_reply(conn, 501, "Bad command syntax");
4285 ofproto = shash_find_data(&all_ofprotos, dpname);
4287 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4292 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4293 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4295 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4296 packet_s += strspn(packet_s, " ");
4297 if (*packet_s != '\0') {
4298 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4301 if (packet.size < ETH_HEADER_LEN) {
4302 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4306 ds_put_cstr(&result, "Packet: ");
4307 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4308 ds_put_cstr(&result, s);
4311 flow_extract(&packet, tun_id, in_port, &flow);
4312 ds_put_cstr(&result, "Flow: ");
4313 flow_format(&result, &flow);
4314 ds_put_char(&result, '\n');
4316 rule = rule_lookup(ofproto, &flow);
4317 trace_format_rule(&result, 0, rule);
4319 struct ofproto_trace trace;
4320 struct ofpbuf *odp_actions;
4322 trace.result = &result;
4324 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4325 trace.ctx.resubmit_hook = trace_resubmit;
4326 odp_actions = xlate_actions(&trace.ctx,
4327 rule->actions, rule->n_actions);
4329 ds_put_char(&result, '\n');
4330 trace_format_flow(&result, 0, "Final flow", &trace);
4331 ds_put_cstr(&result, "Datapath actions: ");
4332 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4333 ofpbuf_delete(odp_actions);
4336 unixctl_command_reply(conn, 200, ds_cstr(&result));
4339 ds_destroy(&result);
4340 ofpbuf_uninit(&packet);
4345 ofproto_unixctl_init(void)
4347 static bool registered;
4353 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4354 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4358 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4359 struct ofpbuf *odp_actions, tag_type *tags,
4360 uint16_t *nf_output_iface, void *ofproto_)
4362 struct ofproto *ofproto = ofproto_;
4363 struct mac_entry *dst_mac;
4365 /* Drop frames for reserved multicast addresses. */
4366 if (eth_addr_is_reserved(flow->dl_dst)) {
4370 /* Learn source MAC (but don't try to learn from revalidation). */
4372 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4373 struct mac_entry *src_mac;
4375 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4376 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4377 /* The log messages here could actually be useful in debugging,
4378 * so keep the rate limit relatively high. */
4379 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4380 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4381 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4383 ofproto_revalidate(ofproto,
4384 mac_learning_changed(ofproto->ml, src_mac));
4385 src_mac->port.i = flow->in_port;
4389 /* Determine output port. */
4390 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4392 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4393 nf_output_iface, odp_actions);
4395 int out_port = dst_mac->port.i;
4396 if (out_port != flow->in_port) {
4397 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4398 *nf_output_iface = out_port;
4407 static const struct ofhooks default_ofhooks = {
4408 default_normal_ofhook_cb,