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"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-mirror.h"
54 #include "ofproto-dpif-sflow.h"
55 #include "ofproto-dpif-upcall.h"
56 #include "ofproto-dpif-xlate.h"
57 #include "poll-loop.h"
62 #include "unaligned.h"
64 #include "vlan-bitmap.h"
67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
69 COVERAGE_DEFINE(ofproto_dpif_expired);
70 COVERAGE_DEFINE(facet_changed_rule);
71 COVERAGE_DEFINE(facet_revalidate);
72 COVERAGE_DEFINE(facet_unexpected);
73 COVERAGE_DEFINE(facet_suppress);
74 COVERAGE_DEFINE(subfacet_install_fail);
75 COVERAGE_DEFINE(packet_in_overflow);
76 COVERAGE_DEFINE(flow_mod_overflow);
78 /* Number of implemented OpenFlow tables. */
79 enum { N_TABLES = 255 };
80 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
91 * - Do include packets and bytes from facets that have been deleted or
92 * whose own statistics have been folded into the rule.
94 * - Do include packets and bytes sent "by hand" that were accounted to
95 * the rule without any facet being involved (this is a rare corner
96 * case in rule_execute()).
98 * - Do not include packet or bytes that can be obtained from any facet's
99 * packet_count or byte_count member or that can be obtained from the
100 * datapath by, e.g., dpif_flow_get() for any subfacet.
102 struct ovs_mutex stats_mutex;
103 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
104 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
107 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
108 static struct rule_dpif *rule_dpif_cast(const struct rule *);
111 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
112 struct ofproto_dpif *ofproto; /* Owning ofproto. */
113 void *aux; /* Key supplied by ofproto's client. */
114 char *name; /* Identifier for log messages. */
117 struct list ports; /* Contains "struct ofport"s. */
118 enum port_vlan_mode vlan_mode; /* VLAN mode */
119 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
120 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
121 * NULL if all VLANs are trunked. */
122 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
123 struct bond *bond; /* Nonnull iff more than one port. */
124 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
127 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
130 static void bundle_remove(struct ofport *);
131 static void bundle_update(struct ofbundle *);
132 static void bundle_destroy(struct ofbundle *);
133 static void bundle_del_port(struct ofport_dpif *);
134 static void bundle_run(struct ofbundle *);
135 static void bundle_wait(struct ofbundle *);
137 static void stp_run(struct ofproto_dpif *ofproto);
138 static void stp_wait(struct ofproto_dpif *ofproto);
139 static int set_stp_port(struct ofport *,
140 const struct ofproto_port_stp_settings *);
142 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
143 enum slow_path_reason,
144 uint64_t *stub, size_t stub_size,
145 const struct nlattr **actionsp,
146 size_t *actions_lenp);
148 /* A subfacet (see "struct subfacet" below) has three possible installation
151 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
152 * case just after the subfacet is created, just before the subfacet is
153 * destroyed, or if the datapath returns an error when we try to install a
156 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
158 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
159 * ofproto_dpif is installed in the datapath.
162 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
163 SF_FAST_PATH, /* Full actions are installed. */
164 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
167 /* A dpif flow and actions associated with a facet.
169 * See also the large comment on struct facet. */
172 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
173 struct list list_node; /* In struct facet's 'facets' list. */
174 struct facet *facet; /* Owning facet. */
175 struct dpif_backer *backer; /* Owning backer. */
177 enum odp_key_fitness key_fitness;
181 long long int used; /* Time last used; time created if not used. */
182 long long int created; /* Time created. */
184 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
185 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
187 enum subfacet_path path; /* Installed in datapath? */
190 #define SUBFACET_DESTROY_MAX_BATCH 50
192 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
193 static struct subfacet *subfacet_find(struct dpif_backer *,
194 const struct nlattr *key, size_t key_len,
196 static void subfacet_destroy(struct subfacet *);
197 static void subfacet_destroy__(struct subfacet *);
198 static void subfacet_destroy_batch(struct dpif_backer *,
199 struct subfacet **, int n);
200 static void subfacet_reset_dp_stats(struct subfacet *,
201 struct dpif_flow_stats *);
202 static void subfacet_update_stats(struct subfacet *,
203 const struct dpif_flow_stats *);
204 static int subfacet_install(struct subfacet *,
205 const struct ofpbuf *odp_actions,
206 struct dpif_flow_stats *);
207 static void subfacet_uninstall(struct subfacet *);
209 /* A unique, non-overlapping instantiation of an OpenFlow flow.
211 * A facet associates a "struct flow", which represents the Open vSwitch
212 * userspace idea of an exact-match flow, with one or more subfacets.
213 * While the facet is created based on an exact-match flow, it is stored
214 * within the ofproto based on the wildcards that could be expressed
215 * based on the flow table and other configuration. (See the 'wc'
216 * description in "struct xlate_out" for more details.)
218 * Each subfacet tracks the datapath's idea of the flow equivalent to
219 * the facet. When the kernel module (or other dpif implementation) and
220 * Open vSwitch userspace agree on the definition of a flow key, there
221 * is exactly one subfacet per facet. If the dpif implementation
222 * supports more-specific flow matching than userspace, however, a facet
223 * can have more than one subfacet. Examples include the dpif
224 * implementation not supporting the same wildcards as userspace or some
225 * distinction in flow that userspace simply doesn't understand.
227 * Flow expiration works in terms of subfacets, so a facet must have at
228 * least one subfacet or it will never expire, leaking memory. */
231 struct ofproto_dpif *ofproto;
234 struct list subfacets;
235 long long int used; /* Time last used; time created if not used. */
238 struct flow flow; /* Flow of the creating subfacet. */
239 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
243 * - Do include packets and bytes sent "by hand", e.g. with
246 * - Do include packets and bytes that were obtained from the datapath
247 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
248 * DPIF_FP_ZERO_STATS).
250 * - Do not include packets or bytes that can be obtained from the
251 * datapath for any existing subfacet.
253 uint64_t packet_count; /* Number of packets received. */
254 uint64_t byte_count; /* Number of bytes received. */
256 /* Resubmit statistics. */
257 uint64_t prev_packet_count; /* Number of packets from last stats push. */
258 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
259 long long int prev_used; /* Used time from last stats push. */
262 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
263 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
264 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
266 struct xlate_out xout;
268 /* Storage for a single subfacet, to reduce malloc() time and space
269 * overhead. (A facet always has at least one subfacet and in the common
270 * case has exactly one subfacet. However, 'one_subfacet' may not
271 * always be valid, since it could have been removed after newer
272 * subfacets were pushed onto the 'subfacets' list.) */
273 struct subfacet one_subfacet;
275 long long int learn_rl; /* Rate limiter for facet_learn(). */
278 static struct facet *facet_create(const struct flow_miss *);
279 static void facet_remove(struct facet *);
280 static void facet_free(struct facet *);
282 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
283 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
284 const struct flow *);
285 static bool facet_revalidate(struct facet *);
286 static bool facet_check_consistency(struct facet *);
288 static void facet_flush_stats(struct facet *);
290 static void facet_reset_counters(struct facet *);
291 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
292 struct dpif_flow_stats *, bool may_learn);
293 static void facet_push_stats(struct facet *, bool may_learn);
294 static void facet_learn(struct facet *);
295 static void facet_account(struct facet *);
296 static void push_all_stats(void);
298 static bool facet_is_controller_flow(struct facet *);
301 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
305 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
306 struct list bundle_node; /* In struct ofbundle's "ports" list. */
307 struct cfm *cfm; /* Connectivity Fault Management, if any. */
308 struct bfd *bfd; /* BFD, if any. */
309 bool may_enable; /* May be enabled in bonds. */
310 bool is_tunnel; /* This port is a tunnel. */
311 long long int carrier_seq; /* Carrier status changes. */
312 struct ofport_dpif *peer; /* Peer if patch port. */
315 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
316 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
317 long long int stp_state_entered;
319 /* Queue to DSCP mapping. */
320 struct ofproto_port_queue *qdscp;
323 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
325 * This is deprecated. It is only for compatibility with broken device
326 * drivers in old versions of Linux that do not properly support VLANs when
327 * VLAN devices are not used. When broken device drivers are no longer in
328 * widespread use, we will delete these interfaces. */
329 ofp_port_t realdev_ofp_port;
333 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
335 * This is deprecated. It is only for compatibility with broken device drivers
336 * in old versions of Linux that do not properly support VLANs when VLAN
337 * devices are not used. When broken device drivers are no longer in
338 * widespread use, we will delete these interfaces. */
339 struct vlan_splinter {
340 struct hmap_node realdev_vid_node;
341 struct hmap_node vlandev_node;
342 ofp_port_t realdev_ofp_port;
343 ofp_port_t vlandev_ofp_port;
347 static void vsp_remove(struct ofport_dpif *);
348 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
350 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
353 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
356 static struct ofport_dpif *
357 ofport_dpif_cast(const struct ofport *ofport)
359 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
362 static void port_run(struct ofport_dpif *);
363 static void port_run_fast(struct ofport_dpif *);
364 static void port_wait(struct ofport_dpif *);
365 static int set_bfd(struct ofport *, const struct smap *);
366 static int set_cfm(struct ofport *, const struct cfm_settings *);
367 static void ofport_update_peer(struct ofport_dpif *);
368 static void run_fast_rl(void);
369 static int run_fast(struct ofproto *);
371 struct dpif_completion {
372 struct list list_node;
373 struct ofoperation *op;
376 /* Reasons that we might need to revalidate every facet, and corresponding
379 * A value of 0 means that there is no need to revalidate.
381 * It would be nice to have some cleaner way to integrate with coverage
382 * counters, but with only a few reasons I guess this is good enough for
384 enum revalidate_reason {
385 REV_RECONFIGURE = 1, /* Switch configuration changed. */
386 REV_STP, /* Spanning tree protocol port status change. */
387 REV_BOND, /* Bonding changed. */
388 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
389 REV_FLOW_TABLE, /* Flow table changed. */
390 REV_MAC_LEARNING, /* Mac learning changed. */
391 REV_INCONSISTENCY /* Facet self-check failed. */
393 COVERAGE_DEFINE(rev_reconfigure);
394 COVERAGE_DEFINE(rev_stp);
395 COVERAGE_DEFINE(rev_bond);
396 COVERAGE_DEFINE(rev_port_toggled);
397 COVERAGE_DEFINE(rev_flow_table);
398 COVERAGE_DEFINE(rev_mac_learning);
399 COVERAGE_DEFINE(rev_inconsistency);
401 struct avg_subfacet_rates {
402 double add_rate; /* Moving average of new flows created per minute. */
403 double del_rate; /* Moving average of flows deleted per minute. */
406 /* All datapaths of a given type share a single dpif backer instance. */
412 struct timer next_expiration;
414 struct ovs_rwlock odp_to_ofport_lock;
415 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
417 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
419 /* Facet revalidation flags applying to facets which use this backer. */
420 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
422 struct hmap drop_keys; /* Set of dropped odp keys. */
423 bool recv_set_enable; /* Enables or disables receiving packets. */
425 struct hmap subfacets;
426 struct governor *governor;
428 /* Subfacet statistics.
430 * These keep track of the total number of subfacets added and deleted and
431 * flow life span. They are useful for computing the flow rates stats
432 * exposed via "ovs-appctl dpif/show". The goal is to learn about
433 * traffic patterns in ways that we can use later to improve Open vSwitch
434 * performance in new situations. */
435 long long int created; /* Time when it is created. */
436 unsigned max_n_subfacet; /* Maximum number of flows */
437 unsigned avg_n_subfacet; /* Average number of flows. */
438 long long int avg_subfacet_life; /* Average life span of subfacets. */
440 /* The average number of subfacets... */
441 struct avg_subfacet_rates hourly; /* ...over the last hour. */
442 struct avg_subfacet_rates daily; /* ...over the last day. */
443 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
444 long long int last_minute; /* Last time 'hourly' was updated. */
446 /* Number of subfacets added or deleted since 'last_minute'. */
447 unsigned subfacet_add_count;
448 unsigned subfacet_del_count;
450 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
451 unsigned long long int total_subfacet_add_count;
452 unsigned long long int total_subfacet_del_count;
454 /* Number of upcall handling threads. */
455 unsigned int n_handler_threads;
458 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
459 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
461 static void drop_key_clear(struct dpif_backer *);
462 static void update_moving_averages(struct dpif_backer *backer);
464 struct ofproto_dpif {
465 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
467 struct dpif_backer *backer;
469 /* Special OpenFlow rules. */
470 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
471 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
472 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
475 struct netflow *netflow;
476 struct dpif_sflow *sflow;
477 struct dpif_ipfix *ipfix;
478 struct hmap bundles; /* Contains "struct ofbundle"s. */
479 struct mac_learning *ml;
480 bool has_bonded_bundles;
481 struct mbridge *mbridge;
484 struct classifier facets; /* Contains 'struct facet's. */
485 long long int consistency_rl;
487 struct netdev_stats stats; /* To account packets generated and consumed in
492 long long int stp_last_tick;
494 /* VLAN splinters. */
495 struct ovs_mutex vsp_mutex;
496 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
497 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
500 struct sset ports; /* Set of standard port names. */
501 struct sset ghost_ports; /* Ports with no datapath port. */
502 struct sset port_poll_set; /* Queued names for port_poll() reply. */
503 int port_poll_errno; /* Last errno for port_poll() reply. */
505 /* Per ofproto's dpif stats. */
510 struct ovs_mutex flow_mod_mutex;
511 struct list flow_mods OVS_GUARDED;
512 size_t n_flow_mods OVS_GUARDED;
514 struct ovs_mutex pin_mutex;
515 struct list pins OVS_GUARDED;
516 size_t n_pins OVS_GUARDED;
519 /* By default, flows in the datapath are wildcarded (megaflows). They
520 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
521 static bool enable_megaflows = true;
523 /* All existing ofproto_dpif instances, indexed by ->up.name. */
524 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
526 static void ofproto_dpif_unixctl_init(void);
528 static inline struct ofproto_dpif *
529 ofproto_dpif_cast(const struct ofproto *ofproto)
531 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
532 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
535 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
536 ofp_port_t ofp_port);
537 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
538 const struct ofpbuf *packet, struct ds *);
541 static void handle_upcalls(struct dpif_backer *);
543 /* Flow expiration. */
544 static int expire(struct dpif_backer *);
547 static void send_netflow_active_timeouts(struct ofproto_dpif *);
550 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
552 /* Global variables. */
553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
555 /* Initial mappings of port to bridge mappings. */
556 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
558 /* Executes and takes ownership of 'fm'. */
560 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
561 struct ofputil_flow_mod *fm)
563 ovs_mutex_lock(&ofproto->flow_mod_mutex);
564 if (ofproto->n_flow_mods > 1024) {
565 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
566 COVERAGE_INC(flow_mod_overflow);
572 list_push_back(&ofproto->flow_mods, &fm->list_node);
573 ofproto->n_flow_mods++;
574 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
577 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
578 * Takes ownership of 'pin' and pin->packet. */
580 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
581 struct ofputil_packet_in *pin)
583 ovs_mutex_lock(&ofproto->pin_mutex);
584 if (ofproto->n_pins > 1024) {
585 ovs_mutex_unlock(&ofproto->pin_mutex);
586 COVERAGE_INC(packet_in_overflow);
587 free(CONST_CAST(void *, pin->packet));
592 list_push_back(&ofproto->pins, &pin->list_node);
594 ovs_mutex_unlock(&ofproto->pin_mutex);
597 /* Factory functions. */
600 init(const struct shash *iface_hints)
602 struct shash_node *node;
604 /* Make a local copy, since we don't own 'iface_hints' elements. */
605 SHASH_FOR_EACH(node, iface_hints) {
606 const struct iface_hint *orig_hint = node->data;
607 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
609 new_hint->br_name = xstrdup(orig_hint->br_name);
610 new_hint->br_type = xstrdup(orig_hint->br_type);
611 new_hint->ofp_port = orig_hint->ofp_port;
613 shash_add(&init_ofp_ports, node->name, new_hint);
618 enumerate_types(struct sset *types)
620 dp_enumerate_types(types);
624 enumerate_names(const char *type, struct sset *names)
626 struct ofproto_dpif *ofproto;
629 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
630 if (strcmp(type, ofproto->up.type)) {
633 sset_add(names, ofproto->up.name);
640 del(const char *type, const char *name)
645 error = dpif_open(name, type, &dpif);
647 error = dpif_delete(dpif);
654 port_open_type(const char *datapath_type, const char *port_type)
656 return dpif_port_open_type(datapath_type, port_type);
659 /* Type functions. */
661 static void process_dpif_port_changes(struct dpif_backer *);
662 static void process_dpif_all_ports_changed(struct dpif_backer *);
663 static void process_dpif_port_change(struct dpif_backer *,
664 const char *devname);
665 static void process_dpif_port_error(struct dpif_backer *, int error);
667 static struct ofproto_dpif *
668 lookup_ofproto_dpif_by_port_name(const char *name)
670 struct ofproto_dpif *ofproto;
672 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
673 if (sset_contains(&ofproto->ports, name)) {
682 type_run(const char *type)
684 static long long int push_timer = LLONG_MIN;
685 struct dpif_backer *backer;
687 backer = shash_find_data(&all_dpif_backers, type);
689 /* This is not necessarily a problem, since backers are only
690 * created on demand. */
694 dpif_run(backer->dpif);
696 /* The most natural place to push facet statistics is when they're pulled
697 * from the datapath. However, when there are many flows in the datapath,
698 * this expensive operation can occur so frequently, that it reduces our
699 * ability to quickly set up flows. To reduce the cost, we push statistics
701 if (time_msec() > push_timer) {
702 push_timer = time_msec() + 2000;
706 /* If vswitchd started with other_config:flow_restore_wait set as "true",
707 * and the configuration has now changed to "false", enable receiving
708 * packets from the datapath. */
709 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
712 backer->recv_set_enable = true;
714 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
716 udpif_recv_set(backer->udpif, 0, false);
717 VLOG_ERR("Failed to enable receiving packets in dpif.");
720 udpif_recv_set(backer->udpif, n_handler_threads,
721 backer->recv_set_enable);
722 dpif_flow_flush(backer->dpif);
723 backer->need_revalidate = REV_RECONFIGURE;
726 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
727 * to reset the handler threads. */
728 if (backer->n_handler_threads != n_handler_threads) {
729 udpif_recv_set(backer->udpif, n_handler_threads,
730 backer->recv_set_enable);
731 backer->n_handler_threads = n_handler_threads;
734 if (backer->need_revalidate) {
735 struct ofproto_dpif *ofproto;
736 struct simap_node *node;
737 struct simap tmp_backers;
739 /* Handle tunnel garbage collection. */
740 simap_init(&tmp_backers);
741 simap_swap(&backer->tnl_backers, &tmp_backers);
743 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
744 struct ofport_dpif *iter;
746 if (backer != ofproto->backer) {
750 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
751 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
754 if (!iter->is_tunnel) {
758 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
759 namebuf, sizeof namebuf);
760 node = simap_find(&tmp_backers, dp_port);
762 simap_put(&backer->tnl_backers, dp_port, node->data);
763 simap_delete(&tmp_backers, node);
764 node = simap_find(&backer->tnl_backers, dp_port);
766 node = simap_find(&backer->tnl_backers, dp_port);
768 odp_port_t odp_port = ODPP_NONE;
770 if (!dpif_port_add(backer->dpif, iter->up.netdev,
772 simap_put(&backer->tnl_backers, dp_port,
773 odp_to_u32(odp_port));
774 node = simap_find(&backer->tnl_backers, dp_port);
779 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
780 if (tnl_port_reconfigure(iter, iter->up.netdev,
782 backer->need_revalidate = REV_RECONFIGURE;
787 SIMAP_FOR_EACH (node, &tmp_backers) {
788 dpif_port_del(backer->dpif, u32_to_odp(node->data));
790 simap_destroy(&tmp_backers);
792 switch (backer->need_revalidate) {
793 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
794 case REV_STP: COVERAGE_INC(rev_stp); break;
795 case REV_BOND: COVERAGE_INC(rev_bond); break;
796 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
797 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
798 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
799 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
801 backer->need_revalidate = 0;
803 /* Clear the drop_keys in case we should now be accepting some
804 * formerly dropped flows. */
805 drop_key_clear(backer);
807 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
808 struct facet *facet, *next;
809 struct ofport_dpif *ofport;
810 struct cls_cursor cursor;
811 struct ofbundle *bundle;
813 if (ofproto->backer != backer) {
817 ovs_rwlock_wrlock(&xlate_rwlock);
818 xlate_ofproto_set(ofproto, ofproto->up.name,
819 ofproto->backer->dpif, ofproto->miss_rule,
820 ofproto->no_packet_in_rule, ofproto->ml,
821 ofproto->stp, ofproto->mbridge,
822 ofproto->sflow, ofproto->ipfix,
823 ofproto->up.frag_handling,
824 ofproto->up.forward_bpdu,
825 connmgr_has_in_band(ofproto->up.connmgr),
826 ofproto->netflow != NULL);
828 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
829 xlate_bundle_set(ofproto, bundle, bundle->name,
830 bundle->vlan_mode, bundle->vlan,
831 bundle->trunks, bundle->use_priority_tags,
832 bundle->bond, bundle->lacp,
836 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
837 int stp_port = ofport->stp_port
838 ? stp_port_no(ofport->stp_port)
840 xlate_ofport_set(ofproto, ofport->bundle, ofport,
841 ofport->up.ofp_port, ofport->odp_port,
842 ofport->up.netdev, ofport->cfm,
843 ofport->bfd, ofport->peer, stp_port,
844 ofport->qdscp, ofport->n_qdscp,
845 ofport->up.pp.config, ofport->is_tunnel,
848 ovs_rwlock_unlock(&xlate_rwlock);
850 /* Only ofproto-dpif cares about the facet classifier so we just
851 * lock cls_cursor_init() to appease the thread safety analysis. */
852 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
853 cls_cursor_init(&cursor, &ofproto->facets, NULL);
854 ovs_rwlock_unlock(&ofproto->facets.rwlock);
855 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
856 facet_revalidate(facet);
861 udpif_revalidate(backer->udpif);
864 if (!backer->recv_set_enable) {
865 /* Wake up before a max of 1000ms. */
866 timer_set_duration(&backer->next_expiration, 1000);
867 } else if (timer_expired(&backer->next_expiration)) {
868 int delay = expire(backer);
869 timer_set_duration(&backer->next_expiration, delay);
872 process_dpif_port_changes(backer);
874 if (backer->governor) {
877 governor_run(backer->governor);
879 /* If the governor has shrunk to its minimum size and the number of
880 * subfacets has dwindled, then drop the governor entirely.
882 * For hysteresis, the number of subfacets to drop the governor is
883 * smaller than the number needed to trigger its creation. */
884 n_subfacets = hmap_count(&backer->subfacets);
885 if (n_subfacets * 4 < flow_eviction_threshold
886 && governor_is_idle(backer->governor)) {
887 governor_destroy(backer->governor);
888 backer->governor = NULL;
895 /* Check for and handle port changes in 'backer''s dpif. */
897 process_dpif_port_changes(struct dpif_backer *backer)
903 error = dpif_port_poll(backer->dpif, &devname);
909 process_dpif_all_ports_changed(backer);
913 process_dpif_port_change(backer, devname);
918 process_dpif_port_error(backer, error);
925 process_dpif_all_ports_changed(struct dpif_backer *backer)
927 struct ofproto_dpif *ofproto;
928 struct dpif_port dpif_port;
929 struct dpif_port_dump dump;
930 struct sset devnames;
933 sset_init(&devnames);
934 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
935 if (ofproto->backer == backer) {
936 struct ofport *ofport;
938 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
939 sset_add(&devnames, netdev_get_name(ofport->netdev));
943 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
944 sset_add(&devnames, dpif_port.name);
947 SSET_FOR_EACH (devname, &devnames) {
948 process_dpif_port_change(backer, devname);
950 sset_destroy(&devnames);
954 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
956 struct ofproto_dpif *ofproto;
957 struct dpif_port port;
959 /* Don't report on the datapath's device. */
960 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
964 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
965 &all_ofproto_dpifs) {
966 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
971 ofproto = lookup_ofproto_dpif_by_port_name(devname);
972 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
973 /* The port was removed. If we know the datapath,
974 * report it through poll_set(). If we don't, it may be
975 * notifying us of a removal we initiated, so ignore it.
976 * If there's a pending ENOBUFS, let it stand, since
977 * everything will be reevaluated. */
978 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
979 sset_add(&ofproto->port_poll_set, devname);
980 ofproto->port_poll_errno = 0;
982 } else if (!ofproto) {
983 /* The port was added, but we don't know with which
984 * ofproto we should associate it. Delete it. */
985 dpif_port_del(backer->dpif, port.port_no);
987 struct ofport_dpif *ofport;
989 ofport = ofport_dpif_cast(shash_find_data(
990 &ofproto->up.port_by_name, devname));
992 && ofport->odp_port != port.port_no
993 && !odp_port_to_ofport(backer, port.port_no))
995 /* 'ofport''s datapath port number has changed from
996 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
997 * structures to match. */
998 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
999 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
1000 ofport->odp_port = port.port_no;
1001 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
1002 hash_odp_port(port.port_no));
1003 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
1004 backer->need_revalidate = REV_RECONFIGURE;
1007 dpif_port_destroy(&port);
1010 /* Propagate 'error' to all ofprotos based on 'backer'. */
1012 process_dpif_port_error(struct dpif_backer *backer, int error)
1014 struct ofproto_dpif *ofproto;
1016 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1017 if (ofproto->backer == backer) {
1018 sset_clear(&ofproto->port_poll_set);
1019 ofproto->port_poll_errno = error;
1025 dpif_backer_run_fast(struct dpif_backer *backer)
1027 udpif_run(backer->udpif);
1028 handle_upcalls(backer);
1034 type_run_fast(const char *type)
1036 struct dpif_backer *backer;
1038 backer = shash_find_data(&all_dpif_backers, type);
1040 /* This is not necessarily a problem, since backers are only
1041 * created on demand. */
1045 return dpif_backer_run_fast(backer);
1051 static long long int port_rl = LLONG_MIN;
1053 if (time_msec() >= port_rl) {
1054 struct ofproto_dpif *ofproto;
1056 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1057 run_fast(&ofproto->up);
1059 port_rl = time_msec() + 200;
1064 type_wait(const char *type)
1066 struct dpif_backer *backer;
1068 backer = shash_find_data(&all_dpif_backers, type);
1070 /* This is not necessarily a problem, since backers are only
1071 * created on demand. */
1075 if (backer->governor) {
1076 governor_wait(backer->governor);
1079 timer_wait(&backer->next_expiration);
1080 dpif_wait(backer->dpif);
1081 udpif_wait(backer->udpif);
1084 /* Basic life-cycle. */
1086 static int add_internal_flows(struct ofproto_dpif *);
1088 static struct ofproto *
1091 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1092 return &ofproto->up;
1096 dealloc(struct ofproto *ofproto_)
1098 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1103 close_dpif_backer(struct dpif_backer *backer)
1105 struct shash_node *node;
1107 ovs_assert(backer->refcount > 0);
1109 if (--backer->refcount) {
1113 drop_key_clear(backer);
1114 hmap_destroy(&backer->drop_keys);
1116 simap_destroy(&backer->tnl_backers);
1117 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1118 hmap_destroy(&backer->odp_to_ofport_map);
1119 node = shash_find(&all_dpif_backers, backer->type);
1121 shash_delete(&all_dpif_backers, node);
1122 udpif_destroy(backer->udpif);
1123 dpif_close(backer->dpif);
1125 ovs_assert(hmap_is_empty(&backer->subfacets));
1126 hmap_destroy(&backer->subfacets);
1127 governor_destroy(backer->governor);
1132 /* Datapath port slated for removal from datapath. */
1133 struct odp_garbage {
1134 struct list list_node;
1135 odp_port_t odp_port;
1139 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1141 struct dpif_backer *backer;
1142 struct dpif_port_dump port_dump;
1143 struct dpif_port port;
1144 struct shash_node *node;
1145 struct list garbage_list;
1146 struct odp_garbage *garbage, *next;
1152 backer = shash_find_data(&all_dpif_backers, type);
1159 backer_name = xasprintf("ovs-%s", type);
1161 /* Remove any existing datapaths, since we assume we're the only
1162 * userspace controlling the datapath. */
1164 dp_enumerate_names(type, &names);
1165 SSET_FOR_EACH(name, &names) {
1166 struct dpif *old_dpif;
1168 /* Don't remove our backer if it exists. */
1169 if (!strcmp(name, backer_name)) {
1173 if (dpif_open(name, type, &old_dpif)) {
1174 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1176 dpif_delete(old_dpif);
1177 dpif_close(old_dpif);
1180 sset_destroy(&names);
1182 backer = xmalloc(sizeof *backer);
1184 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1187 VLOG_ERR("failed to open datapath of type %s: %s", type,
1188 ovs_strerror(error));
1192 backer->udpif = udpif_create(backer, backer->dpif);
1194 backer->type = xstrdup(type);
1195 backer->governor = NULL;
1196 backer->refcount = 1;
1197 hmap_init(&backer->odp_to_ofport_map);
1198 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1199 hmap_init(&backer->drop_keys);
1200 hmap_init(&backer->subfacets);
1201 timer_set_duration(&backer->next_expiration, 1000);
1202 backer->need_revalidate = 0;
1203 simap_init(&backer->tnl_backers);
1204 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1207 if (backer->recv_set_enable) {
1208 dpif_flow_flush(backer->dpif);
1211 /* Loop through the ports already on the datapath and remove any
1212 * that we don't need anymore. */
1213 list_init(&garbage_list);
1214 dpif_port_dump_start(&port_dump, backer->dpif);
1215 while (dpif_port_dump_next(&port_dump, &port)) {
1216 node = shash_find(&init_ofp_ports, port.name);
1217 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1218 garbage = xmalloc(sizeof *garbage);
1219 garbage->odp_port = port.port_no;
1220 list_push_front(&garbage_list, &garbage->list_node);
1223 dpif_port_dump_done(&port_dump);
1225 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1226 dpif_port_del(backer->dpif, garbage->odp_port);
1227 list_remove(&garbage->list_node);
1231 shash_add(&all_dpif_backers, type, backer);
1233 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1235 VLOG_ERR("failed to listen on datapath of type %s: %s",
1236 type, ovs_strerror(error));
1237 close_dpif_backer(backer);
1240 udpif_recv_set(backer->udpif, n_handler_threads,
1241 backer->recv_set_enable);
1242 backer->n_handler_threads = n_handler_threads;
1244 backer->max_n_subfacet = 0;
1245 backer->created = time_msec();
1246 backer->last_minute = backer->created;
1247 memset(&backer->hourly, 0, sizeof backer->hourly);
1248 memset(&backer->daily, 0, sizeof backer->daily);
1249 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1250 backer->subfacet_add_count = 0;
1251 backer->subfacet_del_count = 0;
1252 backer->total_subfacet_add_count = 0;
1253 backer->total_subfacet_del_count = 0;
1254 backer->avg_n_subfacet = 0;
1255 backer->avg_subfacet_life = 0;
1261 construct(struct ofproto *ofproto_)
1263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1264 struct shash_node *node, *next;
1268 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1273 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1274 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1276 ofproto->netflow = NULL;
1277 ofproto->sflow = NULL;
1278 ofproto->ipfix = NULL;
1279 ofproto->stp = NULL;
1280 hmap_init(&ofproto->bundles);
1281 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1282 ofproto->mbridge = mbridge_create();
1283 ofproto->has_bonded_bundles = false;
1284 ovs_mutex_init(&ofproto->vsp_mutex);
1286 classifier_init(&ofproto->facets);
1287 ofproto->consistency_rl = LLONG_MIN;
1289 ovs_mutex_init(&ofproto->flow_mod_mutex);
1290 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1291 list_init(&ofproto->flow_mods);
1292 ofproto->n_flow_mods = 0;
1293 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1295 ovs_mutex_init(&ofproto->pin_mutex);
1296 ovs_mutex_lock(&ofproto->pin_mutex);
1297 list_init(&ofproto->pins);
1298 ofproto->n_pins = 0;
1299 ovs_mutex_unlock(&ofproto->pin_mutex);
1301 ofproto_dpif_unixctl_init();
1303 hmap_init(&ofproto->vlandev_map);
1304 hmap_init(&ofproto->realdev_vid_map);
1306 sset_init(&ofproto->ports);
1307 sset_init(&ofproto->ghost_ports);
1308 sset_init(&ofproto->port_poll_set);
1309 ofproto->port_poll_errno = 0;
1311 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1312 struct iface_hint *iface_hint = node->data;
1314 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1315 /* Check if the datapath already has this port. */
1316 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1317 sset_add(&ofproto->ports, node->name);
1320 free(iface_hint->br_name);
1321 free(iface_hint->br_type);
1323 shash_delete(&init_ofp_ports, node);
1327 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1328 hash_string(ofproto->up.name, 0));
1329 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1331 ofproto_init_tables(ofproto_, N_TABLES);
1332 error = add_internal_flows(ofproto);
1333 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1336 ofproto->n_missed = 0;
1342 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1343 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1345 struct ofputil_flow_mod fm;
1348 match_init_catchall(&fm.match);
1350 match_set_reg(&fm.match, 0, id);
1351 fm.new_cookie = htonll(0);
1352 fm.cookie = htonll(0);
1353 fm.cookie_mask = htonll(0);
1354 fm.modify_cookie = false;
1355 fm.table_id = TBL_INTERNAL;
1356 fm.command = OFPFC_ADD;
1357 fm.idle_timeout = 0;
1358 fm.hard_timeout = 0;
1362 fm.ofpacts = ofpacts->data;
1363 fm.ofpacts_len = ofpacts->size;
1365 error = ofproto_flow_mod(&ofproto->up, &fm);
1367 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1368 id, ofperr_to_string(error));
1372 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1374 rule_dpif_unref(*rulep);
1383 add_internal_flows(struct ofproto_dpif *ofproto)
1385 struct ofpact_controller *controller;
1386 uint64_t ofpacts_stub[128 / 8];
1387 struct ofpbuf ofpacts;
1391 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1394 controller = ofpact_put_CONTROLLER(&ofpacts);
1395 controller->max_len = UINT16_MAX;
1396 controller->controller_id = 0;
1397 controller->reason = OFPR_NO_MATCH;
1398 ofpact_pad(&ofpacts);
1400 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1405 ofpbuf_clear(&ofpacts);
1406 error = add_internal_flow(ofproto, id++, &ofpacts,
1407 &ofproto->no_packet_in_rule);
1412 error = add_internal_flow(ofproto, id++, &ofpacts,
1413 &ofproto->drop_frags_rule);
1418 destruct(struct ofproto *ofproto_)
1420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1421 struct rule_dpif *rule, *next_rule;
1422 struct ofputil_packet_in *pin, *next_pin;
1423 struct ofputil_flow_mod *fm, *next_fm;
1424 struct facet *facet, *next_facet;
1425 struct cls_cursor cursor;
1426 struct oftable *table;
1428 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1429 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1430 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1431 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1432 facet_remove(facet);
1435 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1436 ovs_rwlock_wrlock(&xlate_rwlock);
1437 xlate_remove_ofproto(ofproto);
1438 ovs_rwlock_unlock(&xlate_rwlock);
1440 flow_miss_batch_ofproto_destroyed(ofproto->backer->udpif, ofproto);
1442 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1444 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1445 struct cls_cursor cursor;
1447 ovs_rwlock_wrlock(&table->cls.rwlock);
1448 cls_cursor_init(&cursor, &table->cls, NULL);
1449 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1450 ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1452 ovs_rwlock_unlock(&table->cls.rwlock);
1455 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1456 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1457 list_remove(&fm->list_node);
1458 ofproto->n_flow_mods--;
1462 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1463 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1465 ovs_mutex_lock(&ofproto->pin_mutex);
1466 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1467 list_remove(&pin->list_node);
1469 free(CONST_CAST(void *, pin->packet));
1472 ovs_mutex_unlock(&ofproto->pin_mutex);
1473 ovs_mutex_destroy(&ofproto->pin_mutex);
1475 mbridge_unref(ofproto->mbridge);
1477 netflow_destroy(ofproto->netflow);
1478 dpif_sflow_unref(ofproto->sflow);
1479 hmap_destroy(&ofproto->bundles);
1480 mac_learning_unref(ofproto->ml);
1482 classifier_destroy(&ofproto->facets);
1484 hmap_destroy(&ofproto->vlandev_map);
1485 hmap_destroy(&ofproto->realdev_vid_map);
1487 sset_destroy(&ofproto->ports);
1488 sset_destroy(&ofproto->ghost_ports);
1489 sset_destroy(&ofproto->port_poll_set);
1491 ovs_mutex_destroy(&ofproto->vsp_mutex);
1493 close_dpif_backer(ofproto->backer);
1497 run_fast(struct ofproto *ofproto_)
1499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1500 struct ofputil_packet_in *pin, *next_pin;
1501 struct ofputil_flow_mod *fm, *next_fm;
1502 struct list flow_mods, pins;
1503 struct ofport_dpif *ofport;
1505 /* Do not perform any periodic activity required by 'ofproto' while
1506 * waiting for flow restore to complete. */
1507 if (ofproto_get_flow_restore_wait()) {
1511 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1512 list_move(&flow_mods, &ofproto->flow_mods);
1513 list_init(&ofproto->flow_mods);
1514 ofproto->n_flow_mods = 0;
1515 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1517 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1518 int error = ofproto_flow_mod(&ofproto->up, fm);
1519 if (error && !VLOG_DROP_WARN(&rl)) {
1520 VLOG_WARN("learning action failed to modify flow table (%s)",
1521 ofperr_get_name(error));
1524 list_remove(&fm->list_node);
1529 ovs_mutex_lock(&ofproto->pin_mutex);
1530 list_move(&pins, &ofproto->pins);
1531 list_init(&ofproto->pins);
1532 ofproto->n_pins = 0;
1533 ovs_mutex_unlock(&ofproto->pin_mutex);
1535 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1536 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1537 list_remove(&pin->list_node);
1538 free(CONST_CAST(void *, pin->packet));
1542 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1543 port_run_fast(ofport);
1550 run(struct ofproto *ofproto_)
1552 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1553 struct ofport_dpif *ofport;
1554 struct ofbundle *bundle;
1557 if (mbridge_need_revalidate(ofproto->mbridge)) {
1558 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1559 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1560 mac_learning_flush(ofproto->ml);
1561 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1564 /* Do not perform any periodic activity below required by 'ofproto' while
1565 * waiting for flow restore to complete. */
1566 if (ofproto_get_flow_restore_wait()) {
1570 error = run_fast(ofproto_);
1575 if (ofproto->netflow) {
1576 if (netflow_run(ofproto->netflow)) {
1577 send_netflow_active_timeouts(ofproto);
1580 if (ofproto->sflow) {
1581 dpif_sflow_run(ofproto->sflow);
1583 if (ofproto->ipfix) {
1584 dpif_ipfix_run(ofproto->ipfix);
1587 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1590 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1595 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1596 if (mac_learning_run(ofproto->ml)) {
1597 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1599 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1601 /* Check the consistency of a random facet, to aid debugging. */
1602 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1603 if (time_msec() >= ofproto->consistency_rl
1604 && !classifier_is_empty(&ofproto->facets)
1605 && !ofproto->backer->need_revalidate) {
1606 struct cls_table *table;
1607 struct cls_rule *cr;
1608 struct facet *facet;
1610 ofproto->consistency_rl = time_msec() + 250;
1612 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1613 struct cls_table, hmap_node);
1614 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1616 facet = CONTAINER_OF(cr, struct facet, cr);
1618 if (!facet_check_consistency(facet)) {
1619 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1622 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1628 wait(struct ofproto *ofproto_)
1630 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1631 struct ofport_dpif *ofport;
1632 struct ofbundle *bundle;
1634 if (ofproto_get_flow_restore_wait()) {
1638 if (ofproto->sflow) {
1639 dpif_sflow_wait(ofproto->sflow);
1641 if (ofproto->ipfix) {
1642 dpif_ipfix_wait(ofproto->ipfix);
1644 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1647 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1648 bundle_wait(bundle);
1650 if (ofproto->netflow) {
1651 netflow_wait(ofproto->netflow);
1653 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1654 mac_learning_wait(ofproto->ml);
1655 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1657 if (ofproto->backer->need_revalidate) {
1658 /* Shouldn't happen, but if it does just go around again. */
1659 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1660 poll_immediate_wake();
1665 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1667 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1668 struct cls_cursor cursor;
1669 size_t n_subfacets = 0;
1670 struct facet *facet;
1672 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1673 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1674 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1676 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1677 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1678 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1679 n_subfacets += list_size(&facet->subfacets);
1681 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1682 simap_increase(usage, "subfacets", n_subfacets);
1686 flush(struct ofproto *ofproto_)
1688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1689 struct subfacet *subfacet, *next_subfacet;
1690 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1694 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1695 &ofproto->backer->subfacets) {
1696 if (subfacet->facet->ofproto != ofproto) {
1700 if (subfacet->path != SF_NOT_INSTALLED) {
1701 batch[n_batch++] = subfacet;
1702 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1703 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1707 subfacet_destroy(subfacet);
1712 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1717 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1718 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1720 *arp_match_ip = true;
1721 *actions = (OFPUTIL_A_OUTPUT |
1722 OFPUTIL_A_SET_VLAN_VID |
1723 OFPUTIL_A_SET_VLAN_PCP |
1724 OFPUTIL_A_STRIP_VLAN |
1725 OFPUTIL_A_SET_DL_SRC |
1726 OFPUTIL_A_SET_DL_DST |
1727 OFPUTIL_A_SET_NW_SRC |
1728 OFPUTIL_A_SET_NW_DST |
1729 OFPUTIL_A_SET_NW_TOS |
1730 OFPUTIL_A_SET_TP_SRC |
1731 OFPUTIL_A_SET_TP_DST |
1736 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1739 struct dpif_dp_stats s;
1740 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1743 strcpy(ots->name, "classifier");
1745 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1746 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1747 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1748 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1750 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1751 ots->lookup_count = htonll(n_lookup);
1752 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1755 static struct ofport *
1758 struct ofport_dpif *port = xmalloc(sizeof *port);
1763 port_dealloc(struct ofport *port_)
1765 struct ofport_dpif *port = ofport_dpif_cast(port_);
1770 port_construct(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 struct netdev *netdev = port->up.netdev;
1775 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1776 struct dpif_port dpif_port;
1779 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1780 port->bundle = NULL;
1783 port->may_enable = true;
1784 port->stp_port = NULL;
1785 port->stp_state = STP_DISABLED;
1786 port->is_tunnel = false;
1790 port->realdev_ofp_port = 0;
1791 port->vlandev_vid = 0;
1792 port->carrier_seq = netdev_get_carrier_resets(netdev);
1794 if (netdev_vport_is_patch(netdev)) {
1795 /* By bailing out here, we don't submit the port to the sFlow module
1796 * to be considered for counter polling export. This is correct
1797 * because the patch port represents an interface that sFlow considers
1798 * to be "internal" to the switch as a whole, and therefore not an
1799 * candidate for counter polling. */
1800 port->odp_port = ODPP_NONE;
1801 ofport_update_peer(port);
1805 error = dpif_port_query_by_name(ofproto->backer->dpif,
1806 netdev_vport_get_dpif_port(netdev, namebuf,
1813 port->odp_port = dpif_port.port_no;
1815 if (netdev_get_tunnel_config(netdev)) {
1816 tnl_port_add(port, port->up.netdev, port->odp_port);
1817 port->is_tunnel = true;
1819 /* Sanity-check that a mapping doesn't already exist. This
1820 * shouldn't happen for non-tunnel ports. */
1821 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1822 VLOG_ERR("port %s already has an OpenFlow port number",
1824 dpif_port_destroy(&dpif_port);
1828 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1829 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1830 hash_odp_port(port->odp_port));
1831 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1833 dpif_port_destroy(&dpif_port);
1835 if (ofproto->sflow) {
1836 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1843 port_destruct(struct ofport *port_)
1845 struct ofport_dpif *port = ofport_dpif_cast(port_);
1846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1847 const char *devname = netdev_get_name(port->up.netdev);
1848 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1849 const char *dp_port_name;
1851 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1852 ovs_rwlock_wrlock(&xlate_rwlock);
1853 xlate_ofport_remove(port);
1854 ovs_rwlock_unlock(&xlate_rwlock);
1856 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1858 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1859 /* The underlying device is still there, so delete it. This
1860 * happens when the ofproto is being destroyed, since the caller
1861 * assumes that removal of attached ports will happen as part of
1863 if (!port->is_tunnel) {
1864 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1869 port->peer->peer = NULL;
1873 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1874 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1875 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1876 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1880 sset_find_and_delete(&ofproto->ports, devname);
1881 sset_find_and_delete(&ofproto->ghost_ports, devname);
1882 bundle_remove(port_);
1883 set_cfm(port_, NULL);
1884 set_bfd(port_, NULL);
1885 if (ofproto->sflow) {
1886 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1893 port_modified(struct ofport *port_)
1895 struct ofport_dpif *port = ofport_dpif_cast(port_);
1897 if (port->bundle && port->bundle->bond) {
1898 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1902 cfm_set_netdev(port->cfm, port->up.netdev);
1906 bfd_set_netdev(port->bfd, port->up.netdev);
1909 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1911 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1915 ofport_update_peer(port);
1919 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1921 struct ofport_dpif *port = ofport_dpif_cast(port_);
1922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1923 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1925 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1926 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1927 OFPUTIL_PC_NO_PACKET_IN)) {
1928 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1930 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1931 bundle_update(port->bundle);
1937 set_sflow(struct ofproto *ofproto_,
1938 const struct ofproto_sflow_options *sflow_options)
1940 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1941 struct dpif_sflow *ds = ofproto->sflow;
1943 if (sflow_options) {
1945 struct ofport_dpif *ofport;
1947 ds = ofproto->sflow = dpif_sflow_create();
1948 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1949 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1951 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1953 dpif_sflow_set_options(ds, sflow_options);
1956 dpif_sflow_unref(ds);
1957 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1958 ofproto->sflow = NULL;
1966 struct ofproto *ofproto_,
1967 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1968 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1969 size_t n_flow_exporters_options)
1971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1972 struct dpif_ipfix *di = ofproto->ipfix;
1973 bool has_options = bridge_exporter_options || flow_exporters_options;
1975 if (has_options && !di) {
1976 di = ofproto->ipfix = dpif_ipfix_create();
1980 /* Call set_options in any case to cleanly flush the flow
1981 * caches in the last exporters that are to be destroyed. */
1982 dpif_ipfix_set_options(
1983 di, bridge_exporter_options, flow_exporters_options,
1984 n_flow_exporters_options);
1987 dpif_ipfix_unref(di);
1988 ofproto->ipfix = NULL;
1996 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1998 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2005 struct ofproto_dpif *ofproto;
2007 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2008 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2009 ofport->cfm = cfm_create(ofport->up.netdev);
2012 if (cfm_configure(ofport->cfm, s)) {
2018 cfm_unref(ofport->cfm);
2024 get_cfm_status(const struct ofport *ofport_,
2025 struct ofproto_cfm_status *status)
2027 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2030 status->faults = cfm_get_fault(ofport->cfm);
2031 status->remote_opstate = cfm_get_opup(ofport->cfm);
2032 status->health = cfm_get_health(ofport->cfm);
2033 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2041 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2044 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2048 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2049 cfg, ofport->up.netdev);
2050 if (ofport->bfd != old) {
2051 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2058 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2060 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2063 bfd_get_status(ofport->bfd, smap);
2070 /* Spanning Tree. */
2073 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2075 struct ofproto_dpif *ofproto = ofproto_;
2076 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2077 struct ofport_dpif *ofport;
2079 ofport = stp_port_get_aux(sp);
2081 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2082 ofproto->up.name, port_num);
2084 struct eth_header *eth = pkt->l2;
2086 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2087 if (eth_addr_is_zero(eth->eth_src)) {
2088 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2089 "with unknown MAC", ofproto->up.name, port_num);
2091 send_packet(ofport, pkt);
2097 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2099 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2103 /* Only revalidate flows if the configuration changed. */
2104 if (!s != !ofproto->stp) {
2105 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2109 if (!ofproto->stp) {
2110 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2111 send_bpdu_cb, ofproto);
2112 ofproto->stp_last_tick = time_msec();
2115 stp_set_bridge_id(ofproto->stp, s->system_id);
2116 stp_set_bridge_priority(ofproto->stp, s->priority);
2117 stp_set_hello_time(ofproto->stp, s->hello_time);
2118 stp_set_max_age(ofproto->stp, s->max_age);
2119 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2121 struct ofport *ofport;
2123 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2124 set_stp_port(ofport, NULL);
2127 stp_unref(ofproto->stp);
2128 ofproto->stp = NULL;
2135 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2141 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2142 s->designated_root = stp_get_designated_root(ofproto->stp);
2143 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2152 update_stp_port_state(struct ofport_dpif *ofport)
2154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2155 enum stp_state state;
2157 /* Figure out new state. */
2158 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2162 if (ofport->stp_state != state) {
2163 enum ofputil_port_state of_state;
2166 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2167 netdev_get_name(ofport->up.netdev),
2168 stp_state_name(ofport->stp_state),
2169 stp_state_name(state));
2170 if (stp_learn_in_state(ofport->stp_state)
2171 != stp_learn_in_state(state)) {
2172 /* xxx Learning action flows should also be flushed. */
2173 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2174 mac_learning_flush(ofproto->ml);
2175 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2177 fwd_change = stp_forward_in_state(ofport->stp_state)
2178 != stp_forward_in_state(state);
2180 ofproto->backer->need_revalidate = REV_STP;
2181 ofport->stp_state = state;
2182 ofport->stp_state_entered = time_msec();
2184 if (fwd_change && ofport->bundle) {
2185 bundle_update(ofport->bundle);
2188 /* Update the STP state bits in the OpenFlow port description. */
2189 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2190 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2191 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2192 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2193 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2195 ofproto_port_set_state(&ofport->up, of_state);
2199 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2200 * caller is responsible for assigning STP port numbers and ensuring
2201 * there are no duplicates. */
2203 set_stp_port(struct ofport *ofport_,
2204 const struct ofproto_port_stp_settings *s)
2206 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2208 struct stp_port *sp = ofport->stp_port;
2210 if (!s || !s->enable) {
2212 ofport->stp_port = NULL;
2213 stp_port_disable(sp);
2214 update_stp_port_state(ofport);
2217 } else if (sp && stp_port_no(sp) != s->port_num
2218 && ofport == stp_port_get_aux(sp)) {
2219 /* The port-id changed, so disable the old one if it's not
2220 * already in use by another port. */
2221 stp_port_disable(sp);
2224 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2225 stp_port_enable(sp);
2227 stp_port_set_aux(sp, ofport);
2228 stp_port_set_priority(sp, s->priority);
2229 stp_port_set_path_cost(sp, s->path_cost);
2231 update_stp_port_state(ofport);
2237 get_stp_port_status(struct ofport *ofport_,
2238 struct ofproto_port_stp_status *s)
2240 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2242 struct stp_port *sp = ofport->stp_port;
2244 if (!ofproto->stp || !sp) {
2250 s->port_id = stp_port_get_id(sp);
2251 s->state = stp_port_get_state(sp);
2252 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2253 s->role = stp_port_get_role(sp);
2254 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2260 stp_run(struct ofproto_dpif *ofproto)
2263 long long int now = time_msec();
2264 long long int elapsed = now - ofproto->stp_last_tick;
2265 struct stp_port *sp;
2268 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2269 ofproto->stp_last_tick = now;
2271 while (stp_get_changed_port(ofproto->stp, &sp)) {
2272 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2275 update_stp_port_state(ofport);
2279 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2280 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2281 mac_learning_flush(ofproto->ml);
2282 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2288 stp_wait(struct ofproto_dpif *ofproto)
2291 poll_timer_wait(1000);
2296 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2299 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2302 if (ofport->n_qdscp != n_qdscp
2303 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2304 n_qdscp * sizeof *qdscp))) {
2305 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2306 free(ofport->qdscp);
2307 ofport->qdscp = n_qdscp
2308 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2310 ofport->n_qdscp = n_qdscp;
2318 /* Expires all MAC learning entries associated with 'bundle' and forces its
2319 * ofproto to revalidate every flow.
2321 * Normally MAC learning entries are removed only from the ofproto associated
2322 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2323 * are removed from every ofproto. When patch ports and SLB bonds are in use
2324 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2325 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2326 * with the host from which it migrated. */
2328 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2330 struct ofproto_dpif *ofproto = bundle->ofproto;
2331 struct mac_learning *ml = ofproto->ml;
2332 struct mac_entry *mac, *next_mac;
2334 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2335 ovs_rwlock_wrlock(&ml->rwlock);
2336 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2337 if (mac->port.p == bundle) {
2339 struct ofproto_dpif *o;
2341 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2343 struct mac_entry *e;
2345 ovs_rwlock_wrlock(&o->ml->rwlock);
2346 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2348 mac_learning_expire(o->ml, e);
2350 ovs_rwlock_unlock(&o->ml->rwlock);
2355 mac_learning_expire(ml, mac);
2358 ovs_rwlock_unlock(&ml->rwlock);
2361 static struct ofbundle *
2362 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2364 struct ofbundle *bundle;
2366 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2367 &ofproto->bundles) {
2368 if (bundle->aux == aux) {
2376 bundle_update(struct ofbundle *bundle)
2378 struct ofport_dpif *port;
2380 bundle->floodable = true;
2381 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2382 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2383 || !stp_forward_in_state(port->stp_state)) {
2384 bundle->floodable = false;
2391 bundle_del_port(struct ofport_dpif *port)
2393 struct ofbundle *bundle = port->bundle;
2395 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2397 list_remove(&port->bundle_node);
2398 port->bundle = NULL;
2401 lacp_slave_unregister(bundle->lacp, port);
2404 bond_slave_unregister(bundle->bond, port);
2407 bundle_update(bundle);
2411 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2412 struct lacp_slave_settings *lacp)
2414 struct ofport_dpif *port;
2416 port = get_ofp_port(bundle->ofproto, ofp_port);
2421 if (port->bundle != bundle) {
2422 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2424 bundle_remove(&port->up);
2427 port->bundle = bundle;
2428 list_push_back(&bundle->ports, &port->bundle_node);
2429 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2430 || !stp_forward_in_state(port->stp_state)) {
2431 bundle->floodable = false;
2435 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2436 lacp_slave_register(bundle->lacp, port, lacp);
2443 bundle_destroy(struct ofbundle *bundle)
2445 struct ofproto_dpif *ofproto;
2446 struct ofport_dpif *port, *next_port;
2452 ofproto = bundle->ofproto;
2453 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2455 ovs_rwlock_wrlock(&xlate_rwlock);
2456 xlate_bundle_remove(bundle);
2457 ovs_rwlock_unlock(&xlate_rwlock);
2459 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2460 bundle_del_port(port);
2463 bundle_flush_macs(bundle, true);
2464 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2466 free(bundle->trunks);
2467 lacp_unref(bundle->lacp);
2468 bond_unref(bundle->bond);
2473 bundle_set(struct ofproto *ofproto_, void *aux,
2474 const struct ofproto_bundle_settings *s)
2476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2477 bool need_flush = false;
2478 struct ofport_dpif *port;
2479 struct ofbundle *bundle;
2480 unsigned long *trunks;
2486 bundle_destroy(bundle_lookup(ofproto, aux));
2490 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2491 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2493 bundle = bundle_lookup(ofproto, aux);
2495 bundle = xmalloc(sizeof *bundle);
2497 bundle->ofproto = ofproto;
2498 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2499 hash_pointer(aux, 0));
2501 bundle->name = NULL;
2503 list_init(&bundle->ports);
2504 bundle->vlan_mode = PORT_VLAN_TRUNK;
2506 bundle->trunks = NULL;
2507 bundle->use_priority_tags = s->use_priority_tags;
2508 bundle->lacp = NULL;
2509 bundle->bond = NULL;
2511 bundle->floodable = true;
2512 mbridge_register_bundle(ofproto->mbridge, bundle);
2515 if (!bundle->name || strcmp(s->name, bundle->name)) {
2517 bundle->name = xstrdup(s->name);
2522 if (!bundle->lacp) {
2523 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2524 bundle->lacp = lacp_create();
2526 lacp_configure(bundle->lacp, s->lacp);
2528 lacp_unref(bundle->lacp);
2529 bundle->lacp = NULL;
2532 /* Update set of ports. */
2534 for (i = 0; i < s->n_slaves; i++) {
2535 if (!bundle_add_port(bundle, s->slaves[i],
2536 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2540 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2541 struct ofport_dpif *next_port;
2543 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2544 for (i = 0; i < s->n_slaves; i++) {
2545 if (s->slaves[i] == port->up.ofp_port) {
2550 bundle_del_port(port);
2554 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2556 if (list_is_empty(&bundle->ports)) {
2557 bundle_destroy(bundle);
2561 /* Set VLAN tagging mode */
2562 if (s->vlan_mode != bundle->vlan_mode
2563 || s->use_priority_tags != bundle->use_priority_tags) {
2564 bundle->vlan_mode = s->vlan_mode;
2565 bundle->use_priority_tags = s->use_priority_tags;
2570 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2571 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2573 if (vlan != bundle->vlan) {
2574 bundle->vlan = vlan;
2578 /* Get trunked VLANs. */
2579 switch (s->vlan_mode) {
2580 case PORT_VLAN_ACCESS:
2584 case PORT_VLAN_TRUNK:
2585 trunks = CONST_CAST(unsigned long *, s->trunks);
2588 case PORT_VLAN_NATIVE_UNTAGGED:
2589 case PORT_VLAN_NATIVE_TAGGED:
2590 if (vlan != 0 && (!s->trunks
2591 || !bitmap_is_set(s->trunks, vlan)
2592 || bitmap_is_set(s->trunks, 0))) {
2593 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2595 trunks = bitmap_clone(s->trunks, 4096);
2597 trunks = bitmap_allocate1(4096);
2599 bitmap_set1(trunks, vlan);
2600 bitmap_set0(trunks, 0);
2602 trunks = CONST_CAST(unsigned long *, s->trunks);
2609 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2610 free(bundle->trunks);
2611 if (trunks == s->trunks) {
2612 bundle->trunks = vlan_bitmap_clone(trunks);
2614 bundle->trunks = trunks;
2619 if (trunks != s->trunks) {
2624 if (!list_is_short(&bundle->ports)) {
2625 bundle->ofproto->has_bonded_bundles = true;
2627 if (bond_reconfigure(bundle->bond, s->bond)) {
2628 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2631 bundle->bond = bond_create(s->bond);
2632 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2635 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2636 bond_slave_register(bundle->bond, port, port->up.netdev);
2639 bond_unref(bundle->bond);
2640 bundle->bond = NULL;
2643 /* If we changed something that would affect MAC learning, un-learn
2644 * everything on this port and force flow revalidation. */
2646 bundle_flush_macs(bundle, false);
2653 bundle_remove(struct ofport *port_)
2655 struct ofport_dpif *port = ofport_dpif_cast(port_);
2656 struct ofbundle *bundle = port->bundle;
2659 bundle_del_port(port);
2660 if (list_is_empty(&bundle->ports)) {
2661 bundle_destroy(bundle);
2662 } else if (list_is_short(&bundle->ports)) {
2663 bond_unref(bundle->bond);
2664 bundle->bond = NULL;
2670 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2672 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2673 struct ofport_dpif *port = port_;
2674 uint8_t ea[ETH_ADDR_LEN];
2677 error = netdev_get_etheraddr(port->up.netdev, ea);
2679 struct ofpbuf packet;
2682 ofpbuf_init(&packet, 0);
2683 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2685 memcpy(packet_pdu, pdu, pdu_size);
2687 send_packet(port, &packet);
2688 ofpbuf_uninit(&packet);
2690 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2691 "%s (%s)", port->bundle->name,
2692 netdev_get_name(port->up.netdev), ovs_strerror(error));
2697 bundle_send_learning_packets(struct ofbundle *bundle)
2699 struct ofproto_dpif *ofproto = bundle->ofproto;
2700 struct ofpbuf *learning_packet;
2701 int error, n_packets, n_errors;
2702 struct mac_entry *e;
2703 struct list packets;
2705 list_init(&packets);
2706 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2707 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2708 if (e->port.p != bundle) {
2711 learning_packet = bond_compose_learning_packet(bundle->bond,
2714 learning_packet->private_p = port_void;
2715 list_push_back(&packets, &learning_packet->list_node);
2718 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2720 error = n_packets = n_errors = 0;
2721 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2724 ret = send_packet(learning_packet->private_p, learning_packet);
2731 ofpbuf_list_delete(&packets);
2734 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2735 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2736 "packets, last error was: %s",
2737 bundle->name, n_errors, n_packets, ovs_strerror(error));
2739 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2740 bundle->name, n_packets);
2745 bundle_run(struct ofbundle *bundle)
2748 lacp_run(bundle->lacp, send_pdu_cb);
2751 struct ofport_dpif *port;
2753 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2754 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2757 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2758 bundle->ofproto->backer->need_revalidate = REV_BOND;
2761 if (bond_should_send_learning_packets(bundle->bond)) {
2762 bundle_send_learning_packets(bundle);
2768 bundle_wait(struct ofbundle *bundle)
2771 lacp_wait(bundle->lacp);
2774 bond_wait(bundle->bond);
2781 mirror_set__(struct ofproto *ofproto_, void *aux,
2782 const struct ofproto_mirror_settings *s)
2784 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2785 struct ofbundle **srcs, **dsts;
2790 mirror_destroy(ofproto->mbridge, aux);
2794 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2795 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2797 for (i = 0; i < s->n_srcs; i++) {
2798 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2801 for (i = 0; i < s->n_dsts; i++) {
2802 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2805 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2806 s->n_dsts, s->src_vlans,
2807 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2814 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2815 uint64_t *packets, uint64_t *bytes)
2818 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2823 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2825 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2826 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2827 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2828 mac_learning_flush(ofproto->ml);
2830 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2835 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2838 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2839 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2843 forward_bpdu_changed(struct ofproto *ofproto_)
2845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2846 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2850 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2854 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2855 mac_learning_set_idle_time(ofproto->ml, idle_time);
2856 mac_learning_set_max_entries(ofproto->ml, max_entries);
2857 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2862 static struct ofport_dpif *
2863 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2865 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2866 return ofport ? ofport_dpif_cast(ofport) : NULL;
2869 static struct ofport_dpif *
2870 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2872 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2873 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2877 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2878 struct ofproto_port *ofproto_port,
2879 struct dpif_port *dpif_port)
2881 ofproto_port->name = dpif_port->name;
2882 ofproto_port->type = dpif_port->type;
2883 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2887 ofport_update_peer(struct ofport_dpif *ofport)
2889 const struct ofproto_dpif *ofproto;
2890 struct dpif_backer *backer;
2893 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2897 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2898 backer->need_revalidate = REV_RECONFIGURE;
2901 ofport->peer->peer = NULL;
2902 ofport->peer = NULL;
2905 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2910 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2911 struct ofport *peer_ofport;
2912 struct ofport_dpif *peer;
2915 if (ofproto->backer != backer) {
2919 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2924 peer = ofport_dpif_cast(peer_ofport);
2925 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2926 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2928 ofport->peer = peer;
2929 ofport->peer->peer = ofport;
2939 port_run_fast(struct ofport_dpif *ofport)
2941 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2942 struct ofpbuf packet;
2944 ofpbuf_init(&packet, 0);
2945 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2946 send_packet(ofport, &packet);
2947 ofpbuf_uninit(&packet);
2950 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2951 struct ofpbuf packet;
2953 ofpbuf_init(&packet, 0);
2954 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2955 send_packet(ofport, &packet);
2956 ofpbuf_uninit(&packet);
2961 port_run(struct ofport_dpif *ofport)
2963 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2964 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2965 bool enable = netdev_get_carrier(ofport->up.netdev);
2966 bool cfm_enable = false;
2967 bool bfd_enable = false;
2969 ofport->carrier_seq = carrier_seq;
2971 port_run_fast(ofport);
2974 int cfm_opup = cfm_get_opup(ofport->cfm);
2976 cfm_run(ofport->cfm);
2977 cfm_enable = !cfm_get_fault(ofport->cfm);
2979 if (cfm_opup >= 0) {
2980 cfm_enable = cfm_enable && cfm_opup;
2985 bfd_run(ofport->bfd);
2986 bfd_enable = bfd_forwarding(ofport->bfd);
2989 if (ofport->bfd || ofport->cfm) {
2990 enable = enable && (cfm_enable || bfd_enable);
2993 if (ofport->bundle) {
2994 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2995 if (carrier_changed) {
2996 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3000 if (ofport->may_enable != enable) {
3001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3002 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3005 ofport->may_enable = enable;
3009 port_wait(struct ofport_dpif *ofport)
3012 cfm_wait(ofport->cfm);
3016 bfd_wait(ofport->bfd);
3021 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3022 struct ofproto_port *ofproto_port)
3024 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3025 struct dpif_port dpif_port;
3028 if (sset_contains(&ofproto->ghost_ports, devname)) {
3029 const char *type = netdev_get_type_from_name(devname);
3031 /* We may be called before ofproto->up.port_by_name is populated with
3032 * the appropriate ofport. For this reason, we must get the name and
3033 * type from the netdev layer directly. */
3035 const struct ofport *ofport;
3037 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3038 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3039 ofproto_port->name = xstrdup(devname);
3040 ofproto_port->type = xstrdup(type);
3046 if (!sset_contains(&ofproto->ports, devname)) {
3049 error = dpif_port_query_by_name(ofproto->backer->dpif,
3050 devname, &dpif_port);
3052 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3058 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3061 const char *devname = netdev_get_name(netdev);
3062 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3063 const char *dp_port_name;
3065 if (netdev_vport_is_patch(netdev)) {
3066 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3070 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3071 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3072 odp_port_t port_no = ODPP_NONE;
3075 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3079 if (netdev_get_tunnel_config(netdev)) {
3080 simap_put(&ofproto->backer->tnl_backers,
3081 dp_port_name, odp_to_u32(port_no));
3085 if (netdev_get_tunnel_config(netdev)) {
3086 sset_add(&ofproto->ghost_ports, devname);
3088 sset_add(&ofproto->ports, devname);
3094 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3097 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3104 sset_find_and_delete(&ofproto->ghost_ports,
3105 netdev_get_name(ofport->up.netdev));
3106 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3107 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3108 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3110 /* The caller is going to close ofport->up.netdev. If this is a
3111 * bonded port, then the bond is using that netdev, so remove it
3112 * from the bond. The client will need to reconfigure everything
3113 * after deleting ports, so then the slave will get re-added. */
3114 bundle_remove(&ofport->up);
3121 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3123 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3128 error = netdev_get_stats(ofport->up.netdev, stats);
3130 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3133 /* ofproto->stats.tx_packets represents packets that we created
3134 * internally and sent to some port (e.g. packets sent with
3135 * send_packet()). Account for them as if they had come from
3136 * OFPP_LOCAL and got forwarded. */
3138 if (stats->rx_packets != UINT64_MAX) {
3139 stats->rx_packets += ofproto->stats.tx_packets;
3142 if (stats->rx_bytes != UINT64_MAX) {
3143 stats->rx_bytes += ofproto->stats.tx_bytes;
3146 /* ofproto->stats.rx_packets represents packets that were received on
3147 * some port and we processed internally and dropped (e.g. STP).
3148 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3150 if (stats->tx_packets != UINT64_MAX) {
3151 stats->tx_packets += ofproto->stats.rx_packets;
3154 if (stats->tx_bytes != UINT64_MAX) {
3155 stats->tx_bytes += ofproto->stats.rx_bytes;
3162 struct port_dump_state {
3167 struct ofproto_port port;
3172 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3174 *statep = xzalloc(sizeof(struct port_dump_state));
3179 port_dump_next(const struct ofproto *ofproto_, void *state_,
3180 struct ofproto_port *port)
3182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3183 struct port_dump_state *state = state_;
3184 const struct sset *sset;
3185 struct sset_node *node;
3187 if (state->has_port) {
3188 ofproto_port_destroy(&state->port);
3189 state->has_port = false;
3191 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3192 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3195 error = port_query_by_name(ofproto_, node->name, &state->port);
3197 *port = state->port;
3198 state->has_port = true;
3200 } else if (error != ENODEV) {
3205 if (!state->ghost) {
3206 state->ghost = true;
3209 return port_dump_next(ofproto_, state_, port);
3216 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3218 struct port_dump_state *state = state_;
3220 if (state->has_port) {
3221 ofproto_port_destroy(&state->port);
3228 port_poll(const struct ofproto *ofproto_, char **devnamep)
3230 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3232 if (ofproto->port_poll_errno) {
3233 int error = ofproto->port_poll_errno;
3234 ofproto->port_poll_errno = 0;
3238 if (sset_is_empty(&ofproto->port_poll_set)) {
3242 *devnamep = sset_pop(&ofproto->port_poll_set);
3247 port_poll_wait(const struct ofproto *ofproto_)
3249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3250 dpif_port_poll_wait(ofproto->backer->dpif);
3254 port_is_lacp_current(const struct ofport *ofport_)
3256 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3257 return (ofport->bundle && ofport->bundle->lacp
3258 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3262 /* Upcall handling. */
3264 struct flow_miss_op {
3265 struct dpif_op dpif_op;
3267 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3268 struct xlate_out xout;
3269 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3271 struct ofpbuf mask; /* Flow mask for "put" ops. */
3272 struct odputil_keybuf maskbuf;
3274 /* If this is a "put" op, then a pointer to the subfacet that should
3275 * be marked as uninstalled if the operation fails. */
3276 struct subfacet *subfacet;
3279 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3280 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3281 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3282 * return value of true). However, for short flows the cost of bookkeeping is
3283 * much higher than the benefits, so when the datapath holds a large number of
3284 * flows we impose some heuristics to decide which flows are likely to be worth
3287 flow_miss_should_make_facet(struct flow_miss *miss)
3289 struct dpif_backer *backer = miss->ofproto->backer;
3292 switch (flow_miss_model) {
3293 case OFPROTO_HANDLE_MISS_AUTO:
3295 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3297 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3301 if (!backer->governor) {
3304 n_subfacets = hmap_count(&backer->subfacets);
3305 if (n_subfacets * 2 <= flow_eviction_threshold) {
3309 backer->governor = governor_create();
3312 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3313 return governor_should_install_flow(backer->governor, hash,
3314 list_size(&miss->packets));
3317 /* Handles 'miss', which matches 'facet'. May add any required datapath
3318 * operations to 'ops', incrementing '*n_ops' for each new op.
3320 * All of the packets in 'miss' are considered to have arrived at time
3321 * 'miss->stats.used'. This is really important only for new facets: if we
3322 * just called time_msec() here, then the new subfacet or its packets could
3323 * look (occasionally) as though it was used some time after the facet was
3324 * used. That can make a one-packet flow look like it has a nonzero duration,
3325 * which looks odd in e.g. NetFlow statistics. */
3327 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3328 struct flow_miss_op *ops, size_t *n_ops)
3330 enum subfacet_path want_path;
3331 struct subfacet *subfacet;
3333 facet->packet_count += miss->stats.n_packets;
3334 facet->prev_packet_count += miss->stats.n_packets;
3335 facet->byte_count += miss->stats.n_bytes;
3336 facet->prev_byte_count += miss->stats.n_bytes;
3338 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3340 /* Don't install the flow if it's the result of the "userspace"
3341 * action for an already installed facet. This can occur when a
3342 * datapath flow with wildcards has a "userspace" action and flows
3343 * sent to userspace result in a different subfacet, which will then
3344 * be rejected as overlapping by the datapath. */
3345 if (miss->upcall_type == DPIF_UC_ACTION
3346 && !list_is_empty(&facet->subfacets)) {
3350 subfacet = subfacet_create(facet, miss);
3351 if (subfacet->path != want_path) {
3352 struct flow_miss_op *op = &ops[(*n_ops)++];
3353 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3355 subfacet->path = want_path;
3357 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3358 if (enable_megaflows) {
3359 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3360 &miss->flow, UINT32_MAX);
3363 op->xout_garbage = false;
3364 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3365 op->subfacet = subfacet;
3366 put->flags = DPIF_FP_CREATE;
3367 put->key = miss->key;
3368 put->key_len = miss->key_len;
3369 put->mask = op->mask.data;
3370 put->mask_len = op->mask.size;
3372 if (want_path == SF_FAST_PATH) {
3373 put->actions = facet->xout.odp_actions.data;
3374 put->actions_len = facet->xout.odp_actions.size;
3376 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3377 op->slow_stub, sizeof op->slow_stub,
3378 &put->actions, &put->actions_len);
3384 /* Handles flow miss 'miss'. May add any required datapath operations
3385 * to 'ops', incrementing '*n_ops' for each new op. */
3387 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3390 struct facet *facet;
3392 miss->ofproto->n_missed += list_size(&miss->packets);
3394 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3396 /* There does not exist a bijection between 'struct flow' and datapath
3397 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3398 * assumption used throughout the facet and subfacet handling code.
3399 * Since we have to handle these misses in userspace anyway, we simply
3400 * skip facet creation, avoiding the problem altogether. */
3401 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3402 || !flow_miss_should_make_facet(miss)) {
3406 facet = facet_create(miss);
3408 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3411 static struct drop_key *
3412 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3415 struct drop_key *drop_key;
3417 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3418 &backer->drop_keys) {
3419 if (drop_key->key_len == key_len
3420 && !memcmp(drop_key->key, key, key_len)) {
3428 drop_key_clear(struct dpif_backer *backer)
3430 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3431 struct drop_key *drop_key, *next;
3433 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3436 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3438 if (error && !VLOG_DROP_WARN(&rl)) {
3439 struct ds ds = DS_EMPTY_INITIALIZER;
3440 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3441 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3442 ovs_strerror(error), ds_cstr(&ds));
3446 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3447 drop_key_destroy(drop_key);
3450 udpif_drop_key_clear(backer->udpif);
3454 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3456 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3457 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3458 struct flow_miss *miss;
3461 /* Process each element in the to-do list, constructing the set of
3462 * operations to batch. */
3464 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3465 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3467 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3469 /* Execute batch. */
3470 for (i = 0; i < n_ops; i++) {
3471 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3473 dpif_operate(backer->dpif, dpif_ops, n_ops);
3475 for (i = 0; i < n_ops; i++) {
3476 if (dpif_ops[i]->error != 0
3477 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3478 && flow_miss_ops[i].subfacet) {
3479 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3481 COVERAGE_INC(subfacet_install_fail);
3483 /* Zero-out subfacet counters when installation failed, but
3484 * datapath reported hits. This should not happen and
3485 * indicates a bug, since if the datapath flow exists, we
3486 * should not be attempting to create a new subfacet. A
3487 * buggy datapath could trigger this, so just zero out the
3488 * counters and log an error. */
3489 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3490 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3491 "datapath reported hits");
3492 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3495 subfacet->path = SF_NOT_INSTALLED;
3501 handle_sflow_upcall(struct dpif_backer *backer,
3502 const struct dpif_upcall *upcall)
3504 struct ofproto_dpif *ofproto;
3505 union user_action_cookie cookie;
3507 odp_port_t odp_in_port;
3509 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3510 &flow, NULL, &ofproto, &odp_in_port)
3511 || !ofproto->sflow) {
3515 memset(&cookie, 0, sizeof cookie);
3516 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3517 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3518 odp_in_port, &cookie);
3522 handle_flow_sample_upcall(struct dpif_backer *backer,
3523 const struct dpif_upcall *upcall)
3525 struct ofproto_dpif *ofproto;
3526 union user_action_cookie cookie;
3529 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3530 &flow, NULL, &ofproto, NULL)
3531 || !ofproto->ipfix) {
3535 memset(&cookie, 0, sizeof cookie);
3536 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3538 /* The flow reflects exactly the contents of the packet. Sample
3539 * the packet using it. */
3540 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3541 cookie.flow_sample.collector_set_id,
3542 cookie.flow_sample.probability,
3543 cookie.flow_sample.obs_domain_id,
3544 cookie.flow_sample.obs_point_id);
3548 handle_ipfix_upcall(struct dpif_backer *backer,
3549 const struct dpif_upcall *upcall)
3551 struct ofproto_dpif *ofproto;
3554 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3555 &flow, NULL, &ofproto, NULL)
3556 || !ofproto->ipfix) {
3560 /* The flow reflects exactly the contents of the packet. Sample
3561 * the packet using it. */
3562 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3566 handle_upcalls(struct dpif_backer *backer)
3568 struct flow_miss_batch *fmb;
3571 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3572 struct upcall *upcall = upcall_next(backer->udpif);
3578 switch (upcall->type) {
3580 handle_sflow_upcall(backer, &upcall->dpif_upcall);
3583 case FLOW_SAMPLE_UPCALL:
3584 handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3588 handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3598 upcall_destroy(upcall);
3601 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3602 struct drop_key *drop_key = drop_key_next(backer->udpif);
3607 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3608 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3609 hash_bytes(drop_key->key, drop_key->key_len, 0));
3610 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3611 drop_key->key, drop_key->key_len,
3612 NULL, 0, NULL, 0, NULL);
3614 drop_key_destroy(drop_key);
3618 fmb = flow_miss_batch_next(backer->udpif);
3620 handle_flow_misses(backer, fmb);
3621 flow_miss_batch_destroy(fmb);
3625 /* Flow expiration. */
3627 static int subfacet_max_idle(const struct dpif_backer *);
3628 static void update_stats(struct dpif_backer *);
3629 static void rule_expire(struct rule_dpif *);
3630 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3632 /* This function is called periodically by run(). Its job is to collect
3633 * updates for the flows that have been installed into the datapath, most
3634 * importantly when they last were used, and then use that information to
3635 * expire flows that have not been used recently.
3637 * Returns the number of milliseconds after which it should be called again. */
3639 expire(struct dpif_backer *backer)
3641 struct ofproto_dpif *ofproto;
3645 /* Periodically clear out the drop keys in an effort to keep them
3646 * relatively few. */
3647 drop_key_clear(backer);
3649 /* Update stats for each flow in the backer. */
3650 update_stats(backer);
3652 n_subfacets = hmap_count(&backer->subfacets);
3654 struct subfacet *subfacet;
3655 long long int total, now;
3659 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3660 total += now - subfacet->created;
3662 backer->avg_subfacet_life += total / n_subfacets;
3664 backer->avg_subfacet_life /= 2;
3666 backer->avg_n_subfacet += n_subfacets;
3667 backer->avg_n_subfacet /= 2;
3669 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3671 max_idle = subfacet_max_idle(backer);
3672 expire_subfacets(backer, max_idle);
3674 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3675 struct rule *rule, *next_rule;
3677 if (ofproto->backer != backer) {
3681 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3683 ovs_mutex_lock(&ofproto->up.expirable_mutex);
3684 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3685 &ofproto->up.expirable) {
3686 rule_expire(rule_dpif_cast(rule));
3688 ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3690 /* All outstanding data in existing flows has been accounted, so it's a
3691 * good time to do bond rebalancing. */
3692 if (ofproto->has_bonded_bundles) {
3693 struct ofbundle *bundle;
3695 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3697 bond_rebalance(bundle->bond);
3703 return MIN(max_idle, 1000);
3706 /* Updates flow table statistics given that the datapath just reported 'stats'
3707 * as 'subfacet''s statistics. */
3709 update_subfacet_stats(struct subfacet *subfacet,
3710 const struct dpif_flow_stats *stats)
3712 struct facet *facet = subfacet->facet;
3713 struct dpif_flow_stats diff;
3715 diff.tcp_flags = stats->tcp_flags;
3716 diff.used = stats->used;
3718 if (stats->n_packets >= subfacet->dp_packet_count) {
3719 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3721 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3725 if (stats->n_bytes >= subfacet->dp_byte_count) {
3726 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3728 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3732 facet->ofproto->n_hit += diff.n_packets;
3733 subfacet->dp_packet_count = stats->n_packets;
3734 subfacet->dp_byte_count = stats->n_bytes;
3735 subfacet_update_stats(subfacet, &diff);
3737 if (facet->accounted_bytes < facet->byte_count) {
3739 facet_account(facet);
3740 facet->accounted_bytes = facet->byte_count;
3744 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3745 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3747 delete_unexpected_flow(struct dpif_backer *backer,
3748 const struct nlattr *key, size_t key_len)
3750 if (!VLOG_DROP_WARN(&rl)) {
3754 odp_flow_key_format(key, key_len, &s);
3755 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3759 COVERAGE_INC(facet_unexpected);
3760 dpif_flow_del(backer->dpif, key, key_len, NULL);
3763 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3765 * This function also pushes statistics updates to rules which each facet
3766 * resubmits into. Generally these statistics will be accurate. However, if a
3767 * facet changes the rule it resubmits into at some time in between
3768 * update_stats() runs, it is possible that statistics accrued to the
3769 * old rule will be incorrectly attributed to the new rule. This could be
3770 * avoided by calling update_stats() whenever rules are created or
3771 * deleted. However, the performance impact of making so many calls to the
3772 * datapath do not justify the benefit of having perfectly accurate statistics.
3774 * In addition, this function maintains per ofproto flow hit counts. The patch
3775 * port is not treated specially. e.g. A packet ingress from br0 patched into
3776 * br1 will increase the hit count of br0 by 1, however, does not affect
3777 * the hit or miss counts of br1.
3780 update_stats(struct dpif_backer *backer)
3782 const struct dpif_flow_stats *stats;
3783 struct dpif_flow_dump dump;
3784 const struct nlattr *key, *mask;
3785 size_t key_len, mask_len;
3787 dpif_flow_dump_start(&dump, backer->dpif);
3788 while (dpif_flow_dump_next(&dump, &key, &key_len,
3789 &mask, &mask_len, NULL, NULL, &stats)) {
3790 struct subfacet *subfacet;
3793 key_hash = odp_flow_key_hash(key, key_len);
3794 subfacet = subfacet_find(backer, key, key_len, key_hash);
3795 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3797 update_subfacet_stats(subfacet, stats);
3801 /* Stats are updated per-packet. */
3804 case SF_NOT_INSTALLED:
3806 delete_unexpected_flow(backer, key, key_len);
3811 dpif_flow_dump_done(&dump);
3813 update_moving_averages(backer);
3816 /* Calculates and returns the number of milliseconds of idle time after which
3817 * subfacets should expire from the datapath. When a subfacet expires, we fold
3818 * its statistics into its facet, and when a facet's last subfacet expires, we
3819 * fold its statistic into its rule. */
3821 subfacet_max_idle(const struct dpif_backer *backer)
3824 * Idle time histogram.
3826 * Most of the time a switch has a relatively small number of subfacets.
3827 * When this is the case we might as well keep statistics for all of them
3828 * in userspace and to cache them in the kernel datapath for performance as
3831 * As the number of subfacets increases, the memory required to maintain
3832 * statistics about them in userspace and in the kernel becomes
3833 * significant. However, with a large number of subfacets it is likely
3834 * that only a few of them are "heavy hitters" that consume a large amount
3835 * of bandwidth. At this point, only heavy hitters are worth caching in
3836 * the kernel and maintaining in userspaces; other subfacets we can
3839 * The technique used to compute the idle time is to build a histogram with
3840 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3841 * that is installed in the kernel gets dropped in the appropriate bucket.
3842 * After the histogram has been built, we compute the cutoff so that only
3843 * the most-recently-used 1% of subfacets (but at least
3844 * flow_eviction_threshold flows) are kept cached. At least
3845 * the most-recently-used bucket of subfacets is kept, so actually an
3846 * arbitrary number of subfacets can be kept in any given expiration run
3847 * (though the next run will delete most of those unless they receive
3850 * This requires a second pass through the subfacets, in addition to the
3851 * pass made by update_stats(), because the former function never looks at
3852 * uninstallable subfacets.
3854 enum { BUCKET_WIDTH = 100 };
3855 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3856 int buckets[N_BUCKETS] = { 0 };
3857 int total, subtotal, bucket;
3858 struct subfacet *subfacet;
3862 total = hmap_count(&backer->subfacets);
3863 if (total <= flow_eviction_threshold) {
3864 return N_BUCKETS * BUCKET_WIDTH;
3867 /* Build histogram. */
3869 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3870 long long int idle = now - subfacet->used;
3871 int bucket = (idle <= 0 ? 0
3872 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3873 : (unsigned int) idle / BUCKET_WIDTH);
3877 /* Find the first bucket whose flows should be expired. */
3878 subtotal = bucket = 0;
3880 subtotal += buckets[bucket++];
3881 } while (bucket < N_BUCKETS &&
3882 subtotal < MAX(flow_eviction_threshold, total / 100));
3884 if (VLOG_IS_DBG_ENABLED()) {
3888 ds_put_cstr(&s, "keep");
3889 for (i = 0; i < N_BUCKETS; i++) {
3891 ds_put_cstr(&s, ", drop");
3894 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3897 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3901 return bucket * BUCKET_WIDTH;
3905 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3907 /* Cutoff time for most flows. */
3908 long long int normal_cutoff = time_msec() - dp_max_idle;
3910 /* We really want to keep flows for special protocols around, so use a more
3911 * conservative cutoff. */
3912 long long int special_cutoff = time_msec() - 10000;
3914 struct subfacet *subfacet, *next_subfacet;
3915 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3919 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3920 &backer->subfacets) {
3921 long long int cutoff;
3923 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3927 if (subfacet->used < cutoff) {
3928 if (subfacet->path != SF_NOT_INSTALLED) {
3929 batch[n_batch++] = subfacet;
3930 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3931 subfacet_destroy_batch(backer, batch, n_batch);
3935 subfacet_destroy(subfacet);
3941 subfacet_destroy_batch(backer, batch, n_batch);
3945 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3946 * then delete it entirely. */
3948 rule_expire(struct rule_dpif *rule)
3950 uint16_t idle_timeout, hard_timeout;
3954 ovs_assert(!rule->up.pending);
3956 ovs_mutex_lock(&rule->up.timeout_mutex);
3957 hard_timeout = rule->up.hard_timeout;
3958 idle_timeout = rule->up.idle_timeout;
3959 ovs_mutex_unlock(&rule->up.timeout_mutex);
3961 /* Has 'rule' expired? */
3963 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3964 reason = OFPRR_HARD_TIMEOUT;
3965 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3966 reason = OFPRR_IDLE_TIMEOUT;
3971 COVERAGE_INC(ofproto_dpif_expired);
3972 ofproto_rule_expire(&rule->up, reason);
3977 /* Creates and returns a new facet based on 'miss'.
3979 * The caller must already have determined that no facet with an identical
3980 * 'miss->flow' exists in 'miss->ofproto'.
3982 * 'rule' and 'xout' must have been created based on 'miss'.
3984 * 'facet'' statistics are initialized based on 'stats'.
3986 * The facet will initially have no subfacets. The caller should create (at
3987 * least) one subfacet with subfacet_create(). */
3988 static struct facet *
3989 facet_create(const struct flow_miss *miss)
3991 struct ofproto_dpif *ofproto = miss->ofproto;
3992 struct facet *facet;
3995 facet = xzalloc(sizeof *facet);
3996 facet->ofproto = miss->ofproto;
3997 facet->used = miss->stats.used;
3998 facet->flow = miss->flow;
3999 facet->learn_rl = time_msec() + 500;
4001 list_init(&facet->subfacets);
4002 netflow_flow_init(&facet->nf_flow);
4003 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4005 xlate_out_copy(&facet->xout, &miss->xout);
4007 match_init(&match, &facet->flow, &facet->xout.wc);
4008 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4009 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4010 classifier_insert(&ofproto->facets, &facet->cr);
4011 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4013 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4018 facet_free(struct facet *facet)
4021 xlate_out_uninit(&facet->xout);
4026 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4027 * 'packet', which arrived on 'in_port'. */
4029 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4030 const struct nlattr *odp_actions, size_t actions_len,
4031 struct ofpbuf *packet)
4033 struct odputil_keybuf keybuf;
4037 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4038 odp_flow_key_from_flow(&key, flow,
4039 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4041 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4042 odp_actions, actions_len, packet);
4046 /* Remove 'facet' from its ofproto and free up the associated memory:
4048 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4049 * rule's statistics, via subfacet_uninstall().
4051 * - Removes 'facet' from its rule and from ofproto->facets.
4054 facet_remove(struct facet *facet)
4056 struct subfacet *subfacet, *next_subfacet;
4058 ovs_assert(!list_is_empty(&facet->subfacets));
4060 /* First uninstall all of the subfacets to get final statistics. */
4061 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4062 subfacet_uninstall(subfacet);
4065 /* Flush the final stats to the rule.
4067 * This might require us to have at least one subfacet around so that we
4068 * can use its actions for accounting in facet_account(), which is why we
4069 * have uninstalled but not yet destroyed the subfacets. */
4070 facet_flush_stats(facet);
4072 /* Now we're really all done so destroy everything. */
4073 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4074 &facet->subfacets) {
4075 subfacet_destroy__(subfacet);
4077 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4078 classifier_remove(&facet->ofproto->facets, &facet->cr);
4079 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4080 cls_rule_destroy(&facet->cr);
4084 /* Feed information from 'facet' back into the learning table to keep it in
4085 * sync with what is actually flowing through the datapath. */
4087 facet_learn(struct facet *facet)
4089 long long int now = time_msec();
4091 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4095 facet->learn_rl = now + 500;
4097 if (!facet->xout.has_learn
4098 && !facet->xout.has_normal
4099 && (!facet->xout.has_fin_timeout
4100 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4104 facet_push_stats(facet, true);
4108 facet_account(struct facet *facet)
4110 const struct nlattr *a;
4115 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4118 n_bytes = facet->byte_count - facet->accounted_bytes;
4120 /* This loop feeds byte counters to bond_account() for rebalancing to use
4121 * as a basis. We also need to track the actual VLAN on which the packet
4122 * is going to be sent to ensure that it matches the one passed to
4123 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4126 * We use the actions from an arbitrary subfacet because they should all
4127 * be equally valid for our purpose. */
4128 vlan_tci = facet->flow.vlan_tci;
4129 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4130 facet->xout.odp_actions.size) {
4131 const struct ovs_action_push_vlan *vlan;
4132 struct ofport_dpif *port;
4134 switch (nl_attr_type(a)) {
4135 case OVS_ACTION_ATTR_OUTPUT:
4136 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4137 if (port && port->bundle && port->bundle->bond) {
4138 bond_account(port->bundle->bond, &facet->flow,
4139 vlan_tci_to_vid(vlan_tci), n_bytes);
4143 case OVS_ACTION_ATTR_POP_VLAN:
4144 vlan_tci = htons(0);
4147 case OVS_ACTION_ATTR_PUSH_VLAN:
4148 vlan = nl_attr_get(a);
4149 vlan_tci = vlan->vlan_tci;
4155 /* Returns true if the only action for 'facet' is to send to the controller.
4156 * (We don't report NetFlow expiration messages for such facets because they
4157 * are just part of the control logic for the network, not real traffic). */
4159 facet_is_controller_flow(struct facet *facet)
4162 struct ofproto_dpif *ofproto = facet->ofproto;
4163 const struct ofpact *ofpacts;
4164 struct rule_dpif *rule;
4168 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4169 ofpacts_len = rule->up.actions->ofpacts_len;
4170 ofpacts = rule->up.actions->ofpacts;
4171 is_controller = ofpacts_len > 0
4172 && ofpacts->type == OFPACT_CONTROLLER
4173 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4174 rule_dpif_unref(rule);
4176 return is_controller;
4181 /* Folds all of 'facet''s statistics into its rule. Also updates the
4182 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4183 * 'facet''s statistics in the datapath should have been zeroed and folded into
4184 * its packet and byte counts before this function is called. */
4186 facet_flush_stats(struct facet *facet)
4188 struct ofproto_dpif *ofproto = facet->ofproto;
4189 struct subfacet *subfacet;
4191 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4192 ovs_assert(!subfacet->dp_byte_count);
4193 ovs_assert(!subfacet->dp_packet_count);
4196 facet_push_stats(facet, false);
4197 if (facet->accounted_bytes < facet->byte_count) {
4198 facet_account(facet);
4199 facet->accounted_bytes = facet->byte_count;
4202 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4203 struct ofexpired expired;
4204 expired.flow = facet->flow;
4205 expired.packet_count = facet->packet_count;
4206 expired.byte_count = facet->byte_count;
4207 expired.used = facet->used;
4208 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4211 /* Reset counters to prevent double counting if 'facet' ever gets
4213 facet_reset_counters(facet);
4215 netflow_flow_clear(&facet->nf_flow);
4216 facet->tcp_flags = 0;
4219 /* Searches 'ofproto''s table of facets for one which would be responsible for
4220 * 'flow'. Returns it if found, otherwise a null pointer.
4222 * The returned facet might need revalidation; use facet_lookup_valid()
4223 * instead if that is important. */
4224 static struct facet *
4225 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4227 struct cls_rule *cr;
4229 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4230 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4231 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4232 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4235 /* Searches 'ofproto''s table of facets for one capable that covers
4236 * 'flow'. Returns it if found, otherwise a null pointer.
4238 * The returned facet is guaranteed to be valid. */
4239 static struct facet *
4240 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4242 struct facet *facet;
4244 facet = facet_find(ofproto, flow);
4246 && ofproto->backer->need_revalidate
4247 && !facet_revalidate(facet)) {
4255 facet_check_consistency(struct facet *facet)
4257 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4259 struct xlate_out xout;
4260 struct xlate_in xin;
4262 struct rule_dpif *rule;
4265 /* Check the datapath actions for consistency. */
4266 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4267 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4268 xlate_actions(&xin, &xout);
4269 rule_dpif_unref(rule);
4271 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4272 && facet->xout.slow == xout.slow;
4273 if (!ok && !VLOG_DROP_WARN(&rl)) {
4274 struct ds s = DS_EMPTY_INITIALIZER;
4276 flow_format(&s, &facet->flow);
4277 ds_put_cstr(&s, ": inconsistency in facet");
4279 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4280 ds_put_cstr(&s, " (actions were: ");
4281 format_odp_actions(&s, facet->xout.odp_actions.data,
4282 facet->xout.odp_actions.size);
4283 ds_put_cstr(&s, ") (correct actions: ");
4284 format_odp_actions(&s, xout.odp_actions.data,
4285 xout.odp_actions.size);
4286 ds_put_char(&s, ')');
4289 if (facet->xout.slow != xout.slow) {
4290 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4295 xlate_out_uninit(&xout);
4300 /* Re-searches the classifier for 'facet':
4302 * - If the rule found is different from 'facet''s current rule, moves
4303 * 'facet' to the new rule and recompiles its actions.
4305 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4306 * where it is and recompiles its actions anyway.
4308 * - If any of 'facet''s subfacets correspond to a new flow according to
4309 * xlate_receive(), 'facet' is removed.
4311 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4313 facet_revalidate(struct facet *facet)
4315 struct ofproto_dpif *ofproto = facet->ofproto;
4316 struct rule_dpif *new_rule;
4317 struct subfacet *subfacet;
4318 struct flow_wildcards wc;
4319 struct xlate_out xout;
4320 struct xlate_in xin;
4322 COVERAGE_INC(facet_revalidate);
4324 /* Check that child subfacets still correspond to this facet. Tunnel
4325 * configuration changes could cause a subfacet's OpenFlow in_port to
4327 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4328 struct ofproto_dpif *recv_ofproto;
4329 struct flow recv_flow;
4332 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4333 subfacet->key_len, &recv_flow, NULL,
4334 &recv_ofproto, NULL);
4336 || recv_ofproto != ofproto
4337 || facet != facet_find(ofproto, &recv_flow)) {
4338 facet_remove(facet);
4343 flow_wildcards_init_catchall(&wc);
4344 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4346 /* Calculate new datapath actions.
4348 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4349 * emit a NetFlow expiration and, if so, we need to have the old state
4350 * around to properly compose it. */
4351 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4352 xlate_actions(&xin, &xout);
4353 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4355 /* A facet's slow path reason should only change under dramatic
4356 * circumstances. Rather than try to update everything, it's simpler to
4357 * remove the facet and start over.
4359 * More importantly, if a facet's wildcards change, it will be relatively
4360 * difficult to figure out if its subfacets still belong to it, and if not
4361 * which facet they may belong to. Again, to avoid the complexity, we
4362 * simply give up instead. */
4363 if (facet->xout.slow != xout.slow
4364 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4365 facet_remove(facet);
4366 xlate_out_uninit(&xout);
4367 rule_dpif_unref(new_rule);
4371 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4372 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4373 if (subfacet->path == SF_FAST_PATH) {
4374 struct dpif_flow_stats stats;
4376 subfacet_install(subfacet, &xout.odp_actions, &stats);
4377 subfacet_update_stats(subfacet, &stats);
4381 facet_flush_stats(facet);
4383 ofpbuf_clear(&facet->xout.odp_actions);
4384 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4385 xout.odp_actions.size);
4388 /* Update 'facet' now that we've taken care of all the old state. */
4389 facet->xout.slow = xout.slow;
4390 facet->xout.has_learn = xout.has_learn;
4391 facet->xout.has_normal = xout.has_normal;
4392 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4393 facet->xout.nf_output_iface = xout.nf_output_iface;
4394 facet->xout.mirrors = xout.mirrors;
4395 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4396 facet->used = MAX(facet->used, new_rule->up.created);
4398 xlate_out_uninit(&xout);
4399 rule_dpif_unref(new_rule);
4404 facet_reset_counters(struct facet *facet)
4406 facet->packet_count = 0;
4407 facet->byte_count = 0;
4408 facet->prev_packet_count = 0;
4409 facet->prev_byte_count = 0;
4410 facet->accounted_bytes = 0;
4414 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4415 struct dpif_flow_stats *stats, bool may_learn)
4417 struct ofport_dpif *in_port;
4418 struct rule_dpif *rule;
4419 struct xlate_in xin;
4421 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4422 if (in_port && in_port->is_tunnel) {
4423 netdev_vport_inc_rx(in_port->up.netdev, stats);
4426 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4427 rule_dpif_credit_stats(rule, stats);
4428 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4429 xin.resubmit_stats = stats;
4430 xin.may_learn = may_learn;
4431 xlate_actions_for_side_effects(&xin);
4432 rule_dpif_unref(rule);
4436 facet_push_stats(struct facet *facet, bool may_learn)
4438 struct dpif_flow_stats stats;
4440 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4441 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4442 ovs_assert(facet->used >= facet->prev_used);
4444 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4445 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4446 stats.used = facet->used;
4447 stats.tcp_flags = facet->tcp_flags;
4449 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4450 facet->prev_packet_count = facet->packet_count;
4451 facet->prev_byte_count = facet->byte_count;
4452 facet->prev_used = facet->used;
4454 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4456 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4457 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4458 stats.n_packets, stats.n_bytes);
4459 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4464 push_all_stats__(bool run_fast)
4466 static long long int rl = LLONG_MIN;
4467 struct ofproto_dpif *ofproto;
4469 if (time_msec() < rl) {
4473 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4474 struct cls_cursor cursor;
4475 struct facet *facet;
4477 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4478 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4479 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4480 facet_push_stats(facet, false);
4485 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4488 rl = time_msec() + 100;
4492 push_all_stats(void)
4494 push_all_stats__(true);
4498 rule_dpif_credit_stats(struct rule_dpif *rule,
4499 const struct dpif_flow_stats *stats)
4501 ovs_mutex_lock(&rule->stats_mutex);
4502 rule->packet_count += stats->n_packets;
4503 rule->byte_count += stats->n_bytes;
4504 rule->up.used = MAX(rule->up.used, stats->used);
4505 ovs_mutex_unlock(&rule->stats_mutex);
4509 rule_dpif_fail_open(const struct rule_dpif *rule)
4511 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4515 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4517 return rule->up.flow_cookie;
4521 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4522 uint16_t hard_timeout)
4524 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4527 /* Returns 'rule''s actions. The caller owns a reference on the returned
4528 * actions and must eventually release it (with rule_actions_unref()) to avoid
4530 struct rule_actions *
4531 rule_dpif_get_actions(const struct rule_dpif *rule)
4533 struct rule_actions *actions;
4535 ovs_rwlock_rdlock(&rule->up.rwlock);
4536 actions = rule->up.actions;
4537 rule_actions_ref(actions);
4538 ovs_rwlock_unlock(&rule->up.rwlock);
4545 static struct subfacet *
4546 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4547 size_t key_len, uint32_t key_hash)
4549 struct subfacet *subfacet;
4551 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4552 &backer->subfacets) {
4553 if (subfacet->key_len == key_len
4554 && !memcmp(key, subfacet->key, key_len)) {
4562 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4563 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4564 * existing subfacet if there is one, otherwise creates and returns a
4566 static struct subfacet *
4567 subfacet_create(struct facet *facet, struct flow_miss *miss)
4569 struct dpif_backer *backer = miss->ofproto->backer;
4570 enum odp_key_fitness key_fitness = miss->key_fitness;
4571 const struct nlattr *key = miss->key;
4572 size_t key_len = miss->key_len;
4574 struct subfacet *subfacet;
4576 key_hash = odp_flow_key_hash(key, key_len);
4578 if (list_is_empty(&facet->subfacets)) {
4579 subfacet = &facet->one_subfacet;
4581 subfacet = subfacet_find(backer, key, key_len, key_hash);
4583 if (subfacet->facet == facet) {
4587 /* This shouldn't happen. */
4588 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4589 subfacet_destroy(subfacet);
4592 subfacet = xmalloc(sizeof *subfacet);
4595 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4596 list_push_back(&facet->subfacets, &subfacet->list_node);
4597 subfacet->facet = facet;
4598 subfacet->key_fitness = key_fitness;
4599 subfacet->key = xmemdup(key, key_len);
4600 subfacet->key_len = key_len;
4601 subfacet->used = miss->stats.used;
4602 subfacet->created = subfacet->used;
4603 subfacet->dp_packet_count = 0;
4604 subfacet->dp_byte_count = 0;
4605 subfacet->path = SF_NOT_INSTALLED;
4606 subfacet->backer = backer;
4608 backer->subfacet_add_count++;
4612 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4613 * its facet within 'ofproto', and frees it. */
4615 subfacet_destroy__(struct subfacet *subfacet)
4617 struct facet *facet = subfacet->facet;
4618 struct ofproto_dpif *ofproto = facet->ofproto;
4620 /* Update ofproto stats before uninstall the subfacet. */
4621 ofproto->backer->subfacet_del_count++;
4623 subfacet_uninstall(subfacet);
4624 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4625 list_remove(&subfacet->list_node);
4626 free(subfacet->key);
4627 if (subfacet != &facet->one_subfacet) {
4632 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4633 * last remaining subfacet in its facet destroys the facet too. */
4635 subfacet_destroy(struct subfacet *subfacet)
4637 struct facet *facet = subfacet->facet;
4639 if (list_is_singleton(&facet->subfacets)) {
4640 /* facet_remove() needs at least one subfacet (it will remove it). */
4641 facet_remove(facet);
4643 subfacet_destroy__(subfacet);
4648 subfacet_destroy_batch(struct dpif_backer *backer,
4649 struct subfacet **subfacets, int n)
4651 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4652 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4653 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4656 for (i = 0; i < n; i++) {
4657 ops[i].type = DPIF_OP_FLOW_DEL;
4658 ops[i].u.flow_del.key = subfacets[i]->key;
4659 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4660 ops[i].u.flow_del.stats = &stats[i];
4664 dpif_operate(backer->dpif, opsp, n);
4665 for (i = 0; i < n; i++) {
4666 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4667 subfacets[i]->path = SF_NOT_INSTALLED;
4668 subfacet_destroy(subfacets[i]);
4673 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4674 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4675 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4676 * since 'subfacet' was last updated.
4678 * Returns 0 if successful, otherwise a positive errno value. */
4680 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4681 struct dpif_flow_stats *stats)
4683 struct facet *facet = subfacet->facet;
4684 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4685 const struct nlattr *actions = odp_actions->data;
4686 size_t actions_len = odp_actions->size;
4687 struct odputil_keybuf maskbuf;
4690 uint64_t slow_path_stub[128 / 8];
4691 enum dpif_flow_put_flags flags;
4694 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4697 flags |= DPIF_FP_ZERO_STATS;
4700 if (path == SF_SLOW_PATH) {
4701 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4702 slow_path_stub, sizeof slow_path_stub,
4703 &actions, &actions_len);
4706 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4707 if (enable_megaflows) {
4708 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4709 &facet->flow, UINT32_MAX);
4712 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4713 subfacet->key_len, mask.data, mask.size,
4714 actions, actions_len, stats);
4717 subfacet_reset_dp_stats(subfacet, stats);
4721 COVERAGE_INC(subfacet_install_fail);
4723 subfacet->path = path;
4728 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4730 subfacet_uninstall(struct subfacet *subfacet)
4732 if (subfacet->path != SF_NOT_INSTALLED) {
4733 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4734 struct dpif_flow_stats stats;
4737 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4738 subfacet->key_len, &stats);
4739 subfacet_reset_dp_stats(subfacet, &stats);
4741 subfacet_update_stats(subfacet, &stats);
4743 subfacet->path = SF_NOT_INSTALLED;
4745 ovs_assert(subfacet->dp_packet_count == 0);
4746 ovs_assert(subfacet->dp_byte_count == 0);
4750 /* Resets 'subfacet''s datapath statistics counters. This should be called
4751 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4752 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4753 * was reset in the datapath. 'stats' will be modified to include only
4754 * statistics new since 'subfacet' was last updated. */
4756 subfacet_reset_dp_stats(struct subfacet *subfacet,
4757 struct dpif_flow_stats *stats)
4760 && subfacet->dp_packet_count <= stats->n_packets
4761 && subfacet->dp_byte_count <= stats->n_bytes) {
4762 stats->n_packets -= subfacet->dp_packet_count;
4763 stats->n_bytes -= subfacet->dp_byte_count;
4766 subfacet->dp_packet_count = 0;
4767 subfacet->dp_byte_count = 0;
4770 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4772 * Because of the meaning of a subfacet's counters, it only makes sense to do
4773 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4774 * represents a packet that was sent by hand or if it represents statistics
4775 * that have been cleared out of the datapath. */
4777 subfacet_update_stats(struct subfacet *subfacet,
4778 const struct dpif_flow_stats *stats)
4780 if (stats->n_packets || stats->used > subfacet->used) {
4781 struct facet *facet = subfacet->facet;
4783 subfacet->used = MAX(subfacet->used, stats->used);
4784 facet->used = MAX(facet->used, stats->used);
4785 facet->packet_count += stats->n_packets;
4786 facet->byte_count += stats->n_bytes;
4787 facet->tcp_flags |= stats->tcp_flags;
4793 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4794 * the fields that were relevant as part of the lookup. */
4796 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4797 struct flow_wildcards *wc, struct rule_dpif **rule)
4799 struct ofport_dpif *port;
4801 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4804 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4806 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4807 flow->in_port.ofp_port);
4810 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4811 ofproto->no_packet_in_rule, rule);
4815 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4816 const struct flow *flow, struct flow_wildcards *wc,
4817 uint8_t table_id, struct rule_dpif **rule)
4819 struct cls_rule *cls_rule;
4820 struct classifier *cls;
4824 if (table_id >= N_TABLES) {
4829 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4830 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4833 cls = &ofproto->up.tables[table_id].cls;
4834 ovs_rwlock_rdlock(&cls->rwlock);
4835 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4836 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4837 /* We must pretend that transport ports are unavailable. */
4838 struct flow ofpc_normal_flow = *flow;
4839 ofpc_normal_flow.tp_src = htons(0);
4840 ofpc_normal_flow.tp_dst = htons(0);
4841 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4842 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4843 cls_rule = &ofproto->drop_frags_rule->up.cr;
4845 flow_wildcards_init_exact(wc);
4848 cls_rule = classifier_lookup(cls, flow, wc);
4851 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4852 rule_dpif_ref(*rule);
4853 ovs_rwlock_unlock(&cls->rwlock);
4855 return *rule != NULL;
4858 /* Given a port configuration (specified as zero if there's no port), chooses
4859 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4860 * flow table miss. */
4862 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4863 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4865 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4866 rule_dpif_ref(*rule);
4870 rule_dpif_ref(struct rule_dpif *rule)
4873 ofproto_rule_ref(&rule->up);
4878 rule_dpif_unref(struct rule_dpif *rule)
4881 ofproto_rule_unref(&rule->up);
4886 complete_operation(struct rule_dpif *rule)
4888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4890 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4891 ofoperation_complete(rule->up.pending, 0);
4894 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4896 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4899 static struct rule *
4902 struct rule_dpif *rule = xmalloc(sizeof *rule);
4907 rule_dealloc(struct rule *rule_)
4909 struct rule_dpif *rule = rule_dpif_cast(rule_);
4914 rule_construct(struct rule *rule_)
4916 struct rule_dpif *rule = rule_dpif_cast(rule_);
4917 ovs_mutex_init(&rule->stats_mutex);
4918 ovs_mutex_lock(&rule->stats_mutex);
4919 rule->packet_count = 0;
4920 rule->byte_count = 0;
4921 ovs_mutex_unlock(&rule->stats_mutex);
4926 rule_insert(struct rule *rule_)
4928 struct rule_dpif *rule = rule_dpif_cast(rule_);
4929 complete_operation(rule);
4933 rule_delete(struct rule *rule_)
4935 struct rule_dpif *rule = rule_dpif_cast(rule_);
4936 complete_operation(rule);
4940 rule_destruct(struct rule *rule_)
4942 struct rule_dpif *rule = rule_dpif_cast(rule_);
4943 ovs_mutex_destroy(&rule->stats_mutex);
4947 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4949 struct rule_dpif *rule = rule_dpif_cast(rule_);
4951 /* push_all_stats() can handle flow misses which, when using the learn
4952 * action, can cause rules to be added and deleted. This can corrupt our
4953 * caller's datastructures which assume that rule_get_stats() doesn't have
4954 * an impact on the flow table. To be safe, we disable miss handling. */
4955 push_all_stats__(false);
4957 /* Start from historical data for 'rule' itself that are no longer tracked
4958 * in facets. This counts, for example, facets that have expired. */
4959 ovs_mutex_lock(&rule->stats_mutex);
4960 *packets = rule->packet_count;
4961 *bytes = rule->byte_count;
4962 ovs_mutex_unlock(&rule->stats_mutex);
4966 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4967 struct ofpbuf *packet)
4969 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4970 struct dpif_flow_stats stats;
4971 struct xlate_out xout;
4972 struct xlate_in xin;
4974 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4975 rule_dpif_credit_stats(rule, &stats);
4977 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4978 xin.resubmit_stats = &stats;
4979 xlate_actions(&xin, &xout);
4981 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4982 xout.odp_actions.size, packet);
4984 xlate_out_uninit(&xout);
4988 rule_execute(struct rule *rule, const struct flow *flow,
4989 struct ofpbuf *packet)
4991 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4992 ofpbuf_delete(packet);
4997 rule_modify_actions(struct rule *rule_, bool reset_counters)
4999 struct rule_dpif *rule = rule_dpif_cast(rule_);
5001 if (reset_counters) {
5002 ovs_mutex_lock(&rule->stats_mutex);
5003 rule->packet_count = 0;
5004 rule->byte_count = 0;
5005 ovs_mutex_unlock(&rule->stats_mutex);
5008 complete_operation(rule);
5011 /* Sends 'packet' out 'ofport'.
5012 * May modify 'packet'.
5013 * Returns 0 if successful, otherwise a positive errno value. */
5015 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5018 uint64_t odp_actions_stub[1024 / 8];
5019 struct ofpbuf key, odp_actions;
5020 struct dpif_flow_stats stats;
5021 struct odputil_keybuf keybuf;
5022 struct ofpact_output output;
5023 struct xlate_out xout;
5024 struct xlate_in xin;
5026 union flow_in_port in_port_;
5029 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5030 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5032 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5033 in_port_.ofp_port = OFPP_NONE;
5034 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5035 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5037 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5039 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5040 output.port = ofport->up.ofp_port;
5043 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5044 xin.ofpacts_len = sizeof output;
5045 xin.ofpacts = &output.ofpact;
5046 xin.resubmit_stats = &stats;
5047 xlate_actions(&xin, &xout);
5049 error = dpif_execute(ofproto->backer->dpif,
5051 xout.odp_actions.data, xout.odp_actions.size,
5053 xlate_out_uninit(&xout);
5056 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5057 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5058 ovs_strerror(error));
5061 ofproto->stats.tx_packets++;
5062 ofproto->stats.tx_bytes += packet->size;
5066 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5067 * The action will state 'slow' as the reason that the action is in the slow
5068 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5069 * dump-flows" output to see why a flow is in the slow path.)
5071 * The 'stub_size' bytes in 'stub' will be used to store the action.
5072 * 'stub_size' must be large enough for the action.
5074 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5077 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5078 enum slow_path_reason slow,
5079 uint64_t *stub, size_t stub_size,
5080 const struct nlattr **actionsp, size_t *actions_lenp)
5082 union user_action_cookie cookie;
5085 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5086 cookie.slow_path.unused = 0;
5087 cookie.slow_path.reason = slow;
5089 ofpbuf_use_stack(&buf, stub, stub_size);
5090 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5091 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5093 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5095 odp_port_t odp_port;
5098 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5099 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5100 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5102 *actionsp = buf.data;
5103 *actions_lenp = buf.size;
5107 set_frag_handling(struct ofproto *ofproto_,
5108 enum ofp_config_flags frag_handling)
5110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5111 if (frag_handling != OFPC_FRAG_REASM) {
5112 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5120 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5121 const struct flow *flow,
5122 const struct ofpact *ofpacts, size_t ofpacts_len)
5124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5125 struct odputil_keybuf keybuf;
5126 struct dpif_flow_stats stats;
5127 struct xlate_out xout;
5128 struct xlate_in xin;
5132 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5133 odp_flow_key_from_flow(&key, flow,
5134 ofp_port_to_odp_port(ofproto,
5135 flow->in_port.ofp_port));
5137 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5139 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5140 xin.resubmit_stats = &stats;
5141 xin.ofpacts_len = ofpacts_len;
5142 xin.ofpacts = ofpacts;
5144 xlate_actions(&xin, &xout);
5145 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5146 xout.odp_actions.data, xout.odp_actions.size, packet);
5147 xlate_out_uninit(&xout);
5155 set_netflow(struct ofproto *ofproto_,
5156 const struct netflow_options *netflow_options)
5158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5160 if (netflow_options) {
5161 if (!ofproto->netflow) {
5162 ofproto->netflow = netflow_create();
5163 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5165 return netflow_set_options(ofproto->netflow, netflow_options);
5166 } else if (ofproto->netflow) {
5167 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5168 netflow_destroy(ofproto->netflow);
5169 ofproto->netflow = NULL;
5176 get_netflow_ids(const struct ofproto *ofproto_,
5177 uint8_t *engine_type, uint8_t *engine_id)
5179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5181 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5185 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5187 if (!facet_is_controller_flow(facet) &&
5188 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5189 struct subfacet *subfacet;
5190 struct ofexpired expired;
5192 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5193 if (subfacet->path == SF_FAST_PATH) {
5194 struct dpif_flow_stats stats;
5196 subfacet_install(subfacet, &facet->xout.odp_actions,
5198 subfacet_update_stats(subfacet, &stats);
5202 expired.flow = facet->flow;
5203 expired.packet_count = facet->packet_count;
5204 expired.byte_count = facet->byte_count;
5205 expired.used = facet->used;
5206 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5211 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5213 struct cls_cursor cursor;
5214 struct facet *facet;
5216 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5217 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5218 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5219 send_active_timeout(ofproto, facet);
5221 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5224 static struct ofproto_dpif *
5225 ofproto_dpif_lookup(const char *name)
5227 struct ofproto_dpif *ofproto;
5229 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5230 hash_string(name, 0), &all_ofproto_dpifs) {
5231 if (!strcmp(ofproto->up.name, name)) {
5239 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5240 const char *argv[], void *aux OVS_UNUSED)
5242 struct ofproto_dpif *ofproto;
5245 ofproto = ofproto_dpif_lookup(argv[1]);
5247 unixctl_command_reply_error(conn, "no such bridge");
5250 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5251 mac_learning_flush(ofproto->ml);
5252 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5254 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5255 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5256 mac_learning_flush(ofproto->ml);
5257 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5261 unixctl_command_reply(conn, "table successfully flushed");
5264 static struct ofport_dpif *
5265 ofbundle_get_a_port(const struct ofbundle *bundle)
5267 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5272 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5273 const char *argv[], void *aux OVS_UNUSED)
5275 struct ds ds = DS_EMPTY_INITIALIZER;
5276 const struct ofproto_dpif *ofproto;
5277 const struct mac_entry *e;
5279 ofproto = ofproto_dpif_lookup(argv[1]);
5281 unixctl_command_reply_error(conn, "no such bridge");
5285 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5286 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5287 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5288 struct ofbundle *bundle = e->port.p;
5289 char name[OFP_MAX_PORT_NAME_LEN];
5291 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5293 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5294 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5295 mac_entry_age(ofproto->ml, e));
5297 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5298 unixctl_command_reply(conn, ds_cstr(&ds));
5303 struct xlate_out xout;
5304 struct xlate_in xin;
5310 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5312 ds_put_char_multiple(result, '\t', level);
5314 ds_put_cstr(result, "No match\n");
5318 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5319 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5320 cls_rule_format(&rule->up.cr, result);
5321 ds_put_char(result, '\n');
5323 ds_put_char_multiple(result, '\t', level);
5324 ds_put_cstr(result, "OpenFlow ");
5325 ofpacts_format(rule->up.actions->ofpacts, rule->up.actions->ofpacts_len,
5327 ds_put_char(result, '\n');
5331 trace_format_flow(struct ds *result, int level, const char *title,
5332 struct trace_ctx *trace)
5334 ds_put_char_multiple(result, '\t', level);
5335 ds_put_format(result, "%s: ", title);
5336 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5337 ds_put_cstr(result, "unchanged");
5339 flow_format(result, &trace->xin.flow);
5340 trace->flow = trace->xin.flow;
5342 ds_put_char(result, '\n');
5346 trace_format_regs(struct ds *result, int level, const char *title,
5347 struct trace_ctx *trace)
5351 ds_put_char_multiple(result, '\t', level);
5352 ds_put_format(result, "%s:", title);
5353 for (i = 0; i < FLOW_N_REGS; i++) {
5354 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5356 ds_put_char(result, '\n');
5360 trace_format_odp(struct ds *result, int level, const char *title,
5361 struct trace_ctx *trace)
5363 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5365 ds_put_char_multiple(result, '\t', level);
5366 ds_put_format(result, "%s: ", title);
5367 format_odp_actions(result, odp_actions->data, odp_actions->size);
5368 ds_put_char(result, '\n');
5372 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5374 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5375 struct ds *result = trace->result;
5377 ds_put_char(result, '\n');
5378 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5379 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5380 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5381 trace_format_rule(result, recurse + 1, rule);
5385 trace_report(struct xlate_in *xin, const char *s, int recurse)
5387 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5388 struct ds *result = trace->result;
5390 ds_put_char_multiple(result, '\t', recurse);
5391 ds_put_cstr(result, s);
5392 ds_put_char(result, '\n');
5396 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5397 void *aux OVS_UNUSED)
5399 const struct dpif_backer *backer;
5400 struct ofproto_dpif *ofproto;
5401 struct ofpbuf odp_key, odp_mask;
5402 struct ofpbuf *packet;
5410 ofpbuf_init(&odp_key, 0);
5411 ofpbuf_init(&odp_mask, 0);
5413 /* Handle "-generate" or a hex string as the last argument. */
5414 if (!strcmp(argv[argc - 1], "-generate")) {
5415 packet = ofpbuf_new(0);
5418 const char *error = eth_from_hex(argv[argc - 1], &packet);
5421 } else if (argc == 4) {
5422 /* The 3-argument form must end in "-generate' or a hex string. */
5423 unixctl_command_reply_error(conn, error);
5428 /* Parse the flow and determine whether a datapath or
5429 * bridge is specified. If function odp_flow_key_from_string()
5430 * returns 0, the flow is a odp_flow. If function
5431 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5432 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5433 /* If the odp_flow is the second argument,
5434 * the datapath name is the first argument. */
5436 const char *dp_type;
5437 if (!strncmp(argv[1], "ovs-", 4)) {
5438 dp_type = argv[1] + 4;
5442 backer = shash_find_data(&all_dpif_backers, dp_type);
5444 unixctl_command_reply_error(conn, "Cannot find datapath "
5449 /* No datapath name specified, so there should be only one
5451 struct shash_node *node;
5452 if (shash_count(&all_dpif_backers) != 1) {
5453 unixctl_command_reply_error(conn, "Must specify datapath "
5454 "name, there is more than one type of datapath");
5457 node = shash_first(&all_dpif_backers);
5458 backer = node->data;
5461 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5462 NULL, &ofproto, NULL)) {
5463 unixctl_command_reply_error(conn, "Invalid datapath flow");
5466 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5467 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5469 unixctl_command_reply_error(conn, "Must specify bridge name");
5473 ofproto = ofproto_dpif_lookup(argv[1]);
5475 unixctl_command_reply_error(conn, "Unknown bridge name");
5479 unixctl_command_reply_error(conn, "Bad flow syntax");
5483 /* Generate a packet, if requested. */
5485 if (!packet->size) {
5486 flow_compose(packet, &flow);
5488 union flow_in_port in_port_;
5490 in_port_ = flow.in_port;
5491 ds_put_cstr(&result, "Packet: ");
5492 s = ofp_packet_to_string(packet->data, packet->size);
5493 ds_put_cstr(&result, s);
5496 /* Use the metadata from the flow and the packet argument
5497 * to reconstruct the flow. */
5498 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5503 ofproto_trace(ofproto, &flow, packet, &result);
5504 unixctl_command_reply(conn, ds_cstr(&result));
5507 ds_destroy(&result);
5508 ofpbuf_delete(packet);
5509 ofpbuf_uninit(&odp_key);
5510 ofpbuf_uninit(&odp_mask);
5514 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5515 const struct ofpbuf *packet, struct ds *ds)
5517 struct rule_dpif *rule;
5518 struct flow_wildcards wc;
5520 ds_put_cstr(ds, "Flow: ");
5521 flow_format(ds, flow);
5522 ds_put_char(ds, '\n');
5524 flow_wildcards_init_catchall(&wc);
5525 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5527 trace_format_rule(ds, 0, rule);
5528 if (rule == ofproto->miss_rule) {
5529 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5530 } else if (rule == ofproto->no_packet_in_rule) {
5531 ds_put_cstr(ds, "\nNo match, packets dropped because "
5532 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5533 } else if (rule == ofproto->drop_frags_rule) {
5534 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5535 "and the fragment handling mode is \"drop\".\n");
5539 uint64_t odp_actions_stub[1024 / 8];
5540 struct ofpbuf odp_actions;
5541 struct trace_ctx trace;
5545 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5548 ofpbuf_use_stub(&odp_actions,
5549 odp_actions_stub, sizeof odp_actions_stub);
5550 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5551 trace.xin.resubmit_hook = trace_resubmit;
5552 trace.xin.report_hook = trace_report;
5554 xlate_actions(&trace.xin, &trace.xout);
5555 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5557 ds_put_char(ds, '\n');
5558 trace_format_flow(ds, 0, "Final flow", &trace);
5560 match_init(&match, flow, &trace.xout.wc);
5561 ds_put_cstr(ds, "Relevant fields: ");
5562 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5563 ds_put_char(ds, '\n');
5565 ds_put_cstr(ds, "Datapath actions: ");
5566 format_odp_actions(ds, trace.xout.odp_actions.data,
5567 trace.xout.odp_actions.size);
5569 if (trace.xout.slow) {
5570 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5571 "slow path because it:");
5572 switch (trace.xout.slow) {
5574 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5577 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5580 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5583 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5585 case SLOW_CONTROLLER:
5586 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5587 "to the OpenFlow controller.");
5594 xlate_out_uninit(&trace.xout);
5597 rule_dpif_unref(rule);
5600 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5601 * 'reply' describing the results. */
5603 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5605 struct cls_cursor cursor;
5606 struct facet *facet;
5610 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5611 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5612 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5613 if (!facet_check_consistency(facet)) {
5617 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5619 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5623 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5624 ofproto->up.name, errors);
5626 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5631 ofproto_dpif_self_check(struct unixctl_conn *conn,
5632 int argc, const char *argv[], void *aux OVS_UNUSED)
5634 struct ds reply = DS_EMPTY_INITIALIZER;
5635 struct ofproto_dpif *ofproto;
5638 ofproto = ofproto_dpif_lookup(argv[1]);
5640 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5641 "ofproto/list for help)");
5644 ofproto_dpif_self_check__(ofproto, &reply);
5646 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5647 ofproto_dpif_self_check__(ofproto, &reply);
5651 unixctl_command_reply(conn, ds_cstr(&reply));
5655 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5656 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5657 * to destroy 'ofproto_shash' and free the returned value. */
5658 static const struct shash_node **
5659 get_ofprotos(struct shash *ofproto_shash)
5661 const struct ofproto_dpif *ofproto;
5663 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5664 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5665 shash_add_nocopy(ofproto_shash, name, ofproto);
5668 return shash_sort(ofproto_shash);
5672 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5673 const char *argv[] OVS_UNUSED,
5674 void *aux OVS_UNUSED)
5676 struct ds ds = DS_EMPTY_INITIALIZER;
5677 struct shash ofproto_shash;
5678 const struct shash_node **sorted_ofprotos;
5681 shash_init(&ofproto_shash);
5682 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5683 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5684 const struct shash_node *node = sorted_ofprotos[i];
5685 ds_put_format(&ds, "%s\n", node->name);
5688 shash_destroy(&ofproto_shash);
5689 free(sorted_ofprotos);
5691 unixctl_command_reply(conn, ds_cstr(&ds));
5696 show_dp_rates(struct ds *ds, const char *heading,
5697 const struct avg_subfacet_rates *rates)
5699 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5700 heading, rates->add_rate, rates->del_rate);
5704 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5706 const struct shash_node **ofprotos;
5707 struct ofproto_dpif *ofproto;
5708 struct shash ofproto_shash;
5709 uint64_t n_hit, n_missed;
5710 long long int minutes;
5713 n_hit = n_missed = 0;
5714 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5715 if (ofproto->backer == backer) {
5716 n_missed += ofproto->n_missed;
5717 n_hit += ofproto->n_hit;
5721 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5722 dpif_name(backer->dpif), n_hit, n_missed);
5723 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5724 " life span: %lldms\n", hmap_count(&backer->subfacets),
5725 backer->avg_n_subfacet, backer->max_n_subfacet,
5726 backer->avg_subfacet_life);
5728 minutes = (time_msec() - backer->created) / (1000 * 60);
5729 if (minutes >= 60) {
5730 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5732 if (minutes >= 60 * 24) {
5733 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5735 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5737 shash_init(&ofproto_shash);
5738 ofprotos = get_ofprotos(&ofproto_shash);
5739 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5740 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5741 const struct shash_node **ports;
5744 if (ofproto->backer != backer) {
5748 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5749 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5751 ports = shash_sort(&ofproto->up.port_by_name);
5752 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5753 const struct shash_node *node = ports[j];
5754 struct ofport *ofport = node->data;
5756 odp_port_t odp_port;
5758 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5761 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5762 if (odp_port != ODPP_NONE) {
5763 ds_put_format(ds, "%"PRIu32":", odp_port);
5765 ds_put_cstr(ds, "none:");
5768 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5771 if (!netdev_get_config(ofport->netdev, &config)) {
5772 const struct smap_node **nodes;
5775 nodes = smap_sort(&config);
5776 for (i = 0; i < smap_count(&config); i++) {
5777 const struct smap_node *node = nodes[i];
5778 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5779 node->key, node->value);
5783 smap_destroy(&config);
5785 ds_put_char(ds, ')');
5786 ds_put_char(ds, '\n');
5790 shash_destroy(&ofproto_shash);
5795 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5796 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5798 struct ds ds = DS_EMPTY_INITIALIZER;
5799 const struct shash_node **backers;
5802 backers = shash_sort(&all_dpif_backers);
5803 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5804 dpif_show_backer(backers[i]->data, &ds);
5808 unixctl_command_reply(conn, ds_cstr(&ds));
5812 /* Dump the megaflow (facet) cache. This is useful to check the
5813 * correctness of flow wildcarding, since the same mechanism is used for
5814 * both xlate caching and kernel wildcarding.
5816 * It's important to note that in the output the flow description uses
5817 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5819 * This command is only needed for advanced debugging, so it's not
5820 * documented in the man page. */
5822 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5823 int argc OVS_UNUSED, const char *argv[],
5824 void *aux OVS_UNUSED)
5826 struct ds ds = DS_EMPTY_INITIALIZER;
5827 const struct ofproto_dpif *ofproto;
5828 long long int now = time_msec();
5829 struct cls_cursor cursor;
5830 struct facet *facet;
5832 ofproto = ofproto_dpif_lookup(argv[1]);
5834 unixctl_command_reply_error(conn, "no such bridge");
5838 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5839 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5840 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5841 cls_rule_format(&facet->cr, &ds);
5842 ds_put_cstr(&ds, ", ");
5843 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5844 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5845 ds_put_cstr(&ds, "Datapath actions: ");
5846 if (facet->xout.slow) {
5847 uint64_t slow_path_stub[128 / 8];
5848 const struct nlattr *actions;
5851 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5852 slow_path_stub, sizeof slow_path_stub,
5853 &actions, &actions_len);
5854 format_odp_actions(&ds, actions, actions_len);
5856 format_odp_actions(&ds, facet->xout.odp_actions.data,
5857 facet->xout.odp_actions.size);
5859 ds_put_cstr(&ds, "\n");
5861 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5863 ds_chomp(&ds, '\n');
5864 unixctl_command_reply(conn, ds_cstr(&ds));
5868 /* Disable using the megaflows.
5870 * This command is only needed for advanced debugging, so it's not
5871 * documented in the man page. */
5873 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5874 int argc OVS_UNUSED,
5875 const char *argv[] OVS_UNUSED,
5876 void *aux OVS_UNUSED)
5878 struct ofproto_dpif *ofproto;
5880 enable_megaflows = false;
5882 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5883 flush(&ofproto->up);
5886 unixctl_command_reply(conn, "megaflows disabled");
5889 /* Re-enable using megaflows.
5891 * This command is only needed for advanced debugging, so it's not
5892 * documented in the man page. */
5894 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5895 int argc OVS_UNUSED,
5896 const char *argv[] OVS_UNUSED,
5897 void *aux OVS_UNUSED)
5899 struct ofproto_dpif *ofproto;
5901 enable_megaflows = true;
5903 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5904 flush(&ofproto->up);
5907 unixctl_command_reply(conn, "megaflows enabled");
5911 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5912 int argc OVS_UNUSED, const char *argv[],
5913 void *aux OVS_UNUSED)
5915 struct ds ds = DS_EMPTY_INITIALIZER;
5916 const struct ofproto_dpif *ofproto;
5917 struct subfacet *subfacet;
5919 ofproto = ofproto_dpif_lookup(argv[1]);
5921 unixctl_command_reply_error(conn, "no such bridge");
5925 update_stats(ofproto->backer);
5927 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5928 struct facet *facet = subfacet->facet;
5929 struct odputil_keybuf maskbuf;
5932 if (facet->ofproto != ofproto) {
5936 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5937 if (enable_megaflows) {
5938 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5939 &facet->flow, UINT32_MAX);
5942 odp_flow_format(subfacet->key, subfacet->key_len,
5943 mask.data, mask.size, &ds, false);
5945 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5946 subfacet->dp_packet_count, subfacet->dp_byte_count);
5947 if (subfacet->used) {
5948 ds_put_format(&ds, "%.3fs",
5949 (time_msec() - subfacet->used) / 1000.0);
5951 ds_put_format(&ds, "never");
5953 if (subfacet->facet->tcp_flags) {
5954 ds_put_cstr(&ds, ", flags:");
5955 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5958 ds_put_cstr(&ds, ", actions:");
5959 if (facet->xout.slow) {
5960 uint64_t slow_path_stub[128 / 8];
5961 const struct nlattr *actions;
5964 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5965 slow_path_stub, sizeof slow_path_stub,
5966 &actions, &actions_len);
5967 format_odp_actions(&ds, actions, actions_len);
5969 format_odp_actions(&ds, facet->xout.odp_actions.data,
5970 facet->xout.odp_actions.size);
5972 ds_put_char(&ds, '\n');
5975 unixctl_command_reply(conn, ds_cstr(&ds));
5980 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5981 int argc OVS_UNUSED, const char *argv[],
5982 void *aux OVS_UNUSED)
5984 struct ds ds = DS_EMPTY_INITIALIZER;
5985 struct ofproto_dpif *ofproto;
5987 ofproto = ofproto_dpif_lookup(argv[1]);
5989 unixctl_command_reply_error(conn, "no such bridge");
5993 flush(&ofproto->up);
5995 unixctl_command_reply(conn, ds_cstr(&ds));
6000 ofproto_dpif_unixctl_init(void)
6002 static bool registered;
6008 unixctl_command_register(
6010 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6011 1, 3, ofproto_unixctl_trace, NULL);
6012 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6013 ofproto_unixctl_fdb_flush, NULL);
6014 unixctl_command_register("fdb/show", "bridge", 1, 1,
6015 ofproto_unixctl_fdb_show, NULL);
6016 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6017 ofproto_dpif_self_check, NULL);
6018 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6019 ofproto_unixctl_dpif_dump_dps, NULL);
6020 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6022 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6023 ofproto_unixctl_dpif_dump_flows, NULL);
6024 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6025 ofproto_unixctl_dpif_del_flows, NULL);
6026 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6027 ofproto_unixctl_dpif_dump_megaflows, NULL);
6028 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6029 ofproto_unixctl_dpif_disable_megaflows, NULL);
6030 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6031 ofproto_unixctl_dpif_enable_megaflows, NULL);
6034 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6036 * This is deprecated. It is only for compatibility with broken device drivers
6037 * in old versions of Linux that do not properly support VLANs when VLAN
6038 * devices are not used. When broken device drivers are no longer in
6039 * widespread use, we will delete these interfaces. */
6042 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6044 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6045 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6047 if (realdev_ofp_port == ofport->realdev_ofp_port
6048 && vid == ofport->vlandev_vid) {
6052 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6054 if (ofport->realdev_ofp_port) {
6057 if (realdev_ofp_port && ofport->bundle) {
6058 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6059 * themselves be part of a bundle. */
6060 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6063 ofport->realdev_ofp_port = realdev_ofp_port;
6064 ofport->vlandev_vid = vid;
6066 if (realdev_ofp_port) {
6067 vsp_add(ofport, realdev_ofp_port, vid);
6074 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6076 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6080 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6081 OVS_EXCLUDED(ofproto->vsp_mutex)
6085 ovs_mutex_lock(&ofproto->vsp_mutex);
6086 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6087 ovs_mutex_unlock(&ofproto->vsp_mutex);
6092 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6093 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6094 OVS_REQUIRES(ofproto->vsp_mutex)
6096 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6097 int vid = vlan_tci_to_vid(vlan_tci);
6098 const struct vlan_splinter *vsp;
6100 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6101 hash_realdev_vid(realdev_ofp_port, vid),
6102 &ofproto->realdev_vid_map) {
6103 if (vsp->realdev_ofp_port == realdev_ofp_port
6104 && vsp->vid == vid) {
6105 return vsp->vlandev_ofp_port;
6109 return realdev_ofp_port;
6112 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6113 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6114 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6115 * 'vlan_tci' 9, it would return the port number of eth0.9.
6117 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6118 * function just returns its 'realdev_ofp_port' argument. */
6120 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6121 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6122 OVS_EXCLUDED(ofproto->vsp_mutex)
6126 ovs_mutex_lock(&ofproto->vsp_mutex);
6127 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6128 ovs_mutex_unlock(&ofproto->vsp_mutex);
6132 static struct vlan_splinter *
6133 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6135 struct vlan_splinter *vsp;
6137 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6138 hash_ofp_port(vlandev_ofp_port),
6139 &ofproto->vlandev_map) {
6140 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6148 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6149 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6150 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6151 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6152 * eth0 and store 9 in '*vid'.
6154 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6155 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6158 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6159 ofp_port_t vlandev_ofp_port, int *vid)
6160 OVS_REQUIRES(ofproto->vsp_mutex)
6162 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6163 const struct vlan_splinter *vsp;
6165 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6170 return vsp->realdev_ofp_port;
6176 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6177 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6178 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6179 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6180 * always the case unless VLAN splinters are enabled), returns false without
6181 * making any changes. */
6183 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6184 OVS_EXCLUDED(ofproto->vsp_mutex)
6189 ovs_mutex_lock(&ofproto->vsp_mutex);
6190 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6191 ovs_mutex_unlock(&ofproto->vsp_mutex);
6196 /* Cause the flow to be processed as if it came in on the real device with
6197 * the VLAN device's VLAN ID. */
6198 flow->in_port.ofp_port = realdev;
6199 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6204 vsp_remove(struct ofport_dpif *port)
6206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6207 struct vlan_splinter *vsp;
6209 ovs_mutex_lock(&ofproto->vsp_mutex);
6210 vsp = vlandev_find(ofproto, port->up.ofp_port);
6212 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6213 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6216 port->realdev_ofp_port = 0;
6218 VLOG_ERR("missing vlan device record");
6220 ovs_mutex_unlock(&ofproto->vsp_mutex);
6224 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6226 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6228 ovs_mutex_lock(&ofproto->vsp_mutex);
6229 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6230 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6231 == realdev_ofp_port)) {
6232 struct vlan_splinter *vsp;
6234 vsp = xmalloc(sizeof *vsp);
6235 vsp->realdev_ofp_port = realdev_ofp_port;
6236 vsp->vlandev_ofp_port = port->up.ofp_port;
6239 port->realdev_ofp_port = realdev_ofp_port;
6241 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6242 hash_ofp_port(port->up.ofp_port));
6243 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6244 hash_realdev_vid(realdev_ofp_port, vid));
6246 VLOG_ERR("duplicate vlan device record");
6248 ovs_mutex_unlock(&ofproto->vsp_mutex);
6252 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6254 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6255 return ofport ? ofport->odp_port : ODPP_NONE;
6258 struct ofport_dpif *
6259 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6261 struct ofport_dpif *port;
6263 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6264 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6265 &backer->odp_to_ofport_map) {
6266 if (port->odp_port == odp_port) {
6267 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6272 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6277 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6279 struct ofport_dpif *port;
6281 port = odp_port_to_ofport(ofproto->backer, odp_port);
6282 if (port && &ofproto->up == port->up.ofproto) {
6283 return port->up.ofp_port;
6289 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6290 * most heavily weighted element. 'base' designates the rate of decay: after
6291 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6294 exp_mavg(double *avg, int base, double new)
6296 *avg = (*avg * (base - 1) + new) / base;
6300 update_moving_averages(struct dpif_backer *backer)
6302 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6303 long long int minutes = (time_msec() - backer->created) / min_ms;
6306 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6308 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6311 backer->lifetime.add_rate = 0.0;
6312 backer->lifetime.del_rate = 0.0;
6315 /* Update hourly averages on the minute boundaries. */
6316 if (time_msec() - backer->last_minute >= min_ms) {
6317 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6318 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6320 /* Update daily averages on the hour boundaries. */
6321 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6322 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6323 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6326 backer->total_subfacet_add_count += backer->subfacet_add_count;
6327 backer->total_subfacet_del_count += backer->subfacet_del_count;
6328 backer->subfacet_add_count = 0;
6329 backer->subfacet_del_count = 0;
6330 backer->last_minute += min_ms;
6334 const struct ofproto_class ofproto_dpif_class = {
6369 port_is_lacp_current,
6370 NULL, /* rule_choose_table */
6379 rule_modify_actions,
6393 get_stp_port_status,
6400 is_mirror_output_bundle,
6401 forward_bpdu_changed,
6402 set_mac_table_config,
6404 NULL, /* meter_get_features */
6405 NULL, /* meter_set */
6406 NULL, /* meter_get */
6407 NULL, /* meter_del */
6408 NULL, /* group_alloc */
6409 NULL, /* group_construct */
6410 NULL, /* group_destruct */
6411 NULL, /* group_dealloc */
6412 NULL, /* group_modify */
6413 NULL, /* group_get_stats */