2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
34 #include "guarded-list.h"
38 #include "mac-learning.h"
39 #include "meta-flow.h"
40 #include "multipath.h"
41 #include "netdev-vport.h"
46 #include "odp-execute.h"
49 #include "ofp-actions.h"
50 #include "ofp-parse.h"
51 #include "ofp-print.h"
52 #include "ofproto-dpif-governor.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-sflow.h"
56 #include "ofproto-dpif-upcall.h"
57 #include "ofproto-dpif-xlate.h"
58 #include "poll-loop.h"
63 #include "unaligned.h"
65 #include "vlan-bitmap.h"
68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
70 COVERAGE_DEFINE(ofproto_dpif_expired);
71 COVERAGE_DEFINE(facet_revalidate);
72 COVERAGE_DEFINE(facet_unexpected);
73 COVERAGE_DEFINE(subfacet_install_fail);
74 COVERAGE_DEFINE(packet_in_overflow);
76 /* Number of implemented OpenFlow tables. */
77 enum { N_TABLES = 255 };
78 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
79 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
89 * - Do include packets and bytes from facets that have been deleted or
90 * whose own statistics have been folded into the rule.
92 * - Do include packets and bytes sent "by hand" that were accounted to
93 * the rule without any facet being involved (this is a rare corner
94 * case in rule_execute()).
96 * - Do not include packet or bytes that can be obtained from any facet's
97 * packet_count or byte_count member or that can be obtained from the
98 * datapath by, e.g., dpif_flow_get() for any subfacet.
100 struct ovs_mutex stats_mutex;
101 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
102 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
105 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
106 static struct rule_dpif *rule_dpif_cast(const struct rule *);
109 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
110 struct ofproto_dpif *ofproto; /* Owning ofproto. */
111 void *aux; /* Key supplied by ofproto's client. */
112 char *name; /* Identifier for log messages. */
115 struct list ports; /* Contains "struct ofport"s. */
116 enum port_vlan_mode vlan_mode; /* VLAN mode */
117 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
118 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
119 * NULL if all VLANs are trunked. */
120 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
121 struct bond *bond; /* Nonnull iff more than one port. */
122 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
125 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
128 static void bundle_remove(struct ofport *);
129 static void bundle_update(struct ofbundle *);
130 static void bundle_destroy(struct ofbundle *);
131 static void bundle_del_port(struct ofport_dpif *);
132 static void bundle_run(struct ofbundle *);
133 static void bundle_wait(struct ofbundle *);
135 static void stp_run(struct ofproto_dpif *ofproto);
136 static void stp_wait(struct ofproto_dpif *ofproto);
137 static int set_stp_port(struct ofport *,
138 const struct ofproto_port_stp_settings *);
140 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
141 enum slow_path_reason,
142 uint64_t *stub, size_t stub_size,
143 const struct nlattr **actionsp,
144 size_t *actions_lenp);
146 /* A subfacet (see "struct subfacet" below) has three possible installation
149 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
150 * case just after the subfacet is created, just before the subfacet is
151 * destroyed, or if the datapath returns an error when we try to install a
154 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
156 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
157 * ofproto_dpif is installed in the datapath.
160 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
161 SF_FAST_PATH, /* Full actions are installed. */
162 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
165 /* A dpif flow and actions associated with a facet.
167 * See also the large comment on struct facet. */
170 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
171 struct list list_node; /* In struct facet's 'facets' list. */
172 struct facet *facet; /* Owning facet. */
173 struct dpif_backer *backer; /* Owning backer. */
178 long long int used; /* Time last used; time created if not used. */
179 long long int created; /* Time created. */
181 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
182 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
184 enum subfacet_path path; /* Installed in datapath? */
187 #define SUBFACET_DESTROY_MAX_BATCH 50
189 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
190 static struct subfacet *subfacet_find(struct dpif_backer *,
191 const struct nlattr *key, size_t key_len,
193 static void subfacet_destroy(struct subfacet *);
194 static void subfacet_destroy__(struct subfacet *);
195 static void subfacet_destroy_batch(struct dpif_backer *,
196 struct subfacet **, int n);
197 static void subfacet_reset_dp_stats(struct subfacet *,
198 struct dpif_flow_stats *);
199 static void subfacet_update_stats(struct subfacet *,
200 const struct dpif_flow_stats *);
201 static int subfacet_install(struct subfacet *,
202 const struct ofpbuf *odp_actions,
203 struct dpif_flow_stats *);
204 static void subfacet_uninstall(struct subfacet *);
206 /* A unique, non-overlapping instantiation of an OpenFlow flow.
208 * A facet associates a "struct flow", which represents the Open vSwitch
209 * userspace idea of an exact-match flow, with one or more subfacets.
210 * While the facet is created based on an exact-match flow, it is stored
211 * within the ofproto based on the wildcards that could be expressed
212 * based on the flow table and other configuration. (See the 'wc'
213 * description in "struct xlate_out" for more details.)
215 * Each subfacet tracks the datapath's idea of the flow equivalent to
216 * the facet. When the kernel module (or other dpif implementation) and
217 * Open vSwitch userspace agree on the definition of a flow key, there
218 * is exactly one subfacet per facet. If the dpif implementation
219 * supports more-specific flow matching than userspace, however, a facet
220 * can have more than one subfacet. Examples include the dpif
221 * implementation not supporting the same wildcards as userspace or some
222 * distinction in flow that userspace simply doesn't understand.
224 * Flow expiration works in terms of subfacets, so a facet must have at
225 * least one subfacet or it will never expire, leaking memory. */
228 struct ofproto_dpif *ofproto;
231 struct list subfacets;
232 long long int used; /* Time last used; time created if not used. */
235 struct flow flow; /* Flow of the creating subfacet. */
236 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
240 * - Do include packets and bytes sent "by hand", e.g. with
243 * - Do include packets and bytes that were obtained from the datapath
244 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
245 * DPIF_FP_ZERO_STATS).
247 * - Do not include packets or bytes that can be obtained from the
248 * datapath for any existing subfacet.
250 uint64_t packet_count; /* Number of packets received. */
251 uint64_t byte_count; /* Number of bytes received. */
253 /* Resubmit statistics. */
254 uint64_t prev_packet_count; /* Number of packets from last stats push. */
255 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
256 long long int prev_used; /* Used time from last stats push. */
259 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
260 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
261 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
263 struct xlate_out xout;
265 /* Storage for a single subfacet, to reduce malloc() time and space
266 * overhead. (A facet always has at least one subfacet and in the common
267 * case has exactly one subfacet. However, 'one_subfacet' may not
268 * always be valid, since it could have been removed after newer
269 * subfacets were pushed onto the 'subfacets' list.) */
270 struct subfacet one_subfacet;
272 long long int learn_rl; /* Rate limiter for facet_learn(). */
275 static struct facet *facet_create(const struct flow_miss *);
276 static void facet_remove(struct facet *);
277 static void facet_free(struct facet *);
279 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
280 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
281 const struct flow *);
282 static bool facet_revalidate(struct facet *);
283 static bool facet_check_consistency(struct facet *);
285 static void facet_flush_stats(struct facet *);
287 static void facet_reset_counters(struct facet *);
288 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
289 struct dpif_flow_stats *, bool may_learn);
290 static void facet_push_stats(struct facet *, bool may_learn);
291 static void facet_learn(struct facet *);
292 static void facet_account(struct facet *);
293 static void push_all_stats(void);
295 static bool facet_is_controller_flow(struct facet *);
298 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
302 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
303 struct list bundle_node; /* In struct ofbundle's "ports" list. */
304 struct cfm *cfm; /* Connectivity Fault Management, if any. */
305 struct bfd *bfd; /* BFD, if any. */
306 bool may_enable; /* May be enabled in bonds. */
307 bool is_tunnel; /* This port is a tunnel. */
308 long long int carrier_seq; /* Carrier status changes. */
309 struct ofport_dpif *peer; /* Peer if patch port. */
312 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
313 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
314 long long int stp_state_entered;
316 /* Queue to DSCP mapping. */
317 struct ofproto_port_queue *qdscp;
320 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
322 * This is deprecated. It is only for compatibility with broken device
323 * drivers in old versions of Linux that do not properly support VLANs when
324 * VLAN devices are not used. When broken device drivers are no longer in
325 * widespread use, we will delete these interfaces. */
326 ofp_port_t realdev_ofp_port;
330 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
332 * This is deprecated. It is only for compatibility with broken device drivers
333 * in old versions of Linux that do not properly support VLANs when VLAN
334 * devices are not used. When broken device drivers are no longer in
335 * widespread use, we will delete these interfaces. */
336 struct vlan_splinter {
337 struct hmap_node realdev_vid_node;
338 struct hmap_node vlandev_node;
339 ofp_port_t realdev_ofp_port;
340 ofp_port_t vlandev_ofp_port;
344 static void vsp_remove(struct ofport_dpif *);
345 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
347 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
350 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
353 static struct ofport_dpif *
354 ofport_dpif_cast(const struct ofport *ofport)
356 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
359 static void port_run(struct ofport_dpif *);
360 static void port_run_fast(struct ofport_dpif *);
361 static void port_wait(struct ofport_dpif *);
362 static int set_bfd(struct ofport *, const struct smap *);
363 static int set_cfm(struct ofport *, const struct cfm_settings *);
364 static void ofport_update_peer(struct ofport_dpif *);
365 static void run_fast_rl(void);
366 static int run_fast(struct ofproto *);
368 struct dpif_completion {
369 struct list list_node;
370 struct ofoperation *op;
373 /* Reasons that we might need to revalidate every facet, and corresponding
376 * A value of 0 means that there is no need to revalidate.
378 * It would be nice to have some cleaner way to integrate with coverage
379 * counters, but with only a few reasons I guess this is good enough for
381 enum revalidate_reason {
382 REV_RECONFIGURE = 1, /* Switch configuration changed. */
383 REV_STP, /* Spanning tree protocol port status change. */
384 REV_BOND, /* Bonding changed. */
385 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
386 REV_FLOW_TABLE, /* Flow table changed. */
387 REV_MAC_LEARNING, /* Mac learning changed. */
388 REV_INCONSISTENCY /* Facet self-check failed. */
390 COVERAGE_DEFINE(rev_reconfigure);
391 COVERAGE_DEFINE(rev_stp);
392 COVERAGE_DEFINE(rev_bond);
393 COVERAGE_DEFINE(rev_port_toggled);
394 COVERAGE_DEFINE(rev_flow_table);
395 COVERAGE_DEFINE(rev_mac_learning);
396 COVERAGE_DEFINE(rev_inconsistency);
398 struct avg_subfacet_rates {
399 double add_rate; /* Moving average of new flows created per minute. */
400 double del_rate; /* Moving average of flows deleted per minute. */
403 /* All datapaths of a given type share a single dpif backer instance. */
409 struct timer next_expiration;
411 struct ovs_rwlock odp_to_ofport_lock;
412 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
414 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
416 /* Facet revalidation flags applying to facets which use this backer. */
417 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
419 struct hmap drop_keys; /* Set of dropped odp keys. */
420 bool recv_set_enable; /* Enables or disables receiving packets. */
422 struct hmap subfacets;
423 struct governor *governor;
425 /* Subfacet statistics.
427 * These keep track of the total number of subfacets added and deleted and
428 * flow life span. They are useful for computing the flow rates stats
429 * exposed via "ovs-appctl dpif/show". The goal is to learn about
430 * traffic patterns in ways that we can use later to improve Open vSwitch
431 * performance in new situations. */
432 long long int created; /* Time when it is created. */
433 unsigned max_n_subfacet; /* Maximum number of flows */
434 unsigned avg_n_subfacet; /* Average number of flows. */
435 long long int avg_subfacet_life; /* Average life span of subfacets. */
437 /* The average number of subfacets... */
438 struct avg_subfacet_rates hourly; /* ...over the last hour. */
439 struct avg_subfacet_rates daily; /* ...over the last day. */
440 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
441 long long int last_minute; /* Last time 'hourly' was updated. */
443 /* Number of subfacets added or deleted since 'last_minute'. */
444 unsigned subfacet_add_count;
445 unsigned subfacet_del_count;
447 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
448 unsigned long long int total_subfacet_add_count;
449 unsigned long long int total_subfacet_del_count;
451 /* Number of upcall handling threads. */
452 unsigned int n_handler_threads;
455 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
456 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
458 static void drop_key_clear(struct dpif_backer *);
459 static void update_moving_averages(struct dpif_backer *backer);
461 struct ofproto_dpif {
462 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
464 struct dpif_backer *backer;
466 /* Special OpenFlow rules. */
467 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
468 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
469 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
472 struct netflow *netflow;
473 struct dpif_sflow *sflow;
474 struct dpif_ipfix *ipfix;
475 struct hmap bundles; /* Contains "struct ofbundle"s. */
476 struct mac_learning *ml;
477 bool has_bonded_bundles;
478 struct mbridge *mbridge;
481 struct classifier facets; /* Contains 'struct facet's. */
482 long long int consistency_rl;
484 struct netdev_stats stats; /* To account packets generated and consumed in
489 long long int stp_last_tick;
491 /* VLAN splinters. */
492 struct ovs_mutex vsp_mutex;
493 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
494 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
497 struct sset ports; /* Set of standard port names. */
498 struct sset ghost_ports; /* Ports with no datapath port. */
499 struct sset port_poll_set; /* Queued names for port_poll() reply. */
500 int port_poll_errno; /* Last errno for port_poll() reply. */
502 /* Per ofproto's dpif stats. */
507 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
510 /* By default, flows in the datapath are wildcarded (megaflows). They
511 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
512 static bool enable_megaflows = true;
514 /* All existing ofproto_dpif instances, indexed by ->up.name. */
515 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
517 static void ofproto_dpif_unixctl_init(void);
519 static inline struct ofproto_dpif *
520 ofproto_dpif_cast(const struct ofproto *ofproto)
522 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
523 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
526 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
527 ofp_port_t ofp_port);
528 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
529 const struct ofpbuf *packet, struct ds *);
532 static void handle_upcalls(struct dpif_backer *);
534 /* Flow expiration. */
535 static int expire(struct dpif_backer *);
538 static void send_netflow_active_timeouts(struct ofproto_dpif *);
541 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
543 /* Global variables. */
544 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
546 /* Initial mappings of port to bridge mappings. */
547 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
549 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
552 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
553 struct ofputil_flow_mod *fm)
555 ofproto_flow_mod(&ofproto->up, fm);
558 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
559 * Takes ownership of 'pin' and pin->packet. */
561 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
562 struct ofputil_packet_in *pin)
564 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
565 COVERAGE_INC(packet_in_overflow);
566 free(CONST_CAST(void *, pin->packet));
571 /* Factory functions. */
574 init(const struct shash *iface_hints)
576 struct shash_node *node;
578 /* Make a local copy, since we don't own 'iface_hints' elements. */
579 SHASH_FOR_EACH(node, iface_hints) {
580 const struct iface_hint *orig_hint = node->data;
581 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
583 new_hint->br_name = xstrdup(orig_hint->br_name);
584 new_hint->br_type = xstrdup(orig_hint->br_type);
585 new_hint->ofp_port = orig_hint->ofp_port;
587 shash_add(&init_ofp_ports, node->name, new_hint);
592 enumerate_types(struct sset *types)
594 dp_enumerate_types(types);
598 enumerate_names(const char *type, struct sset *names)
600 struct ofproto_dpif *ofproto;
603 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
604 if (strcmp(type, ofproto->up.type)) {
607 sset_add(names, ofproto->up.name);
614 del(const char *type, const char *name)
619 error = dpif_open(name, type, &dpif);
621 error = dpif_delete(dpif);
628 port_open_type(const char *datapath_type, const char *port_type)
630 return dpif_port_open_type(datapath_type, port_type);
633 /* Type functions. */
635 static void process_dpif_port_changes(struct dpif_backer *);
636 static void process_dpif_all_ports_changed(struct dpif_backer *);
637 static void process_dpif_port_change(struct dpif_backer *,
638 const char *devname);
639 static void process_dpif_port_error(struct dpif_backer *, int error);
641 static struct ofproto_dpif *
642 lookup_ofproto_dpif_by_port_name(const char *name)
644 struct ofproto_dpif *ofproto;
646 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
647 if (sset_contains(&ofproto->ports, name)) {
656 type_run(const char *type)
658 static long long int push_timer = LLONG_MIN;
659 struct dpif_backer *backer;
661 backer = shash_find_data(&all_dpif_backers, type);
663 /* This is not necessarily a problem, since backers are only
664 * created on demand. */
668 dpif_run(backer->dpif);
670 /* The most natural place to push facet statistics is when they're pulled
671 * from the datapath. However, when there are many flows in the datapath,
672 * this expensive operation can occur so frequently, that it reduces our
673 * ability to quickly set up flows. To reduce the cost, we push statistics
675 if (time_msec() > push_timer) {
676 push_timer = time_msec() + 2000;
680 /* If vswitchd started with other_config:flow_restore_wait set as "true",
681 * and the configuration has now changed to "false", enable receiving
682 * packets from the datapath. */
683 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
686 backer->recv_set_enable = true;
688 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
690 udpif_recv_set(backer->udpif, 0, false);
691 VLOG_ERR("Failed to enable receiving packets in dpif.");
694 udpif_recv_set(backer->udpif, n_handler_threads,
695 backer->recv_set_enable);
696 dpif_flow_flush(backer->dpif);
697 backer->need_revalidate = REV_RECONFIGURE;
700 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
701 * to reset the handler threads. */
702 if (backer->n_handler_threads != n_handler_threads) {
703 udpif_recv_set(backer->udpif, n_handler_threads,
704 backer->recv_set_enable);
705 backer->n_handler_threads = n_handler_threads;
708 if (backer->need_revalidate) {
709 struct ofproto_dpif *ofproto;
710 struct simap_node *node;
711 struct simap tmp_backers;
713 /* Handle tunnel garbage collection. */
714 simap_init(&tmp_backers);
715 simap_swap(&backer->tnl_backers, &tmp_backers);
717 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
718 struct ofport_dpif *iter;
720 if (backer != ofproto->backer) {
724 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
725 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
728 if (!iter->is_tunnel) {
732 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
733 namebuf, sizeof namebuf);
734 node = simap_find(&tmp_backers, dp_port);
736 simap_put(&backer->tnl_backers, dp_port, node->data);
737 simap_delete(&tmp_backers, node);
738 node = simap_find(&backer->tnl_backers, dp_port);
740 node = simap_find(&backer->tnl_backers, dp_port);
742 odp_port_t odp_port = ODPP_NONE;
744 if (!dpif_port_add(backer->dpif, iter->up.netdev,
746 simap_put(&backer->tnl_backers, dp_port,
747 odp_to_u32(odp_port));
748 node = simap_find(&backer->tnl_backers, dp_port);
753 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
754 if (tnl_port_reconfigure(iter, iter->up.netdev,
756 backer->need_revalidate = REV_RECONFIGURE;
761 SIMAP_FOR_EACH (node, &tmp_backers) {
762 dpif_port_del(backer->dpif, u32_to_odp(node->data));
764 simap_destroy(&tmp_backers);
766 switch (backer->need_revalidate) {
767 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
768 case REV_STP: COVERAGE_INC(rev_stp); break;
769 case REV_BOND: COVERAGE_INC(rev_bond); break;
770 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
771 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
772 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
773 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
775 backer->need_revalidate = 0;
777 /* Clear the drop_keys in case we should now be accepting some
778 * formerly dropped flows. */
779 drop_key_clear(backer);
781 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
782 struct facet *facet, *next;
783 struct ofport_dpif *ofport;
784 struct cls_cursor cursor;
785 struct ofbundle *bundle;
787 if (ofproto->backer != backer) {
791 ovs_rwlock_wrlock(&xlate_rwlock);
792 xlate_ofproto_set(ofproto, ofproto->up.name,
793 ofproto->backer->dpif, ofproto->miss_rule,
794 ofproto->no_packet_in_rule, ofproto->ml,
795 ofproto->stp, ofproto->mbridge,
796 ofproto->sflow, ofproto->ipfix,
797 ofproto->up.frag_handling,
798 ofproto->up.forward_bpdu,
799 connmgr_has_in_band(ofproto->up.connmgr),
800 ofproto->netflow != NULL);
802 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
803 xlate_bundle_set(ofproto, bundle, bundle->name,
804 bundle->vlan_mode, bundle->vlan,
805 bundle->trunks, bundle->use_priority_tags,
806 bundle->bond, bundle->lacp,
810 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
811 int stp_port = ofport->stp_port
812 ? stp_port_no(ofport->stp_port)
814 xlate_ofport_set(ofproto, ofport->bundle, ofport,
815 ofport->up.ofp_port, ofport->odp_port,
816 ofport->up.netdev, ofport->cfm,
817 ofport->bfd, ofport->peer, stp_port,
818 ofport->qdscp, ofport->n_qdscp,
819 ofport->up.pp.config, ofport->is_tunnel,
822 ovs_rwlock_unlock(&xlate_rwlock);
824 /* Only ofproto-dpif cares about the facet classifier so we just
825 * lock cls_cursor_init() to appease the thread safety analysis. */
826 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
827 cls_cursor_init(&cursor, &ofproto->facets, NULL);
828 ovs_rwlock_unlock(&ofproto->facets.rwlock);
829 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
830 facet_revalidate(facet);
835 udpif_revalidate(backer->udpif);
838 if (!backer->recv_set_enable) {
839 /* Wake up before a max of 1000ms. */
840 timer_set_duration(&backer->next_expiration, 1000);
841 } else if (timer_expired(&backer->next_expiration)) {
842 int delay = expire(backer);
843 timer_set_duration(&backer->next_expiration, delay);
846 process_dpif_port_changes(backer);
848 if (backer->governor) {
851 governor_run(backer->governor);
853 /* If the governor has shrunk to its minimum size and the number of
854 * subfacets has dwindled, then drop the governor entirely.
856 * For hysteresis, the number of subfacets to drop the governor is
857 * smaller than the number needed to trigger its creation. */
858 n_subfacets = hmap_count(&backer->subfacets);
859 if (n_subfacets * 4 < flow_eviction_threshold
860 && governor_is_idle(backer->governor)) {
861 governor_destroy(backer->governor);
862 backer->governor = NULL;
869 /* Check for and handle port changes in 'backer''s dpif. */
871 process_dpif_port_changes(struct dpif_backer *backer)
877 error = dpif_port_poll(backer->dpif, &devname);
883 process_dpif_all_ports_changed(backer);
887 process_dpif_port_change(backer, devname);
892 process_dpif_port_error(backer, error);
899 process_dpif_all_ports_changed(struct dpif_backer *backer)
901 struct ofproto_dpif *ofproto;
902 struct dpif_port dpif_port;
903 struct dpif_port_dump dump;
904 struct sset devnames;
907 sset_init(&devnames);
908 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
909 if (ofproto->backer == backer) {
910 struct ofport *ofport;
912 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
913 sset_add(&devnames, netdev_get_name(ofport->netdev));
917 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
918 sset_add(&devnames, dpif_port.name);
921 SSET_FOR_EACH (devname, &devnames) {
922 process_dpif_port_change(backer, devname);
924 sset_destroy(&devnames);
928 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
930 struct ofproto_dpif *ofproto;
931 struct dpif_port port;
933 /* Don't report on the datapath's device. */
934 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
938 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
939 &all_ofproto_dpifs) {
940 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
945 ofproto = lookup_ofproto_dpif_by_port_name(devname);
946 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
947 /* The port was removed. If we know the datapath,
948 * report it through poll_set(). If we don't, it may be
949 * notifying us of a removal we initiated, so ignore it.
950 * If there's a pending ENOBUFS, let it stand, since
951 * everything will be reevaluated. */
952 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
953 sset_add(&ofproto->port_poll_set, devname);
954 ofproto->port_poll_errno = 0;
956 } else if (!ofproto) {
957 /* The port was added, but we don't know with which
958 * ofproto we should associate it. Delete it. */
959 dpif_port_del(backer->dpif, port.port_no);
961 struct ofport_dpif *ofport;
963 ofport = ofport_dpif_cast(shash_find_data(
964 &ofproto->up.port_by_name, devname));
966 && ofport->odp_port != port.port_no
967 && !odp_port_to_ofport(backer, port.port_no))
969 /* 'ofport''s datapath port number has changed from
970 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
971 * structures to match. */
972 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
973 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
974 ofport->odp_port = port.port_no;
975 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
976 hash_odp_port(port.port_no));
977 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
978 backer->need_revalidate = REV_RECONFIGURE;
981 dpif_port_destroy(&port);
984 /* Propagate 'error' to all ofprotos based on 'backer'. */
986 process_dpif_port_error(struct dpif_backer *backer, int error)
988 struct ofproto_dpif *ofproto;
990 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
991 if (ofproto->backer == backer) {
992 sset_clear(&ofproto->port_poll_set);
993 ofproto->port_poll_errno = error;
999 dpif_backer_run_fast(struct dpif_backer *backer)
1001 handle_upcalls(backer);
1007 type_run_fast(const char *type)
1009 struct dpif_backer *backer;
1011 backer = shash_find_data(&all_dpif_backers, type);
1013 /* This is not necessarily a problem, since backers are only
1014 * created on demand. */
1018 return dpif_backer_run_fast(backer);
1024 static long long int port_rl = LLONG_MIN;
1026 if (time_msec() >= port_rl) {
1027 struct ofproto_dpif *ofproto;
1029 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1030 run_fast(&ofproto->up);
1032 port_rl = time_msec() + 200;
1037 type_wait(const char *type)
1039 struct dpif_backer *backer;
1041 backer = shash_find_data(&all_dpif_backers, type);
1043 /* This is not necessarily a problem, since backers are only
1044 * created on demand. */
1048 if (backer->governor) {
1049 governor_wait(backer->governor);
1052 timer_wait(&backer->next_expiration);
1053 dpif_wait(backer->dpif);
1054 udpif_wait(backer->udpif);
1057 /* Basic life-cycle. */
1059 static int add_internal_flows(struct ofproto_dpif *);
1061 static struct ofproto *
1064 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1065 return &ofproto->up;
1069 dealloc(struct ofproto *ofproto_)
1071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1076 close_dpif_backer(struct dpif_backer *backer)
1078 ovs_assert(backer->refcount > 0);
1080 if (--backer->refcount) {
1084 drop_key_clear(backer);
1085 hmap_destroy(&backer->drop_keys);
1087 udpif_destroy(backer->udpif);
1089 simap_destroy(&backer->tnl_backers);
1090 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1091 hmap_destroy(&backer->odp_to_ofport_map);
1092 shash_find_and_delete(&all_dpif_backers, backer->type);
1094 dpif_close(backer->dpif);
1096 ovs_assert(hmap_is_empty(&backer->subfacets));
1097 hmap_destroy(&backer->subfacets);
1098 governor_destroy(backer->governor);
1103 /* Datapath port slated for removal from datapath. */
1104 struct odp_garbage {
1105 struct list list_node;
1106 odp_port_t odp_port;
1110 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1112 struct dpif_backer *backer;
1113 struct dpif_port_dump port_dump;
1114 struct dpif_port port;
1115 struct shash_node *node;
1116 struct list garbage_list;
1117 struct odp_garbage *garbage, *next;
1123 backer = shash_find_data(&all_dpif_backers, type);
1130 backer_name = xasprintf("ovs-%s", type);
1132 /* Remove any existing datapaths, since we assume we're the only
1133 * userspace controlling the datapath. */
1135 dp_enumerate_names(type, &names);
1136 SSET_FOR_EACH(name, &names) {
1137 struct dpif *old_dpif;
1139 /* Don't remove our backer if it exists. */
1140 if (!strcmp(name, backer_name)) {
1144 if (dpif_open(name, type, &old_dpif)) {
1145 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1147 dpif_delete(old_dpif);
1148 dpif_close(old_dpif);
1151 sset_destroy(&names);
1153 backer = xmalloc(sizeof *backer);
1155 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1158 VLOG_ERR("failed to open datapath of type %s: %s", type,
1159 ovs_strerror(error));
1163 backer->udpif = udpif_create(backer, backer->dpif);
1165 backer->type = xstrdup(type);
1166 backer->governor = NULL;
1167 backer->refcount = 1;
1168 hmap_init(&backer->odp_to_ofport_map);
1169 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1170 hmap_init(&backer->drop_keys);
1171 hmap_init(&backer->subfacets);
1172 timer_set_duration(&backer->next_expiration, 1000);
1173 backer->need_revalidate = 0;
1174 simap_init(&backer->tnl_backers);
1175 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1178 if (backer->recv_set_enable) {
1179 dpif_flow_flush(backer->dpif);
1182 /* Loop through the ports already on the datapath and remove any
1183 * that we don't need anymore. */
1184 list_init(&garbage_list);
1185 dpif_port_dump_start(&port_dump, backer->dpif);
1186 while (dpif_port_dump_next(&port_dump, &port)) {
1187 node = shash_find(&init_ofp_ports, port.name);
1188 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1189 garbage = xmalloc(sizeof *garbage);
1190 garbage->odp_port = port.port_no;
1191 list_push_front(&garbage_list, &garbage->list_node);
1194 dpif_port_dump_done(&port_dump);
1196 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1197 dpif_port_del(backer->dpif, garbage->odp_port);
1198 list_remove(&garbage->list_node);
1202 shash_add(&all_dpif_backers, type, backer);
1204 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1206 VLOG_ERR("failed to listen on datapath of type %s: %s",
1207 type, ovs_strerror(error));
1208 close_dpif_backer(backer);
1211 udpif_recv_set(backer->udpif, n_handler_threads,
1212 backer->recv_set_enable);
1213 backer->n_handler_threads = n_handler_threads;
1215 backer->max_n_subfacet = 0;
1216 backer->created = time_msec();
1217 backer->last_minute = backer->created;
1218 memset(&backer->hourly, 0, sizeof backer->hourly);
1219 memset(&backer->daily, 0, sizeof backer->daily);
1220 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1221 backer->subfacet_add_count = 0;
1222 backer->subfacet_del_count = 0;
1223 backer->total_subfacet_add_count = 0;
1224 backer->total_subfacet_del_count = 0;
1225 backer->avg_n_subfacet = 0;
1226 backer->avg_subfacet_life = 0;
1232 construct(struct ofproto *ofproto_)
1234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1235 struct shash_node *node, *next;
1239 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1244 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1245 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1247 ofproto->netflow = NULL;
1248 ofproto->sflow = NULL;
1249 ofproto->ipfix = NULL;
1250 ofproto->stp = NULL;
1251 hmap_init(&ofproto->bundles);
1252 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1253 ofproto->mbridge = mbridge_create();
1254 ofproto->has_bonded_bundles = false;
1255 ovs_mutex_init(&ofproto->vsp_mutex);
1257 classifier_init(&ofproto->facets);
1258 ofproto->consistency_rl = LLONG_MIN;
1260 guarded_list_init(&ofproto->pins);
1262 ofproto_dpif_unixctl_init();
1264 hmap_init(&ofproto->vlandev_map);
1265 hmap_init(&ofproto->realdev_vid_map);
1267 sset_init(&ofproto->ports);
1268 sset_init(&ofproto->ghost_ports);
1269 sset_init(&ofproto->port_poll_set);
1270 ofproto->port_poll_errno = 0;
1272 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1273 struct iface_hint *iface_hint = node->data;
1275 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1276 /* Check if the datapath already has this port. */
1277 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1278 sset_add(&ofproto->ports, node->name);
1281 free(iface_hint->br_name);
1282 free(iface_hint->br_type);
1284 shash_delete(&init_ofp_ports, node);
1288 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1289 hash_string(ofproto->up.name, 0));
1290 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1292 ofproto_init_tables(ofproto_, N_TABLES);
1293 error = add_internal_flows(ofproto);
1294 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1297 ofproto->n_missed = 0;
1303 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1304 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1306 struct ofputil_flow_mod fm;
1309 match_init_catchall(&fm.match);
1311 match_set_reg(&fm.match, 0, id);
1312 fm.new_cookie = htonll(0);
1313 fm.cookie = htonll(0);
1314 fm.cookie_mask = htonll(0);
1315 fm.modify_cookie = false;
1316 fm.table_id = TBL_INTERNAL;
1317 fm.command = OFPFC_ADD;
1318 fm.idle_timeout = 0;
1319 fm.hard_timeout = 0;
1323 fm.ofpacts = ofpacts->data;
1324 fm.ofpacts_len = ofpacts->size;
1326 error = ofproto_flow_mod(&ofproto->up, &fm);
1328 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1329 id, ofperr_to_string(error));
1333 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1335 rule_dpif_unref(*rulep);
1344 add_internal_flows(struct ofproto_dpif *ofproto)
1346 struct ofpact_controller *controller;
1347 uint64_t ofpacts_stub[128 / 8];
1348 struct ofpbuf ofpacts;
1352 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1355 controller = ofpact_put_CONTROLLER(&ofpacts);
1356 controller->max_len = UINT16_MAX;
1357 controller->controller_id = 0;
1358 controller->reason = OFPR_NO_MATCH;
1359 ofpact_pad(&ofpacts);
1361 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1366 ofpbuf_clear(&ofpacts);
1367 error = add_internal_flow(ofproto, id++, &ofpacts,
1368 &ofproto->no_packet_in_rule);
1373 error = add_internal_flow(ofproto, id++, &ofpacts,
1374 &ofproto->drop_frags_rule);
1379 destruct(struct ofproto *ofproto_)
1381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1382 struct rule_dpif *rule, *next_rule;
1383 struct ofputil_packet_in *pin, *next_pin;
1384 struct facet *facet, *next_facet;
1385 struct cls_cursor cursor;
1386 struct oftable *table;
1389 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1390 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1391 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1392 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1393 facet_remove(facet);
1396 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1397 ovs_rwlock_wrlock(&xlate_rwlock);
1398 xlate_remove_ofproto(ofproto);
1399 ovs_rwlock_unlock(&xlate_rwlock);
1401 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1402 * use-after-free error. */
1403 udpif_revalidate(ofproto->backer->udpif);
1405 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1407 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1408 struct cls_cursor cursor;
1410 ovs_rwlock_rdlock(&table->cls.rwlock);
1411 cls_cursor_init(&cursor, &table->cls, NULL);
1412 ovs_rwlock_unlock(&table->cls.rwlock);
1413 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1414 ofproto_rule_delete(&ofproto->up, &rule->up);
1418 guarded_list_pop_all(&ofproto->pins, &pins);
1419 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1420 list_remove(&pin->list_node);
1421 free(CONST_CAST(void *, pin->packet));
1424 guarded_list_destroy(&ofproto->pins);
1426 mbridge_unref(ofproto->mbridge);
1428 netflow_destroy(ofproto->netflow);
1429 dpif_sflow_unref(ofproto->sflow);
1430 hmap_destroy(&ofproto->bundles);
1431 mac_learning_unref(ofproto->ml);
1433 classifier_destroy(&ofproto->facets);
1435 hmap_destroy(&ofproto->vlandev_map);
1436 hmap_destroy(&ofproto->realdev_vid_map);
1438 sset_destroy(&ofproto->ports);
1439 sset_destroy(&ofproto->ghost_ports);
1440 sset_destroy(&ofproto->port_poll_set);
1442 ovs_mutex_destroy(&ofproto->vsp_mutex);
1444 close_dpif_backer(ofproto->backer);
1448 run_fast(struct ofproto *ofproto_)
1450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1451 struct ofputil_packet_in *pin, *next_pin;
1452 struct ofport_dpif *ofport;
1455 /* Do not perform any periodic activity required by 'ofproto' while
1456 * waiting for flow restore to complete. */
1457 if (ofproto_get_flow_restore_wait()) {
1461 guarded_list_pop_all(&ofproto->pins, &pins);
1462 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1463 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1464 list_remove(&pin->list_node);
1465 free(CONST_CAST(void *, pin->packet));
1469 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1470 port_run_fast(ofport);
1477 run(struct ofproto *ofproto_)
1479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1480 struct ofport_dpif *ofport;
1481 struct ofbundle *bundle;
1484 if (mbridge_need_revalidate(ofproto->mbridge)) {
1485 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1486 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1487 mac_learning_flush(ofproto->ml);
1488 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1491 /* Do not perform any periodic activity below required by 'ofproto' while
1492 * waiting for flow restore to complete. */
1493 if (ofproto_get_flow_restore_wait()) {
1497 error = run_fast(ofproto_);
1502 if (ofproto->netflow) {
1503 if (netflow_run(ofproto->netflow)) {
1504 send_netflow_active_timeouts(ofproto);
1507 if (ofproto->sflow) {
1508 dpif_sflow_run(ofproto->sflow);
1510 if (ofproto->ipfix) {
1511 dpif_ipfix_run(ofproto->ipfix);
1514 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1517 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1522 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1523 if (mac_learning_run(ofproto->ml)) {
1524 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1526 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1528 /* Check the consistency of a random facet, to aid debugging. */
1529 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1530 if (time_msec() >= ofproto->consistency_rl
1531 && !classifier_is_empty(&ofproto->facets)
1532 && !ofproto->backer->need_revalidate) {
1533 struct cls_table *table;
1534 struct cls_rule *cr;
1535 struct facet *facet;
1537 ofproto->consistency_rl = time_msec() + 250;
1539 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1540 struct cls_table, hmap_node);
1541 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1543 facet = CONTAINER_OF(cr, struct facet, cr);
1545 if (!facet_check_consistency(facet)) {
1546 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1549 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1555 wait(struct ofproto *ofproto_)
1557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1558 struct ofport_dpif *ofport;
1559 struct ofbundle *bundle;
1561 if (ofproto_get_flow_restore_wait()) {
1565 if (ofproto->sflow) {
1566 dpif_sflow_wait(ofproto->sflow);
1568 if (ofproto->ipfix) {
1569 dpif_ipfix_wait(ofproto->ipfix);
1571 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1574 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1575 bundle_wait(bundle);
1577 if (ofproto->netflow) {
1578 netflow_wait(ofproto->netflow);
1580 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1581 mac_learning_wait(ofproto->ml);
1582 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1584 if (ofproto->backer->need_revalidate) {
1585 /* Shouldn't happen, but if it does just go around again. */
1586 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1587 poll_immediate_wake();
1592 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1594 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1595 struct cls_cursor cursor;
1596 size_t n_subfacets = 0;
1597 struct facet *facet;
1599 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1600 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1601 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1603 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1604 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1605 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1606 n_subfacets += list_size(&facet->subfacets);
1608 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1609 simap_increase(usage, "subfacets", n_subfacets);
1613 flush(struct ofproto *ofproto_)
1615 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1616 struct subfacet *subfacet, *next_subfacet;
1617 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1621 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1622 &ofproto->backer->subfacets) {
1623 if (subfacet->facet->ofproto != ofproto) {
1627 if (subfacet->path != SF_NOT_INSTALLED) {
1628 batch[n_batch++] = subfacet;
1629 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1630 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1634 subfacet_destroy(subfacet);
1639 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1644 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1645 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1647 *arp_match_ip = true;
1648 *actions = (OFPUTIL_A_OUTPUT |
1649 OFPUTIL_A_SET_VLAN_VID |
1650 OFPUTIL_A_SET_VLAN_PCP |
1651 OFPUTIL_A_STRIP_VLAN |
1652 OFPUTIL_A_SET_DL_SRC |
1653 OFPUTIL_A_SET_DL_DST |
1654 OFPUTIL_A_SET_NW_SRC |
1655 OFPUTIL_A_SET_NW_DST |
1656 OFPUTIL_A_SET_NW_TOS |
1657 OFPUTIL_A_SET_TP_SRC |
1658 OFPUTIL_A_SET_TP_DST |
1663 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1666 struct dpif_dp_stats s;
1667 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1670 strcpy(ots->name, "classifier");
1672 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1673 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1674 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1675 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1677 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1678 ots->lookup_count = htonll(n_lookup);
1679 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1682 static struct ofport *
1685 struct ofport_dpif *port = xmalloc(sizeof *port);
1690 port_dealloc(struct ofport *port_)
1692 struct ofport_dpif *port = ofport_dpif_cast(port_);
1697 port_construct(struct ofport *port_)
1699 struct ofport_dpif *port = ofport_dpif_cast(port_);
1700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1701 const struct netdev *netdev = port->up.netdev;
1702 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1703 struct dpif_port dpif_port;
1706 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1707 port->bundle = NULL;
1710 port->may_enable = true;
1711 port->stp_port = NULL;
1712 port->stp_state = STP_DISABLED;
1713 port->is_tunnel = false;
1717 port->realdev_ofp_port = 0;
1718 port->vlandev_vid = 0;
1719 port->carrier_seq = netdev_get_carrier_resets(netdev);
1721 if (netdev_vport_is_patch(netdev)) {
1722 /* By bailing out here, we don't submit the port to the sFlow module
1723 * to be considered for counter polling export. This is correct
1724 * because the patch port represents an interface that sFlow considers
1725 * to be "internal" to the switch as a whole, and therefore not an
1726 * candidate for counter polling. */
1727 port->odp_port = ODPP_NONE;
1728 ofport_update_peer(port);
1732 error = dpif_port_query_by_name(ofproto->backer->dpif,
1733 netdev_vport_get_dpif_port(netdev, namebuf,
1740 port->odp_port = dpif_port.port_no;
1742 if (netdev_get_tunnel_config(netdev)) {
1743 tnl_port_add(port, port->up.netdev, port->odp_port);
1744 port->is_tunnel = true;
1746 /* Sanity-check that a mapping doesn't already exist. This
1747 * shouldn't happen for non-tunnel ports. */
1748 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1749 VLOG_ERR("port %s already has an OpenFlow port number",
1751 dpif_port_destroy(&dpif_port);
1755 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1756 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1757 hash_odp_port(port->odp_port));
1758 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1760 dpif_port_destroy(&dpif_port);
1762 if (ofproto->sflow) {
1763 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1770 port_destruct(struct ofport *port_)
1772 struct ofport_dpif *port = ofport_dpif_cast(port_);
1773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1774 const char *devname = netdev_get_name(port->up.netdev);
1775 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1776 const char *dp_port_name;
1778 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1779 ovs_rwlock_wrlock(&xlate_rwlock);
1780 xlate_ofport_remove(port);
1781 ovs_rwlock_unlock(&xlate_rwlock);
1783 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1785 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1786 /* The underlying device is still there, so delete it. This
1787 * happens when the ofproto is being destroyed, since the caller
1788 * assumes that removal of attached ports will happen as part of
1790 if (!port->is_tunnel) {
1791 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1796 port->peer->peer = NULL;
1800 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1801 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1802 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1803 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1807 sset_find_and_delete(&ofproto->ports, devname);
1808 sset_find_and_delete(&ofproto->ghost_ports, devname);
1809 bundle_remove(port_);
1810 set_cfm(port_, NULL);
1811 set_bfd(port_, NULL);
1812 if (ofproto->sflow) {
1813 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1820 port_modified(struct ofport *port_)
1822 struct ofport_dpif *port = ofport_dpif_cast(port_);
1824 if (port->bundle && port->bundle->bond) {
1825 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1829 cfm_set_netdev(port->cfm, port->up.netdev);
1833 bfd_set_netdev(port->bfd, port->up.netdev);
1836 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1838 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1842 ofport_update_peer(port);
1846 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1848 struct ofport_dpif *port = ofport_dpif_cast(port_);
1849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1850 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1852 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1853 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1854 OFPUTIL_PC_NO_PACKET_IN)) {
1855 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1857 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1858 bundle_update(port->bundle);
1864 set_sflow(struct ofproto *ofproto_,
1865 const struct ofproto_sflow_options *sflow_options)
1867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1868 struct dpif_sflow *ds = ofproto->sflow;
1870 if (sflow_options) {
1872 struct ofport_dpif *ofport;
1874 ds = ofproto->sflow = dpif_sflow_create();
1875 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1876 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1878 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1880 dpif_sflow_set_options(ds, sflow_options);
1883 dpif_sflow_unref(ds);
1884 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1885 ofproto->sflow = NULL;
1893 struct ofproto *ofproto_,
1894 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1895 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1896 size_t n_flow_exporters_options)
1898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1899 struct dpif_ipfix *di = ofproto->ipfix;
1900 bool has_options = bridge_exporter_options || flow_exporters_options;
1902 if (has_options && !di) {
1903 di = ofproto->ipfix = dpif_ipfix_create();
1907 /* Call set_options in any case to cleanly flush the flow
1908 * caches in the last exporters that are to be destroyed. */
1909 dpif_ipfix_set_options(
1910 di, bridge_exporter_options, flow_exporters_options,
1911 n_flow_exporters_options);
1914 dpif_ipfix_unref(di);
1915 ofproto->ipfix = NULL;
1923 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1925 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1932 struct ofproto_dpif *ofproto;
1934 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1935 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1936 ofport->cfm = cfm_create(ofport->up.netdev);
1939 if (cfm_configure(ofport->cfm, s)) {
1945 cfm_unref(ofport->cfm);
1951 get_cfm_status(const struct ofport *ofport_,
1952 struct ofproto_cfm_status *status)
1954 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1957 status->faults = cfm_get_fault(ofport->cfm);
1958 status->remote_opstate = cfm_get_opup(ofport->cfm);
1959 status->health = cfm_get_health(ofport->cfm);
1960 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1968 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1971 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1975 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1976 cfg, ofport->up.netdev);
1977 if (ofport->bfd != old) {
1978 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1985 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1987 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1990 bfd_get_status(ofport->bfd, smap);
1997 /* Spanning Tree. */
2000 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2002 struct ofproto_dpif *ofproto = ofproto_;
2003 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2004 struct ofport_dpif *ofport;
2006 ofport = stp_port_get_aux(sp);
2008 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2009 ofproto->up.name, port_num);
2011 struct eth_header *eth = pkt->l2;
2013 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2014 if (eth_addr_is_zero(eth->eth_src)) {
2015 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2016 "with unknown MAC", ofproto->up.name, port_num);
2018 send_packet(ofport, pkt);
2024 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2026 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2030 /* Only revalidate flows if the configuration changed. */
2031 if (!s != !ofproto->stp) {
2032 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2036 if (!ofproto->stp) {
2037 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2038 send_bpdu_cb, ofproto);
2039 ofproto->stp_last_tick = time_msec();
2042 stp_set_bridge_id(ofproto->stp, s->system_id);
2043 stp_set_bridge_priority(ofproto->stp, s->priority);
2044 stp_set_hello_time(ofproto->stp, s->hello_time);
2045 stp_set_max_age(ofproto->stp, s->max_age);
2046 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2048 struct ofport *ofport;
2050 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2051 set_stp_port(ofport, NULL);
2054 stp_unref(ofproto->stp);
2055 ofproto->stp = NULL;
2062 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2068 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2069 s->designated_root = stp_get_designated_root(ofproto->stp);
2070 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2079 update_stp_port_state(struct ofport_dpif *ofport)
2081 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2082 enum stp_state state;
2084 /* Figure out new state. */
2085 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2089 if (ofport->stp_state != state) {
2090 enum ofputil_port_state of_state;
2093 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2094 netdev_get_name(ofport->up.netdev),
2095 stp_state_name(ofport->stp_state),
2096 stp_state_name(state));
2097 if (stp_learn_in_state(ofport->stp_state)
2098 != stp_learn_in_state(state)) {
2099 /* xxx Learning action flows should also be flushed. */
2100 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2101 mac_learning_flush(ofproto->ml);
2102 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2104 fwd_change = stp_forward_in_state(ofport->stp_state)
2105 != stp_forward_in_state(state);
2107 ofproto->backer->need_revalidate = REV_STP;
2108 ofport->stp_state = state;
2109 ofport->stp_state_entered = time_msec();
2111 if (fwd_change && ofport->bundle) {
2112 bundle_update(ofport->bundle);
2115 /* Update the STP state bits in the OpenFlow port description. */
2116 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2117 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2118 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2119 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2120 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2122 ofproto_port_set_state(&ofport->up, of_state);
2126 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2127 * caller is responsible for assigning STP port numbers and ensuring
2128 * there are no duplicates. */
2130 set_stp_port(struct ofport *ofport_,
2131 const struct ofproto_port_stp_settings *s)
2133 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2135 struct stp_port *sp = ofport->stp_port;
2137 if (!s || !s->enable) {
2139 ofport->stp_port = NULL;
2140 stp_port_disable(sp);
2141 update_stp_port_state(ofport);
2144 } else if (sp && stp_port_no(sp) != s->port_num
2145 && ofport == stp_port_get_aux(sp)) {
2146 /* The port-id changed, so disable the old one if it's not
2147 * already in use by another port. */
2148 stp_port_disable(sp);
2151 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2152 stp_port_enable(sp);
2154 stp_port_set_aux(sp, ofport);
2155 stp_port_set_priority(sp, s->priority);
2156 stp_port_set_path_cost(sp, s->path_cost);
2158 update_stp_port_state(ofport);
2164 get_stp_port_status(struct ofport *ofport_,
2165 struct ofproto_port_stp_status *s)
2167 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2169 struct stp_port *sp = ofport->stp_port;
2171 if (!ofproto->stp || !sp) {
2177 s->port_id = stp_port_get_id(sp);
2178 s->state = stp_port_get_state(sp);
2179 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2180 s->role = stp_port_get_role(sp);
2181 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2187 stp_run(struct ofproto_dpif *ofproto)
2190 long long int now = time_msec();
2191 long long int elapsed = now - ofproto->stp_last_tick;
2192 struct stp_port *sp;
2195 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2196 ofproto->stp_last_tick = now;
2198 while (stp_get_changed_port(ofproto->stp, &sp)) {
2199 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2202 update_stp_port_state(ofport);
2206 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2207 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2208 mac_learning_flush(ofproto->ml);
2209 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2215 stp_wait(struct ofproto_dpif *ofproto)
2218 poll_timer_wait(1000);
2223 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2226 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2229 if (ofport->n_qdscp != n_qdscp
2230 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2231 n_qdscp * sizeof *qdscp))) {
2232 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2233 free(ofport->qdscp);
2234 ofport->qdscp = n_qdscp
2235 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2237 ofport->n_qdscp = n_qdscp;
2245 /* Expires all MAC learning entries associated with 'bundle' and forces its
2246 * ofproto to revalidate every flow.
2248 * Normally MAC learning entries are removed only from the ofproto associated
2249 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2250 * are removed from every ofproto. When patch ports and SLB bonds are in use
2251 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2252 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2253 * with the host from which it migrated. */
2255 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2257 struct ofproto_dpif *ofproto = bundle->ofproto;
2258 struct mac_learning *ml = ofproto->ml;
2259 struct mac_entry *mac, *next_mac;
2261 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2262 ovs_rwlock_wrlock(&ml->rwlock);
2263 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2264 if (mac->port.p == bundle) {
2266 struct ofproto_dpif *o;
2268 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2270 struct mac_entry *e;
2272 ovs_rwlock_wrlock(&o->ml->rwlock);
2273 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2275 mac_learning_expire(o->ml, e);
2277 ovs_rwlock_unlock(&o->ml->rwlock);
2282 mac_learning_expire(ml, mac);
2285 ovs_rwlock_unlock(&ml->rwlock);
2288 static struct ofbundle *
2289 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2291 struct ofbundle *bundle;
2293 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2294 &ofproto->bundles) {
2295 if (bundle->aux == aux) {
2303 bundle_update(struct ofbundle *bundle)
2305 struct ofport_dpif *port;
2307 bundle->floodable = true;
2308 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2309 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2310 || !stp_forward_in_state(port->stp_state)) {
2311 bundle->floodable = false;
2318 bundle_del_port(struct ofport_dpif *port)
2320 struct ofbundle *bundle = port->bundle;
2322 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2324 list_remove(&port->bundle_node);
2325 port->bundle = NULL;
2328 lacp_slave_unregister(bundle->lacp, port);
2331 bond_slave_unregister(bundle->bond, port);
2334 bundle_update(bundle);
2338 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2339 struct lacp_slave_settings *lacp)
2341 struct ofport_dpif *port;
2343 port = get_ofp_port(bundle->ofproto, ofp_port);
2348 if (port->bundle != bundle) {
2349 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2351 bundle_remove(&port->up);
2354 port->bundle = bundle;
2355 list_push_back(&bundle->ports, &port->bundle_node);
2356 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2357 || !stp_forward_in_state(port->stp_state)) {
2358 bundle->floodable = false;
2362 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2363 lacp_slave_register(bundle->lacp, port, lacp);
2370 bundle_destroy(struct ofbundle *bundle)
2372 struct ofproto_dpif *ofproto;
2373 struct ofport_dpif *port, *next_port;
2379 ofproto = bundle->ofproto;
2380 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2382 ovs_rwlock_wrlock(&xlate_rwlock);
2383 xlate_bundle_remove(bundle);
2384 ovs_rwlock_unlock(&xlate_rwlock);
2386 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2387 bundle_del_port(port);
2390 bundle_flush_macs(bundle, true);
2391 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2393 free(bundle->trunks);
2394 lacp_unref(bundle->lacp);
2395 bond_unref(bundle->bond);
2400 bundle_set(struct ofproto *ofproto_, void *aux,
2401 const struct ofproto_bundle_settings *s)
2403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2404 bool need_flush = false;
2405 struct ofport_dpif *port;
2406 struct ofbundle *bundle;
2407 unsigned long *trunks;
2413 bundle_destroy(bundle_lookup(ofproto, aux));
2417 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2418 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2420 bundle = bundle_lookup(ofproto, aux);
2422 bundle = xmalloc(sizeof *bundle);
2424 bundle->ofproto = ofproto;
2425 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2426 hash_pointer(aux, 0));
2428 bundle->name = NULL;
2430 list_init(&bundle->ports);
2431 bundle->vlan_mode = PORT_VLAN_TRUNK;
2433 bundle->trunks = NULL;
2434 bundle->use_priority_tags = s->use_priority_tags;
2435 bundle->lacp = NULL;
2436 bundle->bond = NULL;
2438 bundle->floodable = true;
2439 mbridge_register_bundle(ofproto->mbridge, bundle);
2442 if (!bundle->name || strcmp(s->name, bundle->name)) {
2444 bundle->name = xstrdup(s->name);
2449 if (!bundle->lacp) {
2450 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2451 bundle->lacp = lacp_create();
2453 lacp_configure(bundle->lacp, s->lacp);
2455 lacp_unref(bundle->lacp);
2456 bundle->lacp = NULL;
2459 /* Update set of ports. */
2461 for (i = 0; i < s->n_slaves; i++) {
2462 if (!bundle_add_port(bundle, s->slaves[i],
2463 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2467 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2468 struct ofport_dpif *next_port;
2470 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2471 for (i = 0; i < s->n_slaves; i++) {
2472 if (s->slaves[i] == port->up.ofp_port) {
2477 bundle_del_port(port);
2481 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2483 if (list_is_empty(&bundle->ports)) {
2484 bundle_destroy(bundle);
2488 /* Set VLAN tagging mode */
2489 if (s->vlan_mode != bundle->vlan_mode
2490 || s->use_priority_tags != bundle->use_priority_tags) {
2491 bundle->vlan_mode = s->vlan_mode;
2492 bundle->use_priority_tags = s->use_priority_tags;
2497 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2498 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2500 if (vlan != bundle->vlan) {
2501 bundle->vlan = vlan;
2505 /* Get trunked VLANs. */
2506 switch (s->vlan_mode) {
2507 case PORT_VLAN_ACCESS:
2511 case PORT_VLAN_TRUNK:
2512 trunks = CONST_CAST(unsigned long *, s->trunks);
2515 case PORT_VLAN_NATIVE_UNTAGGED:
2516 case PORT_VLAN_NATIVE_TAGGED:
2517 if (vlan != 0 && (!s->trunks
2518 || !bitmap_is_set(s->trunks, vlan)
2519 || bitmap_is_set(s->trunks, 0))) {
2520 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2522 trunks = bitmap_clone(s->trunks, 4096);
2524 trunks = bitmap_allocate1(4096);
2526 bitmap_set1(trunks, vlan);
2527 bitmap_set0(trunks, 0);
2529 trunks = CONST_CAST(unsigned long *, s->trunks);
2536 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2537 free(bundle->trunks);
2538 if (trunks == s->trunks) {
2539 bundle->trunks = vlan_bitmap_clone(trunks);
2541 bundle->trunks = trunks;
2546 if (trunks != s->trunks) {
2551 if (!list_is_short(&bundle->ports)) {
2552 bundle->ofproto->has_bonded_bundles = true;
2554 if (bond_reconfigure(bundle->bond, s->bond)) {
2555 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2558 bundle->bond = bond_create(s->bond);
2559 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2562 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2563 bond_slave_register(bundle->bond, port, port->up.netdev);
2566 bond_unref(bundle->bond);
2567 bundle->bond = NULL;
2570 /* If we changed something that would affect MAC learning, un-learn
2571 * everything on this port and force flow revalidation. */
2573 bundle_flush_macs(bundle, false);
2580 bundle_remove(struct ofport *port_)
2582 struct ofport_dpif *port = ofport_dpif_cast(port_);
2583 struct ofbundle *bundle = port->bundle;
2586 bundle_del_port(port);
2587 if (list_is_empty(&bundle->ports)) {
2588 bundle_destroy(bundle);
2589 } else if (list_is_short(&bundle->ports)) {
2590 bond_unref(bundle->bond);
2591 bundle->bond = NULL;
2597 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2599 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2600 struct ofport_dpif *port = port_;
2601 uint8_t ea[ETH_ADDR_LEN];
2604 error = netdev_get_etheraddr(port->up.netdev, ea);
2606 struct ofpbuf packet;
2609 ofpbuf_init(&packet, 0);
2610 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2612 memcpy(packet_pdu, pdu, pdu_size);
2614 send_packet(port, &packet);
2615 ofpbuf_uninit(&packet);
2617 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2618 "%s (%s)", port->bundle->name,
2619 netdev_get_name(port->up.netdev), ovs_strerror(error));
2624 bundle_send_learning_packets(struct ofbundle *bundle)
2626 struct ofproto_dpif *ofproto = bundle->ofproto;
2627 struct ofpbuf *learning_packet;
2628 int error, n_packets, n_errors;
2629 struct mac_entry *e;
2630 struct list packets;
2632 list_init(&packets);
2633 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2634 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2635 if (e->port.p != bundle) {
2638 learning_packet = bond_compose_learning_packet(bundle->bond,
2641 learning_packet->private_p = port_void;
2642 list_push_back(&packets, &learning_packet->list_node);
2645 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2647 error = n_packets = n_errors = 0;
2648 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2651 ret = send_packet(learning_packet->private_p, learning_packet);
2658 ofpbuf_list_delete(&packets);
2661 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2662 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2663 "packets, last error was: %s",
2664 bundle->name, n_errors, n_packets, ovs_strerror(error));
2666 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2667 bundle->name, n_packets);
2672 bundle_run(struct ofbundle *bundle)
2675 lacp_run(bundle->lacp, send_pdu_cb);
2678 struct ofport_dpif *port;
2680 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2681 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2684 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2685 bundle->ofproto->backer->need_revalidate = REV_BOND;
2688 if (bond_should_send_learning_packets(bundle->bond)) {
2689 bundle_send_learning_packets(bundle);
2695 bundle_wait(struct ofbundle *bundle)
2698 lacp_wait(bundle->lacp);
2701 bond_wait(bundle->bond);
2708 mirror_set__(struct ofproto *ofproto_, void *aux,
2709 const struct ofproto_mirror_settings *s)
2711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2712 struct ofbundle **srcs, **dsts;
2717 mirror_destroy(ofproto->mbridge, aux);
2721 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2722 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2724 for (i = 0; i < s->n_srcs; i++) {
2725 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2728 for (i = 0; i < s->n_dsts; i++) {
2729 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2732 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2733 s->n_dsts, s->src_vlans,
2734 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2741 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2742 uint64_t *packets, uint64_t *bytes)
2745 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2750 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2753 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2754 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2755 mac_learning_flush(ofproto->ml);
2757 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2762 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2765 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2766 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2770 forward_bpdu_changed(struct ofproto *ofproto_)
2772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2773 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2777 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2781 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2782 mac_learning_set_idle_time(ofproto->ml, idle_time);
2783 mac_learning_set_max_entries(ofproto->ml, max_entries);
2784 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2789 static struct ofport_dpif *
2790 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2792 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2793 return ofport ? ofport_dpif_cast(ofport) : NULL;
2796 static struct ofport_dpif *
2797 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2799 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2800 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2804 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2805 struct ofproto_port *ofproto_port,
2806 struct dpif_port *dpif_port)
2808 ofproto_port->name = dpif_port->name;
2809 ofproto_port->type = dpif_port->type;
2810 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2814 ofport_update_peer(struct ofport_dpif *ofport)
2816 const struct ofproto_dpif *ofproto;
2817 struct dpif_backer *backer;
2820 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2824 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2825 backer->need_revalidate = REV_RECONFIGURE;
2828 ofport->peer->peer = NULL;
2829 ofport->peer = NULL;
2832 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2837 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2838 struct ofport *peer_ofport;
2839 struct ofport_dpif *peer;
2842 if (ofproto->backer != backer) {
2846 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2851 peer = ofport_dpif_cast(peer_ofport);
2852 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2853 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2855 ofport->peer = peer;
2856 ofport->peer->peer = ofport;
2866 port_run_fast(struct ofport_dpif *ofport)
2868 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2869 struct ofpbuf packet;
2871 ofpbuf_init(&packet, 0);
2872 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2873 send_packet(ofport, &packet);
2874 ofpbuf_uninit(&packet);
2877 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2878 struct ofpbuf packet;
2880 ofpbuf_init(&packet, 0);
2881 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2882 send_packet(ofport, &packet);
2883 ofpbuf_uninit(&packet);
2888 port_run(struct ofport_dpif *ofport)
2890 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2891 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2892 bool enable = netdev_get_carrier(ofport->up.netdev);
2893 bool cfm_enable = false;
2894 bool bfd_enable = false;
2896 ofport->carrier_seq = carrier_seq;
2898 port_run_fast(ofport);
2901 int cfm_opup = cfm_get_opup(ofport->cfm);
2903 cfm_run(ofport->cfm);
2904 cfm_enable = !cfm_get_fault(ofport->cfm);
2906 if (cfm_opup >= 0) {
2907 cfm_enable = cfm_enable && cfm_opup;
2912 bfd_run(ofport->bfd);
2913 bfd_enable = bfd_forwarding(ofport->bfd);
2916 if (ofport->bfd || ofport->cfm) {
2917 enable = enable && (cfm_enable || bfd_enable);
2920 if (ofport->bundle) {
2921 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2922 if (carrier_changed) {
2923 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2927 if (ofport->may_enable != enable) {
2928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2929 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2932 ofport->may_enable = enable;
2936 port_wait(struct ofport_dpif *ofport)
2939 cfm_wait(ofport->cfm);
2943 bfd_wait(ofport->bfd);
2948 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2949 struct ofproto_port *ofproto_port)
2951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2952 struct dpif_port dpif_port;
2955 if (sset_contains(&ofproto->ghost_ports, devname)) {
2956 const char *type = netdev_get_type_from_name(devname);
2958 /* We may be called before ofproto->up.port_by_name is populated with
2959 * the appropriate ofport. For this reason, we must get the name and
2960 * type from the netdev layer directly. */
2962 const struct ofport *ofport;
2964 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2965 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2966 ofproto_port->name = xstrdup(devname);
2967 ofproto_port->type = xstrdup(type);
2973 if (!sset_contains(&ofproto->ports, devname)) {
2976 error = dpif_port_query_by_name(ofproto->backer->dpif,
2977 devname, &dpif_port);
2979 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2985 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2988 const char *devname = netdev_get_name(netdev);
2989 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2990 const char *dp_port_name;
2992 if (netdev_vport_is_patch(netdev)) {
2993 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2997 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2998 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2999 odp_port_t port_no = ODPP_NONE;
3002 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3006 if (netdev_get_tunnel_config(netdev)) {
3007 simap_put(&ofproto->backer->tnl_backers,
3008 dp_port_name, odp_to_u32(port_no));
3012 if (netdev_get_tunnel_config(netdev)) {
3013 sset_add(&ofproto->ghost_ports, devname);
3015 sset_add(&ofproto->ports, devname);
3021 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3024 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3031 sset_find_and_delete(&ofproto->ghost_ports,
3032 netdev_get_name(ofport->up.netdev));
3033 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3034 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3035 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3037 /* The caller is going to close ofport->up.netdev. If this is a
3038 * bonded port, then the bond is using that netdev, so remove it
3039 * from the bond. The client will need to reconfigure everything
3040 * after deleting ports, so then the slave will get re-added. */
3041 bundle_remove(&ofport->up);
3048 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3050 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3055 error = netdev_get_stats(ofport->up.netdev, stats);
3057 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3060 /* ofproto->stats.tx_packets represents packets that we created
3061 * internally and sent to some port (e.g. packets sent with
3062 * send_packet()). Account for them as if they had come from
3063 * OFPP_LOCAL and got forwarded. */
3065 if (stats->rx_packets != UINT64_MAX) {
3066 stats->rx_packets += ofproto->stats.tx_packets;
3069 if (stats->rx_bytes != UINT64_MAX) {
3070 stats->rx_bytes += ofproto->stats.tx_bytes;
3073 /* ofproto->stats.rx_packets represents packets that were received on
3074 * some port and we processed internally and dropped (e.g. STP).
3075 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3077 if (stats->tx_packets != UINT64_MAX) {
3078 stats->tx_packets += ofproto->stats.rx_packets;
3081 if (stats->tx_bytes != UINT64_MAX) {
3082 stats->tx_bytes += ofproto->stats.rx_bytes;
3089 struct port_dump_state {
3094 struct ofproto_port port;
3099 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3101 *statep = xzalloc(sizeof(struct port_dump_state));
3106 port_dump_next(const struct ofproto *ofproto_, void *state_,
3107 struct ofproto_port *port)
3109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3110 struct port_dump_state *state = state_;
3111 const struct sset *sset;
3112 struct sset_node *node;
3114 if (state->has_port) {
3115 ofproto_port_destroy(&state->port);
3116 state->has_port = false;
3118 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3119 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3122 error = port_query_by_name(ofproto_, node->name, &state->port);
3124 *port = state->port;
3125 state->has_port = true;
3127 } else if (error != ENODEV) {
3132 if (!state->ghost) {
3133 state->ghost = true;
3136 return port_dump_next(ofproto_, state_, port);
3143 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3145 struct port_dump_state *state = state_;
3147 if (state->has_port) {
3148 ofproto_port_destroy(&state->port);
3155 port_poll(const struct ofproto *ofproto_, char **devnamep)
3157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3159 if (ofproto->port_poll_errno) {
3160 int error = ofproto->port_poll_errno;
3161 ofproto->port_poll_errno = 0;
3165 if (sset_is_empty(&ofproto->port_poll_set)) {
3169 *devnamep = sset_pop(&ofproto->port_poll_set);
3174 port_poll_wait(const struct ofproto *ofproto_)
3176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3177 dpif_port_poll_wait(ofproto->backer->dpif);
3181 port_is_lacp_current(const struct ofport *ofport_)
3183 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3184 return (ofport->bundle && ofport->bundle->lacp
3185 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3189 /* Upcall handling. */
3191 struct flow_miss_op {
3192 struct dpif_op dpif_op;
3194 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3195 struct xlate_out xout;
3196 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3198 struct ofpbuf mask; /* Flow mask for "put" ops. */
3199 struct odputil_keybuf maskbuf;
3201 /* If this is a "put" op, then a pointer to the subfacet that should
3202 * be marked as uninstalled if the operation fails. */
3203 struct subfacet *subfacet;
3206 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3207 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3208 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3209 * return value of true). However, for short flows the cost of bookkeeping is
3210 * much higher than the benefits, so when the datapath holds a large number of
3211 * flows we impose some heuristics to decide which flows are likely to be worth
3214 flow_miss_should_make_facet(struct flow_miss *miss)
3216 struct dpif_backer *backer = miss->ofproto->backer;
3219 switch (flow_miss_model) {
3220 case OFPROTO_HANDLE_MISS_AUTO:
3222 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3224 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3228 if (!backer->governor) {
3231 n_subfacets = hmap_count(&backer->subfacets);
3232 if (n_subfacets * 2 <= flow_eviction_threshold) {
3236 backer->governor = governor_create();
3239 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3240 return governor_should_install_flow(backer->governor, hash,
3241 miss->stats.n_packets);
3244 /* Handles 'miss', which matches 'facet'. May add any required datapath
3245 * operations to 'ops', incrementing '*n_ops' for each new op.
3247 * All of the packets in 'miss' are considered to have arrived at time
3248 * 'miss->stats.used'. This is really important only for new facets: if we
3249 * just called time_msec() here, then the new subfacet or its packets could
3250 * look (occasionally) as though it was used some time after the facet was
3251 * used. That can make a one-packet flow look like it has a nonzero duration,
3252 * which looks odd in e.g. NetFlow statistics. */
3254 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3255 struct flow_miss_op *ops, size_t *n_ops)
3257 enum subfacet_path want_path;
3258 struct subfacet *subfacet;
3260 facet->packet_count += miss->stats.n_packets;
3261 facet->prev_packet_count += miss->stats.n_packets;
3262 facet->byte_count += miss->stats.n_bytes;
3263 facet->prev_byte_count += miss->stats.n_bytes;
3265 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3267 /* Don't install the flow if it's the result of the "userspace"
3268 * action for an already installed facet. This can occur when a
3269 * datapath flow with wildcards has a "userspace" action and flows
3270 * sent to userspace result in a different subfacet, which will then
3271 * be rejected as overlapping by the datapath. */
3272 if (miss->upcall_type == DPIF_UC_ACTION
3273 && !list_is_empty(&facet->subfacets)) {
3277 subfacet = subfacet_create(facet, miss);
3278 if (subfacet->path != want_path) {
3279 struct flow_miss_op *op = &ops[(*n_ops)++];
3280 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3282 subfacet->path = want_path;
3284 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3285 if (enable_megaflows) {
3286 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3287 &miss->flow, UINT32_MAX);
3290 op->xout_garbage = false;
3291 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3292 op->subfacet = subfacet;
3293 put->flags = DPIF_FP_CREATE;
3294 put->key = miss->key;
3295 put->key_len = miss->key_len;
3296 put->mask = op->mask.data;
3297 put->mask_len = op->mask.size;
3299 if (want_path == SF_FAST_PATH) {
3300 put->actions = facet->xout.odp_actions.data;
3301 put->actions_len = facet->xout.odp_actions.size;
3303 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3304 op->slow_stub, sizeof op->slow_stub,
3305 &put->actions, &put->actions_len);
3311 /* Handles flow miss 'miss'. May add any required datapath operations
3312 * to 'ops', incrementing '*n_ops' for each new op. */
3314 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3317 struct facet *facet;
3319 miss->ofproto->n_missed += miss->stats.n_packets;
3321 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3323 /* There does not exist a bijection between 'struct flow' and datapath
3324 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3325 * assumption used throughout the facet and subfacet handling code.
3326 * Since we have to handle these misses in userspace anyway, we simply
3327 * skip facet creation, avoiding the problem altogether. */
3328 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3329 || !flow_miss_should_make_facet(miss)) {
3333 facet = facet_create(miss);
3335 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3338 static struct drop_key *
3339 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3342 struct drop_key *drop_key;
3344 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3345 &backer->drop_keys) {
3346 if (drop_key->key_len == key_len
3347 && !memcmp(drop_key->key, key, key_len)) {
3355 drop_key_clear(struct dpif_backer *backer)
3357 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3358 struct drop_key *drop_key, *next;
3360 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3363 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3365 if (error && !VLOG_DROP_WARN(&rl)) {
3366 struct ds ds = DS_EMPTY_INITIALIZER;
3367 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3368 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3369 ovs_strerror(error), ds_cstr(&ds));
3373 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3374 drop_key_destroy(drop_key);
3377 udpif_drop_key_clear(backer->udpif);
3381 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3383 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3384 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3385 struct flow_miss *miss;
3388 /* Process each element in the to-do list, constructing the set of
3389 * operations to batch. */
3391 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3392 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3394 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3396 /* Execute batch. */
3397 for (i = 0; i < n_ops; i++) {
3398 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3400 dpif_operate(backer->dpif, dpif_ops, n_ops);
3402 for (i = 0; i < n_ops; i++) {
3403 if (dpif_ops[i]->error != 0
3404 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3405 && flow_miss_ops[i].subfacet) {
3406 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3408 COVERAGE_INC(subfacet_install_fail);
3410 /* Zero-out subfacet counters when installation failed, but
3411 * datapath reported hits. This should not happen and
3412 * indicates a bug, since if the datapath flow exists, we
3413 * should not be attempting to create a new subfacet. A
3414 * buggy datapath could trigger this, so just zero out the
3415 * counters and log an error. */
3416 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3417 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3418 "datapath reported hits");
3419 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3422 subfacet->path = SF_NOT_INSTALLED;
3428 handle_upcalls(struct dpif_backer *backer)
3430 struct flow_miss_batch *fmb;
3433 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3434 struct drop_key *drop_key = drop_key_next(backer->udpif);
3439 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3440 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3441 hash_bytes(drop_key->key, drop_key->key_len, 0));
3442 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3443 drop_key->key, drop_key->key_len,
3444 NULL, 0, NULL, 0, NULL);
3446 drop_key_destroy(drop_key);
3450 fmb = flow_miss_batch_next(backer->udpif);
3452 handle_flow_misses(backer, fmb);
3453 flow_miss_batch_destroy(fmb);
3457 /* Flow expiration. */
3459 static int subfacet_max_idle(const struct dpif_backer *);
3460 static void update_stats(struct dpif_backer *);
3461 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3462 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3464 /* This function is called periodically by run(). Its job is to collect
3465 * updates for the flows that have been installed into the datapath, most
3466 * importantly when they last were used, and then use that information to
3467 * expire flows that have not been used recently.
3469 * Returns the number of milliseconds after which it should be called again. */
3471 expire(struct dpif_backer *backer)
3473 struct ofproto_dpif *ofproto;
3477 /* Periodically clear out the drop keys in an effort to keep them
3478 * relatively few. */
3479 drop_key_clear(backer);
3481 /* Update stats for each flow in the backer. */
3482 update_stats(backer);
3484 n_subfacets = hmap_count(&backer->subfacets);
3486 struct subfacet *subfacet;
3487 long long int total, now;
3491 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3492 total += now - subfacet->created;
3494 backer->avg_subfacet_life += total / n_subfacets;
3496 backer->avg_subfacet_life /= 2;
3498 backer->avg_n_subfacet += n_subfacets;
3499 backer->avg_n_subfacet /= 2;
3501 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3503 max_idle = subfacet_max_idle(backer);
3504 expire_subfacets(backer, max_idle);
3506 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3507 struct rule *rule, *next_rule;
3509 if (ofproto->backer != backer) {
3513 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3515 ovs_mutex_lock(&ofproto_mutex);
3516 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3517 &ofproto->up.expirable) {
3518 rule_expire(rule_dpif_cast(rule));
3520 ovs_mutex_unlock(&ofproto_mutex);
3522 /* All outstanding data in existing flows has been accounted, so it's a
3523 * good time to do bond rebalancing. */
3524 if (ofproto->has_bonded_bundles) {
3525 struct ofbundle *bundle;
3527 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3529 bond_rebalance(bundle->bond);
3535 return MIN(max_idle, 1000);
3538 /* Updates flow table statistics given that the datapath just reported 'stats'
3539 * as 'subfacet''s statistics. */
3541 update_subfacet_stats(struct subfacet *subfacet,
3542 const struct dpif_flow_stats *stats)
3544 struct facet *facet = subfacet->facet;
3545 struct dpif_flow_stats diff;
3547 diff.tcp_flags = stats->tcp_flags;
3548 diff.used = stats->used;
3550 if (stats->n_packets >= subfacet->dp_packet_count) {
3551 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3553 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3557 if (stats->n_bytes >= subfacet->dp_byte_count) {
3558 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3560 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3564 facet->ofproto->n_hit += diff.n_packets;
3565 subfacet->dp_packet_count = stats->n_packets;
3566 subfacet->dp_byte_count = stats->n_bytes;
3567 subfacet_update_stats(subfacet, &diff);
3569 if (facet->accounted_bytes < facet->byte_count) {
3571 facet_account(facet);
3572 facet->accounted_bytes = facet->byte_count;
3576 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3577 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3579 delete_unexpected_flow(struct dpif_backer *backer,
3580 const struct nlattr *key, size_t key_len)
3582 if (!VLOG_DROP_WARN(&rl)) {
3586 odp_flow_key_format(key, key_len, &s);
3587 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3591 COVERAGE_INC(facet_unexpected);
3592 dpif_flow_del(backer->dpif, key, key_len, NULL);
3595 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3597 * This function also pushes statistics updates to rules which each facet
3598 * resubmits into. Generally these statistics will be accurate. However, if a
3599 * facet changes the rule it resubmits into at some time in between
3600 * update_stats() runs, it is possible that statistics accrued to the
3601 * old rule will be incorrectly attributed to the new rule. This could be
3602 * avoided by calling update_stats() whenever rules are created or
3603 * deleted. However, the performance impact of making so many calls to the
3604 * datapath do not justify the benefit of having perfectly accurate statistics.
3606 * In addition, this function maintains per ofproto flow hit counts. The patch
3607 * port is not treated specially. e.g. A packet ingress from br0 patched into
3608 * br1 will increase the hit count of br0 by 1, however, does not affect
3609 * the hit or miss counts of br1.
3612 update_stats(struct dpif_backer *backer)
3614 const struct dpif_flow_stats *stats;
3615 struct dpif_flow_dump dump;
3616 const struct nlattr *key, *mask;
3617 size_t key_len, mask_len;
3619 dpif_flow_dump_start(&dump, backer->dpif);
3620 while (dpif_flow_dump_next(&dump, &key, &key_len,
3621 &mask, &mask_len, NULL, NULL, &stats)) {
3622 struct subfacet *subfacet;
3625 key_hash = odp_flow_key_hash(key, key_len);
3626 subfacet = subfacet_find(backer, key, key_len, key_hash);
3627 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3629 update_subfacet_stats(subfacet, stats);
3633 /* Stats are updated per-packet. */
3636 case SF_NOT_INSTALLED:
3638 delete_unexpected_flow(backer, key, key_len);
3643 dpif_flow_dump_done(&dump);
3645 update_moving_averages(backer);
3648 /* Calculates and returns the number of milliseconds of idle time after which
3649 * subfacets should expire from the datapath. When a subfacet expires, we fold
3650 * its statistics into its facet, and when a facet's last subfacet expires, we
3651 * fold its statistic into its rule. */
3653 subfacet_max_idle(const struct dpif_backer *backer)
3656 * Idle time histogram.
3658 * Most of the time a switch has a relatively small number of subfacets.
3659 * When this is the case we might as well keep statistics for all of them
3660 * in userspace and to cache them in the kernel datapath for performance as
3663 * As the number of subfacets increases, the memory required to maintain
3664 * statistics about them in userspace and in the kernel becomes
3665 * significant. However, with a large number of subfacets it is likely
3666 * that only a few of them are "heavy hitters" that consume a large amount
3667 * of bandwidth. At this point, only heavy hitters are worth caching in
3668 * the kernel and maintaining in userspaces; other subfacets we can
3671 * The technique used to compute the idle time is to build a histogram with
3672 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3673 * that is installed in the kernel gets dropped in the appropriate bucket.
3674 * After the histogram has been built, we compute the cutoff so that only
3675 * the most-recently-used 1% of subfacets (but at least
3676 * flow_eviction_threshold flows) are kept cached. At least
3677 * the most-recently-used bucket of subfacets is kept, so actually an
3678 * arbitrary number of subfacets can be kept in any given expiration run
3679 * (though the next run will delete most of those unless they receive
3682 * This requires a second pass through the subfacets, in addition to the
3683 * pass made by update_stats(), because the former function never looks at
3684 * uninstallable subfacets.
3686 enum { BUCKET_WIDTH = 100 };
3687 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3688 int buckets[N_BUCKETS] = { 0 };
3689 int total, subtotal, bucket;
3690 struct subfacet *subfacet;
3694 total = hmap_count(&backer->subfacets);
3695 if (total <= flow_eviction_threshold) {
3696 return N_BUCKETS * BUCKET_WIDTH;
3699 /* Build histogram. */
3701 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3702 long long int idle = now - subfacet->used;
3703 int bucket = (idle <= 0 ? 0
3704 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3705 : (unsigned int) idle / BUCKET_WIDTH);
3709 /* Find the first bucket whose flows should be expired. */
3710 subtotal = bucket = 0;
3712 subtotal += buckets[bucket++];
3713 } while (bucket < N_BUCKETS &&
3714 subtotal < MAX(flow_eviction_threshold, total / 100));
3716 if (VLOG_IS_DBG_ENABLED()) {
3720 ds_put_cstr(&s, "keep");
3721 for (i = 0; i < N_BUCKETS; i++) {
3723 ds_put_cstr(&s, ", drop");
3726 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3729 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3733 return bucket * BUCKET_WIDTH;
3737 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3739 /* Cutoff time for most flows. */
3740 long long int normal_cutoff = time_msec() - dp_max_idle;
3742 /* We really want to keep flows for special protocols around, so use a more
3743 * conservative cutoff. */
3744 long long int special_cutoff = time_msec() - 10000;
3746 struct subfacet *subfacet, *next_subfacet;
3747 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3751 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3752 &backer->subfacets) {
3753 long long int cutoff;
3755 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3759 if (subfacet->used < cutoff) {
3760 if (subfacet->path != SF_NOT_INSTALLED) {
3761 batch[n_batch++] = subfacet;
3762 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3763 subfacet_destroy_batch(backer, batch, n_batch);
3767 subfacet_destroy(subfacet);
3773 subfacet_destroy_batch(backer, batch, n_batch);
3777 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3778 * then delete it entirely. */
3780 rule_expire(struct rule_dpif *rule)
3781 OVS_REQUIRES(ofproto_mutex)
3783 uint16_t idle_timeout, hard_timeout;
3784 long long int now = time_msec();
3787 ovs_assert(!rule->up.pending);
3789 /* Has 'rule' expired? */
3790 ovs_mutex_lock(&rule->up.mutex);
3791 hard_timeout = rule->up.hard_timeout;
3792 idle_timeout = rule->up.idle_timeout;
3793 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3794 reason = OFPRR_HARD_TIMEOUT;
3795 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3796 reason = OFPRR_IDLE_TIMEOUT;
3800 ovs_mutex_unlock(&rule->up.mutex);
3803 COVERAGE_INC(ofproto_dpif_expired);
3804 ofproto_rule_expire(&rule->up, reason);
3810 /* Creates and returns a new facet based on 'miss'.
3812 * The caller must already have determined that no facet with an identical
3813 * 'miss->flow' exists in 'miss->ofproto'.
3815 * 'rule' and 'xout' must have been created based on 'miss'.
3817 * 'facet'' statistics are initialized based on 'stats'.
3819 * The facet will initially have no subfacets. The caller should create (at
3820 * least) one subfacet with subfacet_create(). */
3821 static struct facet *
3822 facet_create(const struct flow_miss *miss)
3824 struct ofproto_dpif *ofproto = miss->ofproto;
3825 struct facet *facet;
3828 facet = xzalloc(sizeof *facet);
3829 facet->ofproto = miss->ofproto;
3830 facet->used = miss->stats.used;
3831 facet->flow = miss->flow;
3832 facet->learn_rl = time_msec() + 500;
3834 list_init(&facet->subfacets);
3835 netflow_flow_init(&facet->nf_flow);
3836 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3838 xlate_out_copy(&facet->xout, &miss->xout);
3840 match_init(&match, &facet->flow, &facet->xout.wc);
3841 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3842 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3843 classifier_insert(&ofproto->facets, &facet->cr);
3844 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3846 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3851 facet_free(struct facet *facet)
3854 xlate_out_uninit(&facet->xout);
3859 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3860 * 'packet', which arrived on 'in_port'. */
3862 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3863 const struct nlattr *odp_actions, size_t actions_len,
3864 struct ofpbuf *packet)
3866 struct odputil_keybuf keybuf;
3870 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3871 odp_flow_key_from_flow(&key, flow,
3872 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3874 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3875 odp_actions, actions_len, packet);
3879 /* Remove 'facet' from its ofproto and free up the associated memory:
3881 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3882 * rule's statistics, via subfacet_uninstall().
3884 * - Removes 'facet' from its rule and from ofproto->facets.
3887 facet_remove(struct facet *facet)
3889 struct subfacet *subfacet, *next_subfacet;
3891 ovs_assert(!list_is_empty(&facet->subfacets));
3893 /* First uninstall all of the subfacets to get final statistics. */
3894 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3895 subfacet_uninstall(subfacet);
3898 /* Flush the final stats to the rule.
3900 * This might require us to have at least one subfacet around so that we
3901 * can use its actions for accounting in facet_account(), which is why we
3902 * have uninstalled but not yet destroyed the subfacets. */
3903 facet_flush_stats(facet);
3905 /* Now we're really all done so destroy everything. */
3906 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3907 &facet->subfacets) {
3908 subfacet_destroy__(subfacet);
3910 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3911 classifier_remove(&facet->ofproto->facets, &facet->cr);
3912 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3913 cls_rule_destroy(&facet->cr);
3917 /* Feed information from 'facet' back into the learning table to keep it in
3918 * sync with what is actually flowing through the datapath. */
3920 facet_learn(struct facet *facet)
3922 long long int now = time_msec();
3924 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3928 facet->learn_rl = now + 500;
3930 if (!facet->xout.has_learn
3931 && !facet->xout.has_normal
3932 && (!facet->xout.has_fin_timeout
3933 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3937 facet_push_stats(facet, true);
3941 facet_account(struct facet *facet)
3943 const struct nlattr *a;
3948 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3951 n_bytes = facet->byte_count - facet->accounted_bytes;
3953 /* This loop feeds byte counters to bond_account() for rebalancing to use
3954 * as a basis. We also need to track the actual VLAN on which the packet
3955 * is going to be sent to ensure that it matches the one passed to
3956 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3959 * We use the actions from an arbitrary subfacet because they should all
3960 * be equally valid for our purpose. */
3961 vlan_tci = facet->flow.vlan_tci;
3962 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3963 facet->xout.odp_actions.size) {
3964 const struct ovs_action_push_vlan *vlan;
3965 struct ofport_dpif *port;
3967 switch (nl_attr_type(a)) {
3968 case OVS_ACTION_ATTR_OUTPUT:
3969 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3970 if (port && port->bundle && port->bundle->bond) {
3971 bond_account(port->bundle->bond, &facet->flow,
3972 vlan_tci_to_vid(vlan_tci), n_bytes);
3976 case OVS_ACTION_ATTR_POP_VLAN:
3977 vlan_tci = htons(0);
3980 case OVS_ACTION_ATTR_PUSH_VLAN:
3981 vlan = nl_attr_get(a);
3982 vlan_tci = vlan->vlan_tci;
3988 /* Returns true if the only action for 'facet' is to send to the controller.
3989 * (We don't report NetFlow expiration messages for such facets because they
3990 * are just part of the control logic for the network, not real traffic). */
3992 facet_is_controller_flow(struct facet *facet)
3995 struct ofproto_dpif *ofproto = facet->ofproto;
3996 const struct ofpact *ofpacts;
3997 struct rule_actions *actions;
3998 struct rule_dpif *rule;
4002 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4003 actions = rule_dpif_get_actions(rule);
4004 rule_dpif_unref(rule);
4006 ofpacts_len = actions->ofpacts_len;
4007 ofpacts = actions->ofpacts;
4008 is_controller = ofpacts_len > 0
4009 && ofpacts->type == OFPACT_CONTROLLER
4010 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4011 rule_actions_unref(actions);
4013 return is_controller;
4018 /* Folds all of 'facet''s statistics into its rule. Also updates the
4019 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4020 * 'facet''s statistics in the datapath should have been zeroed and folded into
4021 * its packet and byte counts before this function is called. */
4023 facet_flush_stats(struct facet *facet)
4025 struct ofproto_dpif *ofproto = facet->ofproto;
4026 struct subfacet *subfacet;
4028 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4029 ovs_assert(!subfacet->dp_byte_count);
4030 ovs_assert(!subfacet->dp_packet_count);
4033 facet_push_stats(facet, false);
4034 if (facet->accounted_bytes < facet->byte_count) {
4035 facet_account(facet);
4036 facet->accounted_bytes = facet->byte_count;
4039 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4040 struct ofexpired expired;
4041 expired.flow = facet->flow;
4042 expired.packet_count = facet->packet_count;
4043 expired.byte_count = facet->byte_count;
4044 expired.used = facet->used;
4045 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4048 /* Reset counters to prevent double counting if 'facet' ever gets
4050 facet_reset_counters(facet);
4052 netflow_flow_clear(&facet->nf_flow);
4053 facet->tcp_flags = 0;
4056 /* Searches 'ofproto''s table of facets for one which would be responsible for
4057 * 'flow'. Returns it if found, otherwise a null pointer.
4059 * The returned facet might need revalidation; use facet_lookup_valid()
4060 * instead if that is important. */
4061 static struct facet *
4062 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4064 struct cls_rule *cr;
4066 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4067 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4068 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4069 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4072 /* Searches 'ofproto''s table of facets for one capable that covers
4073 * 'flow'. Returns it if found, otherwise a null pointer.
4075 * The returned facet is guaranteed to be valid. */
4076 static struct facet *
4077 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4079 struct facet *facet;
4081 facet = facet_find(ofproto, flow);
4083 && ofproto->backer->need_revalidate
4084 && !facet_revalidate(facet)) {
4092 facet_check_consistency(struct facet *facet)
4094 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4096 struct xlate_out xout;
4097 struct xlate_in xin;
4099 struct rule_dpif *rule;
4102 /* Check the datapath actions for consistency. */
4103 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4104 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4105 xlate_actions(&xin, &xout);
4106 rule_dpif_unref(rule);
4108 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4109 && facet->xout.slow == xout.slow;
4110 if (!ok && !VLOG_DROP_WARN(&rl)) {
4111 struct ds s = DS_EMPTY_INITIALIZER;
4113 flow_format(&s, &facet->flow);
4114 ds_put_cstr(&s, ": inconsistency in facet");
4116 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4117 ds_put_cstr(&s, " (actions were: ");
4118 format_odp_actions(&s, facet->xout.odp_actions.data,
4119 facet->xout.odp_actions.size);
4120 ds_put_cstr(&s, ") (correct actions: ");
4121 format_odp_actions(&s, xout.odp_actions.data,
4122 xout.odp_actions.size);
4123 ds_put_char(&s, ')');
4126 if (facet->xout.slow != xout.slow) {
4127 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4132 xlate_out_uninit(&xout);
4137 /* Re-searches the classifier for 'facet':
4139 * - If the rule found is different from 'facet''s current rule, moves
4140 * 'facet' to the new rule and recompiles its actions.
4142 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4143 * where it is and recompiles its actions anyway.
4145 * - If any of 'facet''s subfacets correspond to a new flow according to
4146 * xlate_receive(), 'facet' is removed.
4148 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4150 facet_revalidate(struct facet *facet)
4152 struct ofproto_dpif *ofproto = facet->ofproto;
4153 struct rule_dpif *new_rule;
4154 struct subfacet *subfacet;
4155 struct flow_wildcards wc;
4156 struct xlate_out xout;
4157 struct xlate_in xin;
4159 COVERAGE_INC(facet_revalidate);
4161 /* Check that child subfacets still correspond to this facet. Tunnel
4162 * configuration changes could cause a subfacet's OpenFlow in_port to
4164 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4165 struct ofproto_dpif *recv_ofproto;
4166 struct flow recv_flow;
4169 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4170 subfacet->key_len, &recv_flow, NULL,
4171 &recv_ofproto, NULL);
4173 || recv_ofproto != ofproto
4174 || facet != facet_find(ofproto, &recv_flow)) {
4175 facet_remove(facet);
4180 flow_wildcards_init_catchall(&wc);
4181 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4183 /* Calculate new datapath actions.
4185 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4186 * emit a NetFlow expiration and, if so, we need to have the old state
4187 * around to properly compose it. */
4188 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4189 xlate_actions(&xin, &xout);
4190 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4192 /* A facet's slow path reason should only change under dramatic
4193 * circumstances. Rather than try to update everything, it's simpler to
4194 * remove the facet and start over.
4196 * More importantly, if a facet's wildcards change, it will be relatively
4197 * difficult to figure out if its subfacets still belong to it, and if not
4198 * which facet they may belong to. Again, to avoid the complexity, we
4199 * simply give up instead. */
4200 if (facet->xout.slow != xout.slow
4201 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4202 facet_remove(facet);
4203 xlate_out_uninit(&xout);
4204 rule_dpif_unref(new_rule);
4208 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4209 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4210 if (subfacet->path == SF_FAST_PATH) {
4211 struct dpif_flow_stats stats;
4213 subfacet_install(subfacet, &xout.odp_actions, &stats);
4214 subfacet_update_stats(subfacet, &stats);
4218 facet_flush_stats(facet);
4220 ofpbuf_clear(&facet->xout.odp_actions);
4221 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4222 xout.odp_actions.size);
4225 /* Update 'facet' now that we've taken care of all the old state. */
4226 facet->xout.slow = xout.slow;
4227 facet->xout.has_learn = xout.has_learn;
4228 facet->xout.has_normal = xout.has_normal;
4229 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4230 facet->xout.nf_output_iface = xout.nf_output_iface;
4231 facet->xout.mirrors = xout.mirrors;
4232 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4234 ovs_mutex_lock(&new_rule->up.mutex);
4235 facet->used = MAX(facet->used, new_rule->up.created);
4236 ovs_mutex_unlock(&new_rule->up.mutex);
4238 xlate_out_uninit(&xout);
4239 rule_dpif_unref(new_rule);
4244 facet_reset_counters(struct facet *facet)
4246 facet->packet_count = 0;
4247 facet->byte_count = 0;
4248 facet->prev_packet_count = 0;
4249 facet->prev_byte_count = 0;
4250 facet->accounted_bytes = 0;
4254 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4255 struct dpif_flow_stats *stats, bool may_learn)
4257 struct ofport_dpif *in_port;
4258 struct rule_dpif *rule;
4259 struct xlate_in xin;
4261 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4262 if (in_port && in_port->is_tunnel) {
4263 netdev_vport_inc_rx(in_port->up.netdev, stats);
4266 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4267 rule_dpif_credit_stats(rule, stats);
4268 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4269 xin.resubmit_stats = stats;
4270 xin.may_learn = may_learn;
4271 xlate_actions_for_side_effects(&xin);
4272 rule_dpif_unref(rule);
4276 facet_push_stats(struct facet *facet, bool may_learn)
4278 struct dpif_flow_stats stats;
4280 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4281 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4282 ovs_assert(facet->used >= facet->prev_used);
4284 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4285 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4286 stats.used = facet->used;
4287 stats.tcp_flags = facet->tcp_flags;
4289 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4290 facet->prev_packet_count = facet->packet_count;
4291 facet->prev_byte_count = facet->byte_count;
4292 facet->prev_used = facet->used;
4294 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4296 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4297 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4298 stats.n_packets, stats.n_bytes);
4299 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4304 push_all_stats__(bool run_fast)
4306 static long long int rl = LLONG_MIN;
4307 struct ofproto_dpif *ofproto;
4309 if (time_msec() < rl) {
4313 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4314 struct cls_cursor cursor;
4315 struct facet *facet;
4317 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4318 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4319 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4320 facet_push_stats(facet, false);
4325 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4328 rl = time_msec() + 100;
4332 push_all_stats(void)
4334 push_all_stats__(true);
4338 rule_dpif_credit_stats(struct rule_dpif *rule,
4339 const struct dpif_flow_stats *stats)
4341 ovs_mutex_lock(&rule->stats_mutex);
4342 rule->packet_count += stats->n_packets;
4343 rule->byte_count += stats->n_bytes;
4344 rule->up.used = MAX(rule->up.used, stats->used);
4345 ovs_mutex_unlock(&rule->stats_mutex);
4349 rule_dpif_fail_open(const struct rule_dpif *rule)
4351 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4355 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4356 OVS_REQUIRES(rule->up.mutex)
4358 return rule->up.flow_cookie;
4362 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4363 uint16_t hard_timeout)
4365 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4368 /* Returns 'rule''s actions. The caller owns a reference on the returned
4369 * actions and must eventually release it (with rule_actions_unref()) to avoid
4371 struct rule_actions *
4372 rule_dpif_get_actions(const struct rule_dpif *rule)
4374 return rule_get_actions(&rule->up);
4379 static struct subfacet *
4380 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4381 size_t key_len, uint32_t key_hash)
4383 struct subfacet *subfacet;
4385 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4386 &backer->subfacets) {
4387 if (subfacet->key_len == key_len
4388 && !memcmp(key, subfacet->key, key_len)) {
4396 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4397 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4398 * existing subfacet if there is one, otherwise creates and returns a
4400 static struct subfacet *
4401 subfacet_create(struct facet *facet, struct flow_miss *miss)
4403 struct dpif_backer *backer = miss->ofproto->backer;
4404 const struct nlattr *key = miss->key;
4405 size_t key_len = miss->key_len;
4407 struct subfacet *subfacet;
4409 key_hash = odp_flow_key_hash(key, key_len);
4411 if (list_is_empty(&facet->subfacets)) {
4412 subfacet = &facet->one_subfacet;
4414 subfacet = subfacet_find(backer, key, key_len, key_hash);
4416 if (subfacet->facet == facet) {
4420 /* This shouldn't happen. */
4421 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4422 subfacet_destroy(subfacet);
4425 subfacet = xmalloc(sizeof *subfacet);
4428 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4429 list_push_back(&facet->subfacets, &subfacet->list_node);
4430 subfacet->facet = facet;
4431 subfacet->key = xmemdup(key, key_len);
4432 subfacet->key_len = key_len;
4433 subfacet->used = miss->stats.used;
4434 subfacet->created = subfacet->used;
4435 subfacet->dp_packet_count = 0;
4436 subfacet->dp_byte_count = 0;
4437 subfacet->path = SF_NOT_INSTALLED;
4438 subfacet->backer = backer;
4440 backer->subfacet_add_count++;
4444 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4445 * its facet within 'ofproto', and frees it. */
4447 subfacet_destroy__(struct subfacet *subfacet)
4449 struct facet *facet = subfacet->facet;
4450 struct ofproto_dpif *ofproto = facet->ofproto;
4452 /* Update ofproto stats before uninstall the subfacet. */
4453 ofproto->backer->subfacet_del_count++;
4455 subfacet_uninstall(subfacet);
4456 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4457 list_remove(&subfacet->list_node);
4458 free(subfacet->key);
4459 if (subfacet != &facet->one_subfacet) {
4464 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4465 * last remaining subfacet in its facet destroys the facet too. */
4467 subfacet_destroy(struct subfacet *subfacet)
4469 struct facet *facet = subfacet->facet;
4471 if (list_is_singleton(&facet->subfacets)) {
4472 /* facet_remove() needs at least one subfacet (it will remove it). */
4473 facet_remove(facet);
4475 subfacet_destroy__(subfacet);
4480 subfacet_destroy_batch(struct dpif_backer *backer,
4481 struct subfacet **subfacets, int n)
4483 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4484 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4485 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4488 for (i = 0; i < n; i++) {
4489 ops[i].type = DPIF_OP_FLOW_DEL;
4490 ops[i].u.flow_del.key = subfacets[i]->key;
4491 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4492 ops[i].u.flow_del.stats = &stats[i];
4496 dpif_operate(backer->dpif, opsp, n);
4497 for (i = 0; i < n; i++) {
4498 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4499 subfacets[i]->path = SF_NOT_INSTALLED;
4500 subfacet_destroy(subfacets[i]);
4505 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4506 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4507 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4508 * since 'subfacet' was last updated.
4510 * Returns 0 if successful, otherwise a positive errno value. */
4512 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4513 struct dpif_flow_stats *stats)
4515 struct facet *facet = subfacet->facet;
4516 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4517 const struct nlattr *actions = odp_actions->data;
4518 size_t actions_len = odp_actions->size;
4519 struct odputil_keybuf maskbuf;
4522 uint64_t slow_path_stub[128 / 8];
4523 enum dpif_flow_put_flags flags;
4526 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4529 flags |= DPIF_FP_ZERO_STATS;
4532 if (path == SF_SLOW_PATH) {
4533 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4534 slow_path_stub, sizeof slow_path_stub,
4535 &actions, &actions_len);
4538 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4539 if (enable_megaflows) {
4540 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4541 &facet->flow, UINT32_MAX);
4544 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4545 subfacet->key_len, mask.data, mask.size,
4546 actions, actions_len, stats);
4549 subfacet_reset_dp_stats(subfacet, stats);
4553 COVERAGE_INC(subfacet_install_fail);
4555 subfacet->path = path;
4560 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4562 subfacet_uninstall(struct subfacet *subfacet)
4564 if (subfacet->path != SF_NOT_INSTALLED) {
4565 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4566 struct dpif_flow_stats stats;
4569 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4570 subfacet->key_len, &stats);
4571 subfacet_reset_dp_stats(subfacet, &stats);
4573 subfacet_update_stats(subfacet, &stats);
4575 subfacet->path = SF_NOT_INSTALLED;
4577 ovs_assert(subfacet->dp_packet_count == 0);
4578 ovs_assert(subfacet->dp_byte_count == 0);
4582 /* Resets 'subfacet''s datapath statistics counters. This should be called
4583 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4584 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4585 * was reset in the datapath. 'stats' will be modified to include only
4586 * statistics new since 'subfacet' was last updated. */
4588 subfacet_reset_dp_stats(struct subfacet *subfacet,
4589 struct dpif_flow_stats *stats)
4592 && subfacet->dp_packet_count <= stats->n_packets
4593 && subfacet->dp_byte_count <= stats->n_bytes) {
4594 stats->n_packets -= subfacet->dp_packet_count;
4595 stats->n_bytes -= subfacet->dp_byte_count;
4598 subfacet->dp_packet_count = 0;
4599 subfacet->dp_byte_count = 0;
4602 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4604 * Because of the meaning of a subfacet's counters, it only makes sense to do
4605 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4606 * represents a packet that was sent by hand or if it represents statistics
4607 * that have been cleared out of the datapath. */
4609 subfacet_update_stats(struct subfacet *subfacet,
4610 const struct dpif_flow_stats *stats)
4612 if (stats->n_packets || stats->used > subfacet->used) {
4613 struct facet *facet = subfacet->facet;
4615 subfacet->used = MAX(subfacet->used, stats->used);
4616 facet->used = MAX(facet->used, stats->used);
4617 facet->packet_count += stats->n_packets;
4618 facet->byte_count += stats->n_bytes;
4619 facet->tcp_flags |= stats->tcp_flags;
4625 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4626 * the fields that were relevant as part of the lookup. */
4628 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4629 struct flow_wildcards *wc, struct rule_dpif **rule)
4631 struct ofport_dpif *port;
4633 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4636 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4638 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4639 flow->in_port.ofp_port);
4642 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4643 ofproto->no_packet_in_rule, rule);
4647 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4648 const struct flow *flow, struct flow_wildcards *wc,
4649 uint8_t table_id, struct rule_dpif **rule)
4651 const struct cls_rule *cls_rule;
4652 struct classifier *cls;
4656 if (table_id >= N_TABLES) {
4661 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4662 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4665 cls = &ofproto->up.tables[table_id].cls;
4666 ovs_rwlock_rdlock(&cls->rwlock);
4667 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4668 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4669 /* We must pretend that transport ports are unavailable. */
4670 struct flow ofpc_normal_flow = *flow;
4671 ofpc_normal_flow.tp_src = htons(0);
4672 ofpc_normal_flow.tp_dst = htons(0);
4673 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4674 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4675 cls_rule = &ofproto->drop_frags_rule->up.cr;
4677 flow_wildcards_init_exact(wc);
4680 cls_rule = classifier_lookup(cls, flow, wc);
4683 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4684 rule_dpif_ref(*rule);
4685 ovs_rwlock_unlock(&cls->rwlock);
4687 return *rule != NULL;
4690 /* Given a port configuration (specified as zero if there's no port), chooses
4691 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4692 * flow table miss. */
4694 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4695 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4697 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4698 rule_dpif_ref(*rule);
4702 rule_dpif_ref(struct rule_dpif *rule)
4705 ofproto_rule_ref(&rule->up);
4710 rule_dpif_unref(struct rule_dpif *rule)
4713 ofproto_rule_unref(&rule->up);
4718 complete_operation(struct rule_dpif *rule)
4719 OVS_REQUIRES(ofproto_mutex)
4721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4723 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4724 ofoperation_complete(rule->up.pending, 0);
4727 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4729 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4732 static struct rule *
4735 struct rule_dpif *rule = xmalloc(sizeof *rule);
4740 rule_dealloc(struct rule *rule_)
4742 struct rule_dpif *rule = rule_dpif_cast(rule_);
4747 rule_construct(struct rule *rule_)
4749 struct rule_dpif *rule = rule_dpif_cast(rule_);
4750 ovs_mutex_init(&rule->stats_mutex);
4751 ovs_mutex_lock(&rule->stats_mutex);
4752 rule->packet_count = 0;
4753 rule->byte_count = 0;
4754 ovs_mutex_unlock(&rule->stats_mutex);
4759 rule_insert(struct rule *rule_)
4760 OVS_REQUIRES(ofproto_mutex)
4762 struct rule_dpif *rule = rule_dpif_cast(rule_);
4763 complete_operation(rule);
4767 rule_delete(struct rule *rule_)
4768 OVS_REQUIRES(ofproto_mutex)
4770 struct rule_dpif *rule = rule_dpif_cast(rule_);
4771 complete_operation(rule);
4775 rule_destruct(struct rule *rule_)
4777 struct rule_dpif *rule = rule_dpif_cast(rule_);
4778 ovs_mutex_destroy(&rule->stats_mutex);
4782 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4784 struct rule_dpif *rule = rule_dpif_cast(rule_);
4786 /* push_all_stats() can handle flow misses which, when using the learn
4787 * action, can cause rules to be added and deleted. This can corrupt our
4788 * caller's datastructures which assume that rule_get_stats() doesn't have
4789 * an impact on the flow table. To be safe, we disable miss handling. */
4790 push_all_stats__(false);
4792 /* Start from historical data for 'rule' itself that are no longer tracked
4793 * in facets. This counts, for example, facets that have expired. */
4794 ovs_mutex_lock(&rule->stats_mutex);
4795 *packets = rule->packet_count;
4796 *bytes = rule->byte_count;
4797 ovs_mutex_unlock(&rule->stats_mutex);
4801 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4802 struct ofpbuf *packet)
4804 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4805 struct dpif_flow_stats stats;
4806 struct xlate_out xout;
4807 struct xlate_in xin;
4809 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4810 rule_dpif_credit_stats(rule, &stats);
4812 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4813 xin.resubmit_stats = &stats;
4814 xlate_actions(&xin, &xout);
4816 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4817 xout.odp_actions.size, packet);
4819 xlate_out_uninit(&xout);
4823 rule_execute(struct rule *rule, const struct flow *flow,
4824 struct ofpbuf *packet)
4826 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4827 ofpbuf_delete(packet);
4832 rule_modify_actions(struct rule *rule_, bool reset_counters)
4833 OVS_REQUIRES(ofproto_mutex)
4835 struct rule_dpif *rule = rule_dpif_cast(rule_);
4837 if (reset_counters) {
4838 ovs_mutex_lock(&rule->stats_mutex);
4839 rule->packet_count = 0;
4840 rule->byte_count = 0;
4841 ovs_mutex_unlock(&rule->stats_mutex);
4844 complete_operation(rule);
4847 /* Sends 'packet' out 'ofport'.
4848 * May modify 'packet'.
4849 * Returns 0 if successful, otherwise a positive errno value. */
4851 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4854 uint64_t odp_actions_stub[1024 / 8];
4855 struct ofpbuf key, odp_actions;
4856 struct dpif_flow_stats stats;
4857 struct odputil_keybuf keybuf;
4858 struct ofpact_output output;
4859 struct xlate_out xout;
4860 struct xlate_in xin;
4862 union flow_in_port in_port_;
4865 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4866 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4868 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4869 in_port_.ofp_port = OFPP_NONE;
4870 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4871 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4873 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4875 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4876 output.port = ofport->up.ofp_port;
4879 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4880 xin.ofpacts_len = sizeof output;
4881 xin.ofpacts = &output.ofpact;
4882 xin.resubmit_stats = &stats;
4883 xlate_actions(&xin, &xout);
4885 error = dpif_execute(ofproto->backer->dpif,
4887 xout.odp_actions.data, xout.odp_actions.size,
4889 xlate_out_uninit(&xout);
4892 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
4893 ofproto->up.name, netdev_get_name(ofport->up.netdev),
4894 ovs_strerror(error));
4897 ofproto->stats.tx_packets++;
4898 ofproto->stats.tx_bytes += packet->size;
4902 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4903 * The action will state 'slow' as the reason that the action is in the slow
4904 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4905 * dump-flows" output to see why a flow is in the slow path.)
4907 * The 'stub_size' bytes in 'stub' will be used to store the action.
4908 * 'stub_size' must be large enough for the action.
4910 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4913 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4914 enum slow_path_reason slow,
4915 uint64_t *stub, size_t stub_size,
4916 const struct nlattr **actionsp, size_t *actions_lenp)
4918 union user_action_cookie cookie;
4921 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4922 cookie.slow_path.unused = 0;
4923 cookie.slow_path.reason = slow;
4925 ofpbuf_use_stack(&buf, stub, stub_size);
4926 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4927 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4929 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4931 odp_port_t odp_port;
4934 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4935 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4936 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4938 *actionsp = buf.data;
4939 *actions_lenp = buf.size;
4943 set_frag_handling(struct ofproto *ofproto_,
4944 enum ofp_config_flags frag_handling)
4946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4947 if (frag_handling != OFPC_FRAG_REASM) {
4948 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4956 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4957 const struct flow *flow,
4958 const struct ofpact *ofpacts, size_t ofpacts_len)
4960 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4961 struct odputil_keybuf keybuf;
4962 struct dpif_flow_stats stats;
4963 struct xlate_out xout;
4964 struct xlate_in xin;
4968 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4969 odp_flow_key_from_flow(&key, flow,
4970 ofp_port_to_odp_port(ofproto,
4971 flow->in_port.ofp_port));
4973 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4975 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
4976 xin.resubmit_stats = &stats;
4977 xin.ofpacts_len = ofpacts_len;
4978 xin.ofpacts = ofpacts;
4980 xlate_actions(&xin, &xout);
4981 dpif_execute(ofproto->backer->dpif, key.data, key.size,
4982 xout.odp_actions.data, xout.odp_actions.size, packet);
4983 xlate_out_uninit(&xout);
4991 set_netflow(struct ofproto *ofproto_,
4992 const struct netflow_options *netflow_options)
4994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4996 if (netflow_options) {
4997 if (!ofproto->netflow) {
4998 ofproto->netflow = netflow_create();
4999 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5001 return netflow_set_options(ofproto->netflow, netflow_options);
5002 } else if (ofproto->netflow) {
5003 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5004 netflow_destroy(ofproto->netflow);
5005 ofproto->netflow = NULL;
5012 get_netflow_ids(const struct ofproto *ofproto_,
5013 uint8_t *engine_type, uint8_t *engine_id)
5015 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5017 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5021 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5023 if (!facet_is_controller_flow(facet) &&
5024 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5025 struct subfacet *subfacet;
5026 struct ofexpired expired;
5028 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5029 if (subfacet->path == SF_FAST_PATH) {
5030 struct dpif_flow_stats stats;
5032 subfacet_install(subfacet, &facet->xout.odp_actions,
5034 subfacet_update_stats(subfacet, &stats);
5038 expired.flow = facet->flow;
5039 expired.packet_count = facet->packet_count;
5040 expired.byte_count = facet->byte_count;
5041 expired.used = facet->used;
5042 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5047 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5049 struct cls_cursor cursor;
5050 struct facet *facet;
5052 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5053 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5054 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5055 send_active_timeout(ofproto, facet);
5057 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5060 static struct ofproto_dpif *
5061 ofproto_dpif_lookup(const char *name)
5063 struct ofproto_dpif *ofproto;
5065 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5066 hash_string(name, 0), &all_ofproto_dpifs) {
5067 if (!strcmp(ofproto->up.name, name)) {
5075 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5076 const char *argv[], void *aux OVS_UNUSED)
5078 struct ofproto_dpif *ofproto;
5081 ofproto = ofproto_dpif_lookup(argv[1]);
5083 unixctl_command_reply_error(conn, "no such bridge");
5086 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5087 mac_learning_flush(ofproto->ml);
5088 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5090 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5091 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5092 mac_learning_flush(ofproto->ml);
5093 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5097 unixctl_command_reply(conn, "table successfully flushed");
5100 static struct ofport_dpif *
5101 ofbundle_get_a_port(const struct ofbundle *bundle)
5103 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5108 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5109 const char *argv[], void *aux OVS_UNUSED)
5111 struct ds ds = DS_EMPTY_INITIALIZER;
5112 const struct ofproto_dpif *ofproto;
5113 const struct mac_entry *e;
5115 ofproto = ofproto_dpif_lookup(argv[1]);
5117 unixctl_command_reply_error(conn, "no such bridge");
5121 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5122 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5123 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5124 struct ofbundle *bundle = e->port.p;
5125 char name[OFP_MAX_PORT_NAME_LEN];
5127 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5129 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5130 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5131 mac_entry_age(ofproto->ml, e));
5133 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5134 unixctl_command_reply(conn, ds_cstr(&ds));
5139 struct xlate_out xout;
5140 struct xlate_in xin;
5146 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5148 struct rule_actions *actions;
5151 ds_put_char_multiple(result, '\t', level);
5153 ds_put_cstr(result, "No match\n");
5157 ovs_mutex_lock(&rule->up.mutex);
5158 cookie = rule->up.flow_cookie;
5159 ovs_mutex_unlock(&rule->up.mutex);
5161 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5162 rule ? rule->up.table_id : 0, ntohll(cookie));
5163 cls_rule_format(&rule->up.cr, result);
5164 ds_put_char(result, '\n');
5166 actions = rule_dpif_get_actions(rule);
5168 ds_put_char_multiple(result, '\t', level);
5169 ds_put_cstr(result, "OpenFlow ");
5170 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5171 ds_put_char(result, '\n');
5173 rule_actions_unref(actions);
5177 trace_format_flow(struct ds *result, int level, const char *title,
5178 struct trace_ctx *trace)
5180 ds_put_char_multiple(result, '\t', level);
5181 ds_put_format(result, "%s: ", title);
5182 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5183 ds_put_cstr(result, "unchanged");
5185 flow_format(result, &trace->xin.flow);
5186 trace->flow = trace->xin.flow;
5188 ds_put_char(result, '\n');
5192 trace_format_regs(struct ds *result, int level, const char *title,
5193 struct trace_ctx *trace)
5197 ds_put_char_multiple(result, '\t', level);
5198 ds_put_format(result, "%s:", title);
5199 for (i = 0; i < FLOW_N_REGS; i++) {
5200 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5202 ds_put_char(result, '\n');
5206 trace_format_odp(struct ds *result, int level, const char *title,
5207 struct trace_ctx *trace)
5209 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5211 ds_put_char_multiple(result, '\t', level);
5212 ds_put_format(result, "%s: ", title);
5213 format_odp_actions(result, odp_actions->data, odp_actions->size);
5214 ds_put_char(result, '\n');
5218 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5220 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5221 struct ds *result = trace->result;
5223 ds_put_char(result, '\n');
5224 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5225 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5226 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5227 trace_format_rule(result, recurse + 1, rule);
5231 trace_report(struct xlate_in *xin, const char *s, int recurse)
5233 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5234 struct ds *result = trace->result;
5236 ds_put_char_multiple(result, '\t', recurse);
5237 ds_put_cstr(result, s);
5238 ds_put_char(result, '\n');
5242 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5243 void *aux OVS_UNUSED)
5245 const struct dpif_backer *backer;
5246 struct ofproto_dpif *ofproto;
5247 struct ofpbuf odp_key, odp_mask;
5248 struct ofpbuf *packet;
5256 ofpbuf_init(&odp_key, 0);
5257 ofpbuf_init(&odp_mask, 0);
5259 /* Handle "-generate" or a hex string as the last argument. */
5260 if (!strcmp(argv[argc - 1], "-generate")) {
5261 packet = ofpbuf_new(0);
5264 const char *error = eth_from_hex(argv[argc - 1], &packet);
5267 } else if (argc == 4) {
5268 /* The 3-argument form must end in "-generate' or a hex string. */
5269 unixctl_command_reply_error(conn, error);
5274 /* Parse the flow and determine whether a datapath or
5275 * bridge is specified. If function odp_flow_key_from_string()
5276 * returns 0, the flow is a odp_flow. If function
5277 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5278 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5279 /* If the odp_flow is the second argument,
5280 * the datapath name is the first argument. */
5282 const char *dp_type;
5283 if (!strncmp(argv[1], "ovs-", 4)) {
5284 dp_type = argv[1] + 4;
5288 backer = shash_find_data(&all_dpif_backers, dp_type);
5290 unixctl_command_reply_error(conn, "Cannot find datapath "
5295 /* No datapath name specified, so there should be only one
5297 struct shash_node *node;
5298 if (shash_count(&all_dpif_backers) != 1) {
5299 unixctl_command_reply_error(conn, "Must specify datapath "
5300 "name, there is more than one type of datapath");
5303 node = shash_first(&all_dpif_backers);
5304 backer = node->data;
5307 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5308 NULL, &ofproto, NULL)) {
5309 unixctl_command_reply_error(conn, "Invalid datapath flow");
5312 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5313 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5315 unixctl_command_reply_error(conn, "Must specify bridge name");
5319 ofproto = ofproto_dpif_lookup(argv[1]);
5321 unixctl_command_reply_error(conn, "Unknown bridge name");
5325 unixctl_command_reply_error(conn, "Bad flow syntax");
5329 /* Generate a packet, if requested. */
5331 if (!packet->size) {
5332 flow_compose(packet, &flow);
5334 union flow_in_port in_port_;
5336 in_port_ = flow.in_port;
5337 ds_put_cstr(&result, "Packet: ");
5338 s = ofp_packet_to_string(packet->data, packet->size);
5339 ds_put_cstr(&result, s);
5342 /* Use the metadata from the flow and the packet argument
5343 * to reconstruct the flow. */
5344 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5349 ofproto_trace(ofproto, &flow, packet, &result);
5350 unixctl_command_reply(conn, ds_cstr(&result));
5353 ds_destroy(&result);
5354 ofpbuf_delete(packet);
5355 ofpbuf_uninit(&odp_key);
5356 ofpbuf_uninit(&odp_mask);
5360 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5361 const struct ofpbuf *packet, struct ds *ds)
5363 struct rule_dpif *rule;
5364 struct flow_wildcards wc;
5366 ds_put_cstr(ds, "Flow: ");
5367 flow_format(ds, flow);
5368 ds_put_char(ds, '\n');
5370 flow_wildcards_init_catchall(&wc);
5371 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5373 trace_format_rule(ds, 0, rule);
5374 if (rule == ofproto->miss_rule) {
5375 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5376 } else if (rule == ofproto->no_packet_in_rule) {
5377 ds_put_cstr(ds, "\nNo match, packets dropped because "
5378 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5379 } else if (rule == ofproto->drop_frags_rule) {
5380 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5381 "and the fragment handling mode is \"drop\".\n");
5385 uint64_t odp_actions_stub[1024 / 8];
5386 struct ofpbuf odp_actions;
5387 struct trace_ctx trace;
5391 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5394 ofpbuf_use_stub(&odp_actions,
5395 odp_actions_stub, sizeof odp_actions_stub);
5396 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5397 trace.xin.resubmit_hook = trace_resubmit;
5398 trace.xin.report_hook = trace_report;
5400 xlate_actions(&trace.xin, &trace.xout);
5401 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5403 ds_put_char(ds, '\n');
5404 trace_format_flow(ds, 0, "Final flow", &trace);
5406 match_init(&match, flow, &trace.xout.wc);
5407 ds_put_cstr(ds, "Relevant fields: ");
5408 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5409 ds_put_char(ds, '\n');
5411 ds_put_cstr(ds, "Datapath actions: ");
5412 format_odp_actions(ds, trace.xout.odp_actions.data,
5413 trace.xout.odp_actions.size);
5415 if (trace.xout.slow) {
5416 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5417 "slow path because it:");
5418 switch (trace.xout.slow) {
5420 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5423 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5426 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5429 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5431 case SLOW_CONTROLLER:
5432 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5433 "to the OpenFlow controller.");
5440 xlate_out_uninit(&trace.xout);
5443 rule_dpif_unref(rule);
5446 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5447 * 'reply' describing the results. */
5449 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5451 struct cls_cursor cursor;
5452 struct facet *facet;
5456 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5457 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5458 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5459 if (!facet_check_consistency(facet)) {
5463 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5465 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5469 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5470 ofproto->up.name, errors);
5472 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5477 ofproto_dpif_self_check(struct unixctl_conn *conn,
5478 int argc, const char *argv[], void *aux OVS_UNUSED)
5480 struct ds reply = DS_EMPTY_INITIALIZER;
5481 struct ofproto_dpif *ofproto;
5484 ofproto = ofproto_dpif_lookup(argv[1]);
5486 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5487 "ofproto/list for help)");
5490 ofproto_dpif_self_check__(ofproto, &reply);
5492 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5493 ofproto_dpif_self_check__(ofproto, &reply);
5497 unixctl_command_reply(conn, ds_cstr(&reply));
5501 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5502 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5503 * to destroy 'ofproto_shash' and free the returned value. */
5504 static const struct shash_node **
5505 get_ofprotos(struct shash *ofproto_shash)
5507 const struct ofproto_dpif *ofproto;
5509 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5510 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5511 shash_add_nocopy(ofproto_shash, name, ofproto);
5514 return shash_sort(ofproto_shash);
5518 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5519 const char *argv[] OVS_UNUSED,
5520 void *aux OVS_UNUSED)
5522 struct ds ds = DS_EMPTY_INITIALIZER;
5523 struct shash ofproto_shash;
5524 const struct shash_node **sorted_ofprotos;
5527 shash_init(&ofproto_shash);
5528 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5529 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5530 const struct shash_node *node = sorted_ofprotos[i];
5531 ds_put_format(&ds, "%s\n", node->name);
5534 shash_destroy(&ofproto_shash);
5535 free(sorted_ofprotos);
5537 unixctl_command_reply(conn, ds_cstr(&ds));
5542 show_dp_rates(struct ds *ds, const char *heading,
5543 const struct avg_subfacet_rates *rates)
5545 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5546 heading, rates->add_rate, rates->del_rate);
5550 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5552 const struct shash_node **ofprotos;
5553 struct ofproto_dpif *ofproto;
5554 struct shash ofproto_shash;
5555 uint64_t n_hit, n_missed;
5556 long long int minutes;
5559 n_hit = n_missed = 0;
5560 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5561 if (ofproto->backer == backer) {
5562 n_missed += ofproto->n_missed;
5563 n_hit += ofproto->n_hit;
5567 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5568 dpif_name(backer->dpif), n_hit, n_missed);
5569 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5570 " life span: %lldms\n", hmap_count(&backer->subfacets),
5571 backer->avg_n_subfacet, backer->max_n_subfacet,
5572 backer->avg_subfacet_life);
5574 minutes = (time_msec() - backer->created) / (1000 * 60);
5575 if (minutes >= 60) {
5576 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5578 if (minutes >= 60 * 24) {
5579 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5581 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5583 shash_init(&ofproto_shash);
5584 ofprotos = get_ofprotos(&ofproto_shash);
5585 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5586 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5587 const struct shash_node **ports;
5590 if (ofproto->backer != backer) {
5594 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5595 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5597 ports = shash_sort(&ofproto->up.port_by_name);
5598 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5599 const struct shash_node *node = ports[j];
5600 struct ofport *ofport = node->data;
5602 odp_port_t odp_port;
5604 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5607 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5608 if (odp_port != ODPP_NONE) {
5609 ds_put_format(ds, "%"PRIu32":", odp_port);
5611 ds_put_cstr(ds, "none:");
5614 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5617 if (!netdev_get_config(ofport->netdev, &config)) {
5618 const struct smap_node **nodes;
5621 nodes = smap_sort(&config);
5622 for (i = 0; i < smap_count(&config); i++) {
5623 const struct smap_node *node = nodes[i];
5624 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5625 node->key, node->value);
5629 smap_destroy(&config);
5631 ds_put_char(ds, ')');
5632 ds_put_char(ds, '\n');
5636 shash_destroy(&ofproto_shash);
5641 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5642 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5644 struct ds ds = DS_EMPTY_INITIALIZER;
5645 const struct shash_node **backers;
5648 backers = shash_sort(&all_dpif_backers);
5649 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5650 dpif_show_backer(backers[i]->data, &ds);
5654 unixctl_command_reply(conn, ds_cstr(&ds));
5658 /* Dump the megaflow (facet) cache. This is useful to check the
5659 * correctness of flow wildcarding, since the same mechanism is used for
5660 * both xlate caching and kernel wildcarding.
5662 * It's important to note that in the output the flow description uses
5663 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5665 * This command is only needed for advanced debugging, so it's not
5666 * documented in the man page. */
5668 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5669 int argc OVS_UNUSED, const char *argv[],
5670 void *aux OVS_UNUSED)
5672 struct ds ds = DS_EMPTY_INITIALIZER;
5673 const struct ofproto_dpif *ofproto;
5674 long long int now = time_msec();
5675 struct cls_cursor cursor;
5676 struct facet *facet;
5678 ofproto = ofproto_dpif_lookup(argv[1]);
5680 unixctl_command_reply_error(conn, "no such bridge");
5684 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5685 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5686 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5687 cls_rule_format(&facet->cr, &ds);
5688 ds_put_cstr(&ds, ", ");
5689 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5690 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5691 ds_put_cstr(&ds, "Datapath actions: ");
5692 if (facet->xout.slow) {
5693 uint64_t slow_path_stub[128 / 8];
5694 const struct nlattr *actions;
5697 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5698 slow_path_stub, sizeof slow_path_stub,
5699 &actions, &actions_len);
5700 format_odp_actions(&ds, actions, actions_len);
5702 format_odp_actions(&ds, facet->xout.odp_actions.data,
5703 facet->xout.odp_actions.size);
5705 ds_put_cstr(&ds, "\n");
5707 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5709 ds_chomp(&ds, '\n');
5710 unixctl_command_reply(conn, ds_cstr(&ds));
5714 /* Disable using the megaflows.
5716 * This command is only needed for advanced debugging, so it's not
5717 * documented in the man page. */
5719 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5720 int argc OVS_UNUSED,
5721 const char *argv[] OVS_UNUSED,
5722 void *aux OVS_UNUSED)
5724 struct ofproto_dpif *ofproto;
5726 enable_megaflows = false;
5728 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5729 flush(&ofproto->up);
5732 unixctl_command_reply(conn, "megaflows disabled");
5735 /* Re-enable using megaflows.
5737 * This command is only needed for advanced debugging, so it's not
5738 * documented in the man page. */
5740 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5741 int argc OVS_UNUSED,
5742 const char *argv[] OVS_UNUSED,
5743 void *aux OVS_UNUSED)
5745 struct ofproto_dpif *ofproto;
5747 enable_megaflows = true;
5749 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5750 flush(&ofproto->up);
5753 unixctl_command_reply(conn, "megaflows enabled");
5757 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5758 int argc OVS_UNUSED, const char *argv[],
5759 void *aux OVS_UNUSED)
5761 struct ds ds = DS_EMPTY_INITIALIZER;
5762 const struct ofproto_dpif *ofproto;
5763 struct subfacet *subfacet;
5765 ofproto = ofproto_dpif_lookup(argv[1]);
5767 unixctl_command_reply_error(conn, "no such bridge");
5771 update_stats(ofproto->backer);
5773 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5774 struct facet *facet = subfacet->facet;
5775 struct odputil_keybuf maskbuf;
5778 if (facet->ofproto != ofproto) {
5782 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5783 if (enable_megaflows) {
5784 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5785 &facet->flow, UINT32_MAX);
5788 odp_flow_format(subfacet->key, subfacet->key_len,
5789 mask.data, mask.size, &ds, false);
5791 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5792 subfacet->dp_packet_count, subfacet->dp_byte_count);
5793 if (subfacet->used) {
5794 ds_put_format(&ds, "%.3fs",
5795 (time_msec() - subfacet->used) / 1000.0);
5797 ds_put_format(&ds, "never");
5799 if (subfacet->facet->tcp_flags) {
5800 ds_put_cstr(&ds, ", flags:");
5801 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5804 ds_put_cstr(&ds, ", actions:");
5805 if (facet->xout.slow) {
5806 uint64_t slow_path_stub[128 / 8];
5807 const struct nlattr *actions;
5810 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5811 slow_path_stub, sizeof slow_path_stub,
5812 &actions, &actions_len);
5813 format_odp_actions(&ds, actions, actions_len);
5815 format_odp_actions(&ds, facet->xout.odp_actions.data,
5816 facet->xout.odp_actions.size);
5818 ds_put_char(&ds, '\n');
5821 unixctl_command_reply(conn, ds_cstr(&ds));
5826 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5827 int argc OVS_UNUSED, const char *argv[],
5828 void *aux OVS_UNUSED)
5830 struct ds ds = DS_EMPTY_INITIALIZER;
5831 struct ofproto_dpif *ofproto;
5833 ofproto = ofproto_dpif_lookup(argv[1]);
5835 unixctl_command_reply_error(conn, "no such bridge");
5839 flush(&ofproto->up);
5841 unixctl_command_reply(conn, ds_cstr(&ds));
5846 ofproto_dpif_unixctl_init(void)
5848 static bool registered;
5854 unixctl_command_register(
5856 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5857 1, 3, ofproto_unixctl_trace, NULL);
5858 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5859 ofproto_unixctl_fdb_flush, NULL);
5860 unixctl_command_register("fdb/show", "bridge", 1, 1,
5861 ofproto_unixctl_fdb_show, NULL);
5862 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5863 ofproto_dpif_self_check, NULL);
5864 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5865 ofproto_unixctl_dpif_dump_dps, NULL);
5866 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5868 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5869 ofproto_unixctl_dpif_dump_flows, NULL);
5870 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5871 ofproto_unixctl_dpif_del_flows, NULL);
5872 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5873 ofproto_unixctl_dpif_dump_megaflows, NULL);
5874 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5875 ofproto_unixctl_dpif_disable_megaflows, NULL);
5876 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5877 ofproto_unixctl_dpif_enable_megaflows, NULL);
5880 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5882 * This is deprecated. It is only for compatibility with broken device drivers
5883 * in old versions of Linux that do not properly support VLANs when VLAN
5884 * devices are not used. When broken device drivers are no longer in
5885 * widespread use, we will delete these interfaces. */
5888 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5890 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5891 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5893 if (realdev_ofp_port == ofport->realdev_ofp_port
5894 && vid == ofport->vlandev_vid) {
5898 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5900 if (ofport->realdev_ofp_port) {
5903 if (realdev_ofp_port && ofport->bundle) {
5904 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5905 * themselves be part of a bundle. */
5906 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5909 ofport->realdev_ofp_port = realdev_ofp_port;
5910 ofport->vlandev_vid = vid;
5912 if (realdev_ofp_port) {
5913 vsp_add(ofport, realdev_ofp_port, vid);
5920 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5922 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5926 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5927 OVS_EXCLUDED(ofproto->vsp_mutex)
5931 ovs_mutex_lock(&ofproto->vsp_mutex);
5932 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
5933 ovs_mutex_unlock(&ofproto->vsp_mutex);
5938 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5939 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5940 OVS_REQUIRES(ofproto->vsp_mutex)
5942 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5943 int vid = vlan_tci_to_vid(vlan_tci);
5944 const struct vlan_splinter *vsp;
5946 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5947 hash_realdev_vid(realdev_ofp_port, vid),
5948 &ofproto->realdev_vid_map) {
5949 if (vsp->realdev_ofp_port == realdev_ofp_port
5950 && vsp->vid == vid) {
5951 return vsp->vlandev_ofp_port;
5955 return realdev_ofp_port;
5958 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5959 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5960 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5961 * 'vlan_tci' 9, it would return the port number of eth0.9.
5963 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5964 * function just returns its 'realdev_ofp_port' argument. */
5966 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5967 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5968 OVS_EXCLUDED(ofproto->vsp_mutex)
5972 ovs_mutex_lock(&ofproto->vsp_mutex);
5973 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5974 ovs_mutex_unlock(&ofproto->vsp_mutex);
5978 static struct vlan_splinter *
5979 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5981 struct vlan_splinter *vsp;
5983 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5984 hash_ofp_port(vlandev_ofp_port),
5985 &ofproto->vlandev_map) {
5986 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5994 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5995 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5996 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5997 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5998 * eth0 and store 9 in '*vid'.
6000 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6001 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6004 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6005 ofp_port_t vlandev_ofp_port, int *vid)
6006 OVS_REQUIRES(ofproto->vsp_mutex)
6008 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6009 const struct vlan_splinter *vsp;
6011 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6016 return vsp->realdev_ofp_port;
6022 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6023 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6024 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6025 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6026 * always the case unless VLAN splinters are enabled), returns false without
6027 * making any changes. */
6029 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6030 OVS_EXCLUDED(ofproto->vsp_mutex)
6035 ovs_mutex_lock(&ofproto->vsp_mutex);
6036 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6037 ovs_mutex_unlock(&ofproto->vsp_mutex);
6042 /* Cause the flow to be processed as if it came in on the real device with
6043 * the VLAN device's VLAN ID. */
6044 flow->in_port.ofp_port = realdev;
6045 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6050 vsp_remove(struct ofport_dpif *port)
6052 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6053 struct vlan_splinter *vsp;
6055 ovs_mutex_lock(&ofproto->vsp_mutex);
6056 vsp = vlandev_find(ofproto, port->up.ofp_port);
6058 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6059 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6062 port->realdev_ofp_port = 0;
6064 VLOG_ERR("missing vlan device record");
6066 ovs_mutex_unlock(&ofproto->vsp_mutex);
6070 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6074 ovs_mutex_lock(&ofproto->vsp_mutex);
6075 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6076 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6077 == realdev_ofp_port)) {
6078 struct vlan_splinter *vsp;
6080 vsp = xmalloc(sizeof *vsp);
6081 vsp->realdev_ofp_port = realdev_ofp_port;
6082 vsp->vlandev_ofp_port = port->up.ofp_port;
6085 port->realdev_ofp_port = realdev_ofp_port;
6087 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6088 hash_ofp_port(port->up.ofp_port));
6089 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6090 hash_realdev_vid(realdev_ofp_port, vid));
6092 VLOG_ERR("duplicate vlan device record");
6094 ovs_mutex_unlock(&ofproto->vsp_mutex);
6098 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6100 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6101 return ofport ? ofport->odp_port : ODPP_NONE;
6104 struct ofport_dpif *
6105 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6107 struct ofport_dpif *port;
6109 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6110 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6111 &backer->odp_to_ofport_map) {
6112 if (port->odp_port == odp_port) {
6113 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6118 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6123 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6125 struct ofport_dpif *port;
6127 port = odp_port_to_ofport(ofproto->backer, odp_port);
6128 if (port && &ofproto->up == port->up.ofproto) {
6129 return port->up.ofp_port;
6135 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6136 * most heavily weighted element. 'base' designates the rate of decay: after
6137 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6140 exp_mavg(double *avg, int base, double new)
6142 *avg = (*avg * (base - 1) + new) / base;
6146 update_moving_averages(struct dpif_backer *backer)
6148 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6149 long long int minutes = (time_msec() - backer->created) / min_ms;
6152 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6154 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6157 backer->lifetime.add_rate = 0.0;
6158 backer->lifetime.del_rate = 0.0;
6161 /* Update hourly averages on the minute boundaries. */
6162 if (time_msec() - backer->last_minute >= min_ms) {
6163 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6164 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6166 /* Update daily averages on the hour boundaries. */
6167 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6168 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6169 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6172 backer->total_subfacet_add_count += backer->subfacet_add_count;
6173 backer->total_subfacet_del_count += backer->subfacet_del_count;
6174 backer->subfacet_add_count = 0;
6175 backer->subfacet_del_count = 0;
6176 backer->last_minute += min_ms;
6180 const struct ofproto_class ofproto_dpif_class = {
6215 port_is_lacp_current,
6216 NULL, /* rule_choose_table */
6225 rule_modify_actions,
6239 get_stp_port_status,
6246 is_mirror_output_bundle,
6247 forward_bpdu_changed,
6248 set_mac_table_config,
6250 NULL, /* meter_get_features */
6251 NULL, /* meter_set */
6252 NULL, /* meter_get */
6253 NULL, /* meter_del */
6254 NULL, /* group_alloc */
6255 NULL, /* group_construct */
6256 NULL, /* group_destruct */
6257 NULL, /* group_dealloc */
6258 NULL, /* group_modify */
6259 NULL, /* group_get_stats */