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);
86 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
89 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
90 struct ofproto_dpif *ofproto; /* Owning ofproto. */
91 void *aux; /* Key supplied by ofproto's client. */
92 char *name; /* Identifier for log messages. */
95 struct list ports; /* Contains "struct ofport"s. */
96 enum port_vlan_mode vlan_mode; /* VLAN mode */
97 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
98 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
99 * NULL if all VLANs are trunked. */
100 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
101 struct bond *bond; /* Nonnull iff more than one port. */
102 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
105 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
108 static void bundle_remove(struct ofport *);
109 static void bundle_update(struct ofbundle *);
110 static void bundle_destroy(struct ofbundle *);
111 static void bundle_del_port(struct ofport_dpif *);
112 static void bundle_run(struct ofbundle *);
113 static void bundle_wait(struct ofbundle *);
115 static void stp_run(struct ofproto_dpif *ofproto);
116 static void stp_wait(struct ofproto_dpif *ofproto);
117 static int set_stp_port(struct ofport *,
118 const struct ofproto_port_stp_settings *);
120 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
121 enum slow_path_reason,
122 uint64_t *stub, size_t stub_size,
123 const struct nlattr **actionsp,
124 size_t *actions_lenp);
126 /* A subfacet (see "struct subfacet" below) has three possible installation
129 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
130 * case just after the subfacet is created, just before the subfacet is
131 * destroyed, or if the datapath returns an error when we try to install a
134 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
136 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
137 * ofproto_dpif is installed in the datapath.
140 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
141 SF_FAST_PATH, /* Full actions are installed. */
142 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
145 /* A dpif flow and actions associated with a facet.
147 * See also the large comment on struct facet. */
150 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
151 struct list list_node; /* In struct facet's 'facets' list. */
152 struct facet *facet; /* Owning facet. */
153 struct dpif_backer *backer; /* Owning backer. */
155 enum odp_key_fitness key_fitness;
159 long long int used; /* Time last used; time created if not used. */
160 long long int created; /* Time created. */
162 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
163 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
165 enum subfacet_path path; /* Installed in datapath? */
168 #define SUBFACET_DESTROY_MAX_BATCH 50
170 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
171 static struct subfacet *subfacet_find(struct dpif_backer *,
172 const struct nlattr *key, size_t key_len,
174 static void subfacet_destroy(struct subfacet *);
175 static void subfacet_destroy__(struct subfacet *);
176 static void subfacet_destroy_batch(struct dpif_backer *,
177 struct subfacet **, int n);
178 static void subfacet_reset_dp_stats(struct subfacet *,
179 struct dpif_flow_stats *);
180 static void subfacet_update_stats(struct subfacet *,
181 const struct dpif_flow_stats *);
182 static int subfacet_install(struct subfacet *,
183 const struct ofpbuf *odp_actions,
184 struct dpif_flow_stats *);
185 static void subfacet_uninstall(struct subfacet *);
187 /* A unique, non-overlapping instantiation of an OpenFlow flow.
189 * A facet associates a "struct flow", which represents the Open vSwitch
190 * userspace idea of an exact-match flow, with one or more subfacets.
191 * While the facet is created based on an exact-match flow, it is stored
192 * within the ofproto based on the wildcards that could be expressed
193 * based on the flow table and other configuration. (See the 'wc'
194 * description in "struct xlate_out" for more details.)
196 * Each subfacet tracks the datapath's idea of the flow equivalent to
197 * the facet. When the kernel module (or other dpif implementation) and
198 * Open vSwitch userspace agree on the definition of a flow key, there
199 * is exactly one subfacet per facet. If the dpif implementation
200 * supports more-specific flow matching than userspace, however, a facet
201 * can have more than one subfacet. Examples include the dpif
202 * implementation not supporting the same wildcards as userspace or some
203 * distinction in flow that userspace simply doesn't understand.
205 * Flow expiration works in terms of subfacets, so a facet must have at
206 * least one subfacet or it will never expire, leaking memory. */
209 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
210 struct ofproto_dpif *ofproto;
213 struct list subfacets;
214 long long int used; /* Time last used; time created if not used. */
217 struct flow flow; /* Flow of the creating subfacet. */
218 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
222 * - Do include packets and bytes sent "by hand", e.g. with
225 * - Do include packets and bytes that were obtained from the datapath
226 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
227 * DPIF_FP_ZERO_STATS).
229 * - Do not include packets or bytes that can be obtained from the
230 * datapath for any existing subfacet.
232 uint64_t packet_count; /* Number of packets received. */
233 uint64_t byte_count; /* Number of bytes received. */
235 /* Resubmit statistics. */
236 uint64_t prev_packet_count; /* Number of packets from last stats push. */
237 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
238 long long int prev_used; /* Used time from last stats push. */
241 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
242 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
243 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
245 struct xlate_out xout;
247 /* Storage for a single subfacet, to reduce malloc() time and space
248 * overhead. (A facet always has at least one subfacet and in the common
249 * case has exactly one subfacet. However, 'one_subfacet' may not
250 * always be valid, since it could have been removed after newer
251 * subfacets were pushed onto the 'subfacets' list.) */
252 struct subfacet one_subfacet;
254 long long int learn_rl; /* Rate limiter for facet_learn(). */
257 static struct facet *facet_create(const struct flow_miss *);
258 static void facet_remove(struct facet *);
259 static void facet_free(struct facet *);
261 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
262 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
263 const struct flow *);
264 static bool facet_revalidate(struct facet *);
265 static bool facet_check_consistency(struct facet *);
267 static void facet_flush_stats(struct facet *);
269 static void facet_reset_counters(struct facet *);
270 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
271 struct dpif_flow_stats *, bool may_learn);
272 static void facet_push_stats(struct facet *, bool may_learn);
273 static void facet_learn(struct facet *);
274 static void facet_account(struct facet *);
275 static void push_all_stats(void);
277 static bool facet_is_controller_flow(struct facet *);
280 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
284 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
285 struct list bundle_node; /* In struct ofbundle's "ports" list. */
286 struct cfm *cfm; /* Connectivity Fault Management, if any. */
287 struct bfd *bfd; /* BFD, if any. */
288 bool may_enable; /* May be enabled in bonds. */
289 bool is_tunnel; /* This port is a tunnel. */
290 long long int carrier_seq; /* Carrier status changes. */
291 struct ofport_dpif *peer; /* Peer if patch port. */
294 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
295 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
296 long long int stp_state_entered;
298 /* Queue to DSCP mapping. */
299 struct ofproto_port_queue *qdscp;
302 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
304 * This is deprecated. It is only for compatibility with broken device
305 * drivers in old versions of Linux that do not properly support VLANs when
306 * VLAN devices are not used. When broken device drivers are no longer in
307 * widespread use, we will delete these interfaces. */
308 ofp_port_t realdev_ofp_port;
312 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
314 * This is deprecated. It is only for compatibility with broken device drivers
315 * in old versions of Linux that do not properly support VLANs when VLAN
316 * devices are not used. When broken device drivers are no longer in
317 * widespread use, we will delete these interfaces. */
318 struct vlan_splinter {
319 struct hmap_node realdev_vid_node;
320 struct hmap_node vlandev_node;
321 ofp_port_t realdev_ofp_port;
322 ofp_port_t vlandev_ofp_port;
326 static void vsp_remove(struct ofport_dpif *);
327 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
329 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
332 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
335 static struct ofport_dpif *
336 ofport_dpif_cast(const struct ofport *ofport)
338 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
341 static void port_run(struct ofport_dpif *);
342 static void port_run_fast(struct ofport_dpif *);
343 static void port_wait(struct ofport_dpif *);
344 static int set_bfd(struct ofport *, const struct smap *);
345 static int set_cfm(struct ofport *, const struct cfm_settings *);
346 static void ofport_update_peer(struct ofport_dpif *);
347 static void run_fast_rl(void);
348 static int run_fast(struct ofproto *);
350 struct dpif_completion {
351 struct list list_node;
352 struct ofoperation *op;
355 /* Reasons that we might need to revalidate every facet, and corresponding
358 * A value of 0 means that there is no need to revalidate.
360 * It would be nice to have some cleaner way to integrate with coverage
361 * counters, but with only a few reasons I guess this is good enough for
363 enum revalidate_reason {
364 REV_RECONFIGURE = 1, /* Switch configuration changed. */
365 REV_STP, /* Spanning tree protocol port status change. */
366 REV_BOND, /* Bonding changed. */
367 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
368 REV_FLOW_TABLE, /* Flow table changed. */
369 REV_MAC_LEARNING, /* Mac learning changed. */
370 REV_INCONSISTENCY /* Facet self-check failed. */
372 COVERAGE_DEFINE(rev_reconfigure);
373 COVERAGE_DEFINE(rev_stp);
374 COVERAGE_DEFINE(rev_bond);
375 COVERAGE_DEFINE(rev_port_toggled);
376 COVERAGE_DEFINE(rev_flow_table);
377 COVERAGE_DEFINE(rev_mac_learning);
378 COVERAGE_DEFINE(rev_inconsistency);
380 struct avg_subfacet_rates {
381 double add_rate; /* Moving average of new flows created per minute. */
382 double del_rate; /* Moving average of flows deleted per minute. */
385 /* All datapaths of a given type share a single dpif backer instance. */
391 struct timer next_expiration;
393 struct ovs_rwlock odp_to_ofport_lock;
394 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
396 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
398 /* Facet revalidation flags applying to facets which use this backer. */
399 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
401 struct hmap drop_keys; /* Set of dropped odp keys. */
402 bool recv_set_enable; /* Enables or disables receiving packets. */
404 struct hmap subfacets;
405 struct governor *governor;
407 /* Subfacet statistics.
409 * These keep track of the total number of subfacets added and deleted and
410 * flow life span. They are useful for computing the flow rates stats
411 * exposed via "ovs-appctl dpif/show". The goal is to learn about
412 * traffic patterns in ways that we can use later to improve Open vSwitch
413 * performance in new situations. */
414 long long int created; /* Time when it is created. */
415 unsigned max_n_subfacet; /* Maximum number of flows */
416 unsigned avg_n_subfacet; /* Average number of flows. */
417 long long int avg_subfacet_life; /* Average life span of subfacets. */
419 /* The average number of subfacets... */
420 struct avg_subfacet_rates hourly; /* ...over the last hour. */
421 struct avg_subfacet_rates daily; /* ...over the last day. */
422 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
423 long long int last_minute; /* Last time 'hourly' was updated. */
425 /* Number of subfacets added or deleted since 'last_minute'. */
426 unsigned subfacet_add_count;
427 unsigned subfacet_del_count;
429 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
430 unsigned long long int total_subfacet_add_count;
431 unsigned long long int total_subfacet_del_count;
433 /* Number of upcall handling threads. */
434 unsigned int n_handler_threads;
437 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
438 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
440 static void drop_key_clear(struct dpif_backer *);
441 static void update_moving_averages(struct dpif_backer *backer);
443 struct ofproto_dpif {
444 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
446 struct dpif_backer *backer;
448 /* Special OpenFlow rules. */
449 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
450 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
451 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
454 struct netflow *netflow;
455 struct dpif_sflow *sflow;
456 struct dpif_ipfix *ipfix;
457 struct hmap bundles; /* Contains "struct ofbundle"s. */
458 struct mac_learning *ml;
459 bool has_bonded_bundles;
460 struct mbridge *mbridge;
463 struct classifier facets; /* Contains 'struct facet's. */
464 long long int consistency_rl;
466 /* Support for debugging async flow mods. */
467 struct list completions;
469 struct netdev_stats stats; /* To account packets generated and consumed in
474 long long int stp_last_tick;
476 /* VLAN splinters. */
477 struct ovs_mutex vsp_mutex;
478 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
479 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
482 struct sset ports; /* Set of standard port names. */
483 struct sset ghost_ports; /* Ports with no datapath port. */
484 struct sset port_poll_set; /* Queued names for port_poll() reply. */
485 int port_poll_errno; /* Last errno for port_poll() reply. */
487 /* Per ofproto's dpif stats. */
492 struct ovs_mutex flow_mod_mutex;
493 struct list flow_mods OVS_GUARDED;
494 size_t n_flow_mods OVS_GUARDED;
496 struct ovs_mutex pin_mutex;
497 struct list pins OVS_GUARDED;
498 size_t n_pins OVS_GUARDED;
501 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
502 * for debugging the asynchronous flow_mod implementation.) */
505 /* By default, flows in the datapath are wildcarded (megaflows). They
506 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
507 static bool enable_megaflows = true;
509 /* All existing ofproto_dpif instances, indexed by ->up.name. */
510 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
512 static void ofproto_dpif_unixctl_init(void);
514 static inline struct ofproto_dpif *
515 ofproto_dpif_cast(const struct ofproto *ofproto)
517 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
518 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
521 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
522 ofp_port_t ofp_port);
523 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
524 const struct ofpbuf *packet, struct ds *);
527 static void handle_upcalls(struct dpif_backer *);
529 /* Flow expiration. */
530 static int expire(struct dpif_backer *);
533 static void send_netflow_active_timeouts(struct ofproto_dpif *);
536 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
538 /* Global variables. */
539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
541 /* Initial mappings of port to bridge mappings. */
542 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
544 /* Executes and takes ownership of 'fm'. */
546 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
547 struct ofputil_flow_mod *fm)
549 ovs_mutex_lock(&ofproto->flow_mod_mutex);
550 if (ofproto->n_flow_mods > 1024) {
551 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
552 COVERAGE_INC(flow_mod_overflow);
558 list_push_back(&ofproto->flow_mods, &fm->list_node);
559 ofproto->n_flow_mods++;
560 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
563 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
564 * Takes ownership of 'pin' and pin->packet. */
566 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
567 struct ofputil_packet_in *pin)
569 ovs_mutex_lock(&ofproto->pin_mutex);
570 if (ofproto->n_pins > 1024) {
571 ovs_mutex_unlock(&ofproto->pin_mutex);
572 COVERAGE_INC(packet_in_overflow);
573 free(CONST_CAST(void *, pin->packet));
578 list_push_back(&ofproto->pins, &pin->list_node);
580 ovs_mutex_unlock(&ofproto->pin_mutex);
583 /* Factory functions. */
586 init(const struct shash *iface_hints)
588 struct shash_node *node;
590 /* Make a local copy, since we don't own 'iface_hints' elements. */
591 SHASH_FOR_EACH(node, iface_hints) {
592 const struct iface_hint *orig_hint = node->data;
593 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
595 new_hint->br_name = xstrdup(orig_hint->br_name);
596 new_hint->br_type = xstrdup(orig_hint->br_type);
597 new_hint->ofp_port = orig_hint->ofp_port;
599 shash_add(&init_ofp_ports, node->name, new_hint);
604 enumerate_types(struct sset *types)
606 dp_enumerate_types(types);
610 enumerate_names(const char *type, struct sset *names)
612 struct ofproto_dpif *ofproto;
615 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
616 if (strcmp(type, ofproto->up.type)) {
619 sset_add(names, ofproto->up.name);
626 del(const char *type, const char *name)
631 error = dpif_open(name, type, &dpif);
633 error = dpif_delete(dpif);
640 port_open_type(const char *datapath_type, const char *port_type)
642 return dpif_port_open_type(datapath_type, port_type);
645 /* Type functions. */
647 static void process_dpif_port_changes(struct dpif_backer *);
648 static void process_dpif_all_ports_changed(struct dpif_backer *);
649 static void process_dpif_port_change(struct dpif_backer *,
650 const char *devname);
651 static void process_dpif_port_error(struct dpif_backer *, int error);
653 static struct ofproto_dpif *
654 lookup_ofproto_dpif_by_port_name(const char *name)
656 struct ofproto_dpif *ofproto;
658 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
659 if (sset_contains(&ofproto->ports, name)) {
668 type_run(const char *type)
670 static long long int push_timer = LLONG_MIN;
671 struct dpif_backer *backer;
673 backer = shash_find_data(&all_dpif_backers, type);
675 /* This is not necessarily a problem, since backers are only
676 * created on demand. */
680 dpif_run(backer->dpif);
682 /* The most natural place to push facet statistics is when they're pulled
683 * from the datapath. However, when there are many flows in the datapath,
684 * this expensive operation can occur so frequently, that it reduces our
685 * ability to quickly set up flows. To reduce the cost, we push statistics
687 if (time_msec() > push_timer) {
688 push_timer = time_msec() + 2000;
692 /* If vswitchd started with other_config:flow_restore_wait set as "true",
693 * and the configuration has now changed to "false", enable receiving
694 * packets from the datapath. */
695 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
698 backer->recv_set_enable = true;
700 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
702 udpif_recv_set(backer->udpif, 0, false);
703 VLOG_ERR("Failed to enable receiving packets in dpif.");
706 udpif_recv_set(backer->udpif, n_handler_threads,
707 backer->recv_set_enable);
708 dpif_flow_flush(backer->dpif);
709 backer->need_revalidate = REV_RECONFIGURE;
712 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
713 * to reset the handler threads. */
714 if (backer->n_handler_threads != n_handler_threads) {
715 udpif_recv_set(backer->udpif, n_handler_threads,
716 backer->recv_set_enable);
717 backer->n_handler_threads = n_handler_threads;
720 if (backer->need_revalidate) {
721 struct ofproto_dpif *ofproto;
722 struct simap_node *node;
723 struct simap tmp_backers;
725 /* Handle tunnel garbage collection. */
726 simap_init(&tmp_backers);
727 simap_swap(&backer->tnl_backers, &tmp_backers);
729 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
730 struct ofport_dpif *iter;
732 if (backer != ofproto->backer) {
736 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
737 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
740 if (!iter->is_tunnel) {
744 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
745 namebuf, sizeof namebuf);
746 node = simap_find(&tmp_backers, dp_port);
748 simap_put(&backer->tnl_backers, dp_port, node->data);
749 simap_delete(&tmp_backers, node);
750 node = simap_find(&backer->tnl_backers, dp_port);
752 node = simap_find(&backer->tnl_backers, dp_port);
754 odp_port_t odp_port = ODPP_NONE;
756 if (!dpif_port_add(backer->dpif, iter->up.netdev,
758 simap_put(&backer->tnl_backers, dp_port,
759 odp_to_u32(odp_port));
760 node = simap_find(&backer->tnl_backers, dp_port);
765 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
766 if (tnl_port_reconfigure(iter, iter->up.netdev,
768 backer->need_revalidate = REV_RECONFIGURE;
773 SIMAP_FOR_EACH (node, &tmp_backers) {
774 dpif_port_del(backer->dpif, u32_to_odp(node->data));
776 simap_destroy(&tmp_backers);
778 switch (backer->need_revalidate) {
779 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
780 case REV_STP: COVERAGE_INC(rev_stp); break;
781 case REV_BOND: COVERAGE_INC(rev_bond); break;
782 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
783 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
784 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
785 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
787 backer->need_revalidate = 0;
789 /* Clear the drop_keys in case we should now be accepting some
790 * formerly dropped flows. */
791 drop_key_clear(backer);
793 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
794 struct facet *facet, *next;
795 struct ofport_dpif *ofport;
796 struct cls_cursor cursor;
797 struct ofbundle *bundle;
799 if (ofproto->backer != backer) {
803 ovs_rwlock_wrlock(&xlate_rwlock);
804 xlate_ofproto_set(ofproto, ofproto->up.name,
805 ofproto->backer->dpif, ofproto->miss_rule,
806 ofproto->no_packet_in_rule, ofproto->ml,
807 ofproto->stp, ofproto->mbridge,
808 ofproto->sflow, ofproto->ipfix,
809 ofproto->up.frag_handling,
810 ofproto->up.forward_bpdu,
811 connmgr_has_in_band(ofproto->up.connmgr),
812 ofproto->netflow != NULL);
814 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
815 xlate_bundle_set(ofproto, bundle, bundle->name,
816 bundle->vlan_mode, bundle->vlan,
817 bundle->trunks, bundle->use_priority_tags,
818 bundle->bond, bundle->lacp,
822 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
823 int stp_port = ofport->stp_port
824 ? stp_port_no(ofport->stp_port)
826 xlate_ofport_set(ofproto, ofport->bundle, ofport,
827 ofport->up.ofp_port, ofport->odp_port,
828 ofport->up.netdev, ofport->cfm,
829 ofport->bfd, ofport->peer, stp_port,
830 ofport->qdscp, ofport->n_qdscp,
831 ofport->up.pp.config, ofport->is_tunnel,
834 ovs_rwlock_unlock(&xlate_rwlock);
836 /* Only ofproto-dpif cares about the facet classifier so we just
837 * lock cls_cursor_init() to appease the thread safety analysis. */
838 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
839 cls_cursor_init(&cursor, &ofproto->facets, NULL);
840 ovs_rwlock_unlock(&ofproto->facets.rwlock);
841 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
842 facet_revalidate(facet);
847 udpif_revalidate(backer->udpif);
850 if (!backer->recv_set_enable) {
851 /* Wake up before a max of 1000ms. */
852 timer_set_duration(&backer->next_expiration, 1000);
853 } else if (timer_expired(&backer->next_expiration)) {
854 int delay = expire(backer);
855 timer_set_duration(&backer->next_expiration, delay);
858 process_dpif_port_changes(backer);
860 if (backer->governor) {
863 governor_run(backer->governor);
865 /* If the governor has shrunk to its minimum size and the number of
866 * subfacets has dwindled, then drop the governor entirely.
868 * For hysteresis, the number of subfacets to drop the governor is
869 * smaller than the number needed to trigger its creation. */
870 n_subfacets = hmap_count(&backer->subfacets);
871 if (n_subfacets * 4 < flow_eviction_threshold
872 && governor_is_idle(backer->governor)) {
873 governor_destroy(backer->governor);
874 backer->governor = NULL;
881 /* Check for and handle port changes in 'backer''s dpif. */
883 process_dpif_port_changes(struct dpif_backer *backer)
889 error = dpif_port_poll(backer->dpif, &devname);
895 process_dpif_all_ports_changed(backer);
899 process_dpif_port_change(backer, devname);
904 process_dpif_port_error(backer, error);
911 process_dpif_all_ports_changed(struct dpif_backer *backer)
913 struct ofproto_dpif *ofproto;
914 struct dpif_port dpif_port;
915 struct dpif_port_dump dump;
916 struct sset devnames;
919 sset_init(&devnames);
920 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
921 if (ofproto->backer == backer) {
922 struct ofport *ofport;
924 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
925 sset_add(&devnames, netdev_get_name(ofport->netdev));
929 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
930 sset_add(&devnames, dpif_port.name);
933 SSET_FOR_EACH (devname, &devnames) {
934 process_dpif_port_change(backer, devname);
936 sset_destroy(&devnames);
940 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
942 struct ofproto_dpif *ofproto;
943 struct dpif_port port;
945 /* Don't report on the datapath's device. */
946 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
950 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
951 &all_ofproto_dpifs) {
952 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
957 ofproto = lookup_ofproto_dpif_by_port_name(devname);
958 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
959 /* The port was removed. If we know the datapath,
960 * report it through poll_set(). If we don't, it may be
961 * notifying us of a removal we initiated, so ignore it.
962 * If there's a pending ENOBUFS, let it stand, since
963 * everything will be reevaluated. */
964 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
965 sset_add(&ofproto->port_poll_set, devname);
966 ofproto->port_poll_errno = 0;
968 } else if (!ofproto) {
969 /* The port was added, but we don't know with which
970 * ofproto we should associate it. Delete it. */
971 dpif_port_del(backer->dpif, port.port_no);
973 struct ofport_dpif *ofport;
975 ofport = ofport_dpif_cast(shash_find_data(
976 &ofproto->up.port_by_name, devname));
978 && ofport->odp_port != port.port_no
979 && !odp_port_to_ofport(backer, port.port_no))
981 /* 'ofport''s datapath port number has changed from
982 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
983 * structures to match. */
984 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
985 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
986 ofport->odp_port = port.port_no;
987 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
988 hash_odp_port(port.port_no));
989 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
990 backer->need_revalidate = REV_RECONFIGURE;
993 dpif_port_destroy(&port);
996 /* Propagate 'error' to all ofprotos based on 'backer'. */
998 process_dpif_port_error(struct dpif_backer *backer, int error)
1000 struct ofproto_dpif *ofproto;
1002 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1003 if (ofproto->backer == backer) {
1004 sset_clear(&ofproto->port_poll_set);
1005 ofproto->port_poll_errno = error;
1011 dpif_backer_run_fast(struct dpif_backer *backer)
1013 udpif_run(backer->udpif);
1014 handle_upcalls(backer);
1020 type_run_fast(const char *type)
1022 struct dpif_backer *backer;
1024 backer = shash_find_data(&all_dpif_backers, type);
1026 /* This is not necessarily a problem, since backers are only
1027 * created on demand. */
1031 return dpif_backer_run_fast(backer);
1037 static long long int port_rl = LLONG_MIN;
1039 if (time_msec() >= port_rl) {
1040 struct ofproto_dpif *ofproto;
1042 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1043 run_fast(&ofproto->up);
1045 port_rl = time_msec() + 200;
1050 type_wait(const char *type)
1052 struct dpif_backer *backer;
1054 backer = shash_find_data(&all_dpif_backers, type);
1056 /* This is not necessarily a problem, since backers are only
1057 * created on demand. */
1061 if (backer->governor) {
1062 governor_wait(backer->governor);
1065 timer_wait(&backer->next_expiration);
1066 dpif_wait(backer->dpif);
1067 udpif_wait(backer->udpif);
1070 /* Basic life-cycle. */
1072 static int add_internal_flows(struct ofproto_dpif *);
1074 static struct ofproto *
1077 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1078 return &ofproto->up;
1082 dealloc(struct ofproto *ofproto_)
1084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1089 close_dpif_backer(struct dpif_backer *backer)
1091 struct shash_node *node;
1093 ovs_assert(backer->refcount > 0);
1095 if (--backer->refcount) {
1099 drop_key_clear(backer);
1100 hmap_destroy(&backer->drop_keys);
1102 simap_destroy(&backer->tnl_backers);
1103 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1104 hmap_destroy(&backer->odp_to_ofport_map);
1105 node = shash_find(&all_dpif_backers, backer->type);
1107 shash_delete(&all_dpif_backers, node);
1108 udpif_destroy(backer->udpif);
1109 dpif_close(backer->dpif);
1111 ovs_assert(hmap_is_empty(&backer->subfacets));
1112 hmap_destroy(&backer->subfacets);
1113 governor_destroy(backer->governor);
1118 /* Datapath port slated for removal from datapath. */
1119 struct odp_garbage {
1120 struct list list_node;
1121 odp_port_t odp_port;
1125 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1127 struct dpif_backer *backer;
1128 struct dpif_port_dump port_dump;
1129 struct dpif_port port;
1130 struct shash_node *node;
1131 struct list garbage_list;
1132 struct odp_garbage *garbage, *next;
1138 backer = shash_find_data(&all_dpif_backers, type);
1145 backer_name = xasprintf("ovs-%s", type);
1147 /* Remove any existing datapaths, since we assume we're the only
1148 * userspace controlling the datapath. */
1150 dp_enumerate_names(type, &names);
1151 SSET_FOR_EACH(name, &names) {
1152 struct dpif *old_dpif;
1154 /* Don't remove our backer if it exists. */
1155 if (!strcmp(name, backer_name)) {
1159 if (dpif_open(name, type, &old_dpif)) {
1160 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1162 dpif_delete(old_dpif);
1163 dpif_close(old_dpif);
1166 sset_destroy(&names);
1168 backer = xmalloc(sizeof *backer);
1170 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1173 VLOG_ERR("failed to open datapath of type %s: %s", type,
1174 ovs_strerror(error));
1178 backer->udpif = udpif_create(backer, backer->dpif);
1180 backer->type = xstrdup(type);
1181 backer->governor = NULL;
1182 backer->refcount = 1;
1183 hmap_init(&backer->odp_to_ofport_map);
1184 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1185 hmap_init(&backer->drop_keys);
1186 hmap_init(&backer->subfacets);
1187 timer_set_duration(&backer->next_expiration, 1000);
1188 backer->need_revalidate = 0;
1189 simap_init(&backer->tnl_backers);
1190 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1193 if (backer->recv_set_enable) {
1194 dpif_flow_flush(backer->dpif);
1197 /* Loop through the ports already on the datapath and remove any
1198 * that we don't need anymore. */
1199 list_init(&garbage_list);
1200 dpif_port_dump_start(&port_dump, backer->dpif);
1201 while (dpif_port_dump_next(&port_dump, &port)) {
1202 node = shash_find(&init_ofp_ports, port.name);
1203 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1204 garbage = xmalloc(sizeof *garbage);
1205 garbage->odp_port = port.port_no;
1206 list_push_front(&garbage_list, &garbage->list_node);
1209 dpif_port_dump_done(&port_dump);
1211 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1212 dpif_port_del(backer->dpif, garbage->odp_port);
1213 list_remove(&garbage->list_node);
1217 shash_add(&all_dpif_backers, type, backer);
1219 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1221 VLOG_ERR("failed to listen on datapath of type %s: %s",
1222 type, ovs_strerror(error));
1223 close_dpif_backer(backer);
1226 udpif_recv_set(backer->udpif, n_handler_threads,
1227 backer->recv_set_enable);
1228 backer->n_handler_threads = n_handler_threads;
1230 backer->max_n_subfacet = 0;
1231 backer->created = time_msec();
1232 backer->last_minute = backer->created;
1233 memset(&backer->hourly, 0, sizeof backer->hourly);
1234 memset(&backer->daily, 0, sizeof backer->daily);
1235 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1236 backer->subfacet_add_count = 0;
1237 backer->subfacet_del_count = 0;
1238 backer->total_subfacet_add_count = 0;
1239 backer->total_subfacet_del_count = 0;
1240 backer->avg_n_subfacet = 0;
1241 backer->avg_subfacet_life = 0;
1247 construct(struct ofproto *ofproto_)
1249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1250 struct shash_node *node, *next;
1254 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1259 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1260 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1262 ofproto->netflow = NULL;
1263 ofproto->sflow = NULL;
1264 ofproto->ipfix = NULL;
1265 ofproto->stp = NULL;
1266 hmap_init(&ofproto->bundles);
1267 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1268 ofproto->mbridge = mbridge_create();
1269 ofproto->has_bonded_bundles = false;
1270 ovs_mutex_init(&ofproto->vsp_mutex);
1272 classifier_init(&ofproto->facets);
1273 ofproto->consistency_rl = LLONG_MIN;
1275 list_init(&ofproto->completions);
1277 ovs_mutex_init(&ofproto->flow_mod_mutex);
1278 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1279 list_init(&ofproto->flow_mods);
1280 ofproto->n_flow_mods = 0;
1281 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1283 ovs_mutex_init(&ofproto->pin_mutex);
1284 ovs_mutex_lock(&ofproto->pin_mutex);
1285 list_init(&ofproto->pins);
1286 ofproto->n_pins = 0;
1287 ovs_mutex_unlock(&ofproto->pin_mutex);
1289 ofproto_dpif_unixctl_init();
1291 hmap_init(&ofproto->vlandev_map);
1292 hmap_init(&ofproto->realdev_vid_map);
1294 sset_init(&ofproto->ports);
1295 sset_init(&ofproto->ghost_ports);
1296 sset_init(&ofproto->port_poll_set);
1297 ofproto->port_poll_errno = 0;
1299 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1300 struct iface_hint *iface_hint = node->data;
1302 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1303 /* Check if the datapath already has this port. */
1304 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1305 sset_add(&ofproto->ports, node->name);
1308 free(iface_hint->br_name);
1309 free(iface_hint->br_type);
1311 shash_delete(&init_ofp_ports, node);
1315 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1316 hash_string(ofproto->up.name, 0));
1317 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1319 ofproto_init_tables(ofproto_, N_TABLES);
1320 error = add_internal_flows(ofproto);
1321 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1324 ofproto->n_missed = 0;
1330 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1331 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1333 struct ofputil_flow_mod fm;
1336 match_init_catchall(&fm.match);
1338 match_set_reg(&fm.match, 0, id);
1339 fm.new_cookie = htonll(0);
1340 fm.cookie = htonll(0);
1341 fm.cookie_mask = htonll(0);
1342 fm.modify_cookie = false;
1343 fm.table_id = TBL_INTERNAL;
1344 fm.command = OFPFC_ADD;
1345 fm.idle_timeout = 0;
1346 fm.hard_timeout = 0;
1350 fm.ofpacts = ofpacts->data;
1351 fm.ofpacts_len = ofpacts->size;
1353 error = ofproto_flow_mod(&ofproto->up, &fm);
1355 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1356 id, ofperr_to_string(error));
1360 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1362 ovs_rwlock_unlock(&(*rulep)->up.evict);
1371 add_internal_flows(struct ofproto_dpif *ofproto)
1373 struct ofpact_controller *controller;
1374 uint64_t ofpacts_stub[128 / 8];
1375 struct ofpbuf ofpacts;
1379 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1382 controller = ofpact_put_CONTROLLER(&ofpacts);
1383 controller->max_len = UINT16_MAX;
1384 controller->controller_id = 0;
1385 controller->reason = OFPR_NO_MATCH;
1386 ofpact_pad(&ofpacts);
1388 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1393 ofpbuf_clear(&ofpacts);
1394 error = add_internal_flow(ofproto, id++, &ofpacts,
1395 &ofproto->no_packet_in_rule);
1400 error = add_internal_flow(ofproto, id++, &ofpacts,
1401 &ofproto->drop_frags_rule);
1406 complete_operations(struct ofproto_dpif *ofproto)
1408 struct dpif_completion *c, *next;
1410 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1411 ofoperation_complete(c->op, 0);
1412 list_remove(&c->list_node);
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 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1441 complete_operations(ofproto);
1443 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1444 struct cls_cursor cursor;
1446 ovs_rwlock_wrlock(&table->cls.rwlock);
1447 cls_cursor_init(&cursor, &table->cls, NULL);
1448 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1449 ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1451 ovs_rwlock_unlock(&table->cls.rwlock);
1453 complete_operations(ofproto);
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;
1558 complete_operations(ofproto);
1561 if (mbridge_need_revalidate(ofproto->mbridge)) {
1562 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1563 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1564 mac_learning_flush(ofproto->ml);
1565 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1568 /* Do not perform any periodic activity below required by 'ofproto' while
1569 * waiting for flow restore to complete. */
1570 if (ofproto_get_flow_restore_wait()) {
1574 error = run_fast(ofproto_);
1579 if (ofproto->netflow) {
1580 if (netflow_run(ofproto->netflow)) {
1581 send_netflow_active_timeouts(ofproto);
1584 if (ofproto->sflow) {
1585 dpif_sflow_run(ofproto->sflow);
1587 if (ofproto->ipfix) {
1588 dpif_ipfix_run(ofproto->ipfix);
1591 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1594 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1599 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1600 if (mac_learning_run(ofproto->ml)) {
1601 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1603 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1605 /* Check the consistency of a random facet, to aid debugging. */
1606 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1607 if (time_msec() >= ofproto->consistency_rl
1608 && !classifier_is_empty(&ofproto->facets)
1609 && !ofproto->backer->need_revalidate) {
1610 struct cls_table *table;
1611 struct cls_rule *cr;
1612 struct facet *facet;
1614 ofproto->consistency_rl = time_msec() + 250;
1616 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1617 struct cls_table, hmap_node);
1618 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1620 facet = CONTAINER_OF(cr, struct facet, cr);
1622 if (!facet_check_consistency(facet)) {
1623 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1626 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1632 wait(struct ofproto *ofproto_)
1634 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1635 struct ofport_dpif *ofport;
1636 struct ofbundle *bundle;
1638 if (!clogged && !list_is_empty(&ofproto->completions)) {
1639 poll_immediate_wake();
1642 if (ofproto_get_flow_restore_wait()) {
1646 if (ofproto->sflow) {
1647 dpif_sflow_wait(ofproto->sflow);
1649 if (ofproto->ipfix) {
1650 dpif_ipfix_wait(ofproto->ipfix);
1652 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1655 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1656 bundle_wait(bundle);
1658 if (ofproto->netflow) {
1659 netflow_wait(ofproto->netflow);
1661 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1662 mac_learning_wait(ofproto->ml);
1663 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1665 if (ofproto->backer->need_revalidate) {
1666 /* Shouldn't happen, but if it does just go around again. */
1667 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1668 poll_immediate_wake();
1673 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1675 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1676 struct cls_cursor cursor;
1677 size_t n_subfacets = 0;
1678 struct facet *facet;
1680 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1681 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1682 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1684 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1685 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1686 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1687 n_subfacets += list_size(&facet->subfacets);
1689 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1690 simap_increase(usage, "subfacets", n_subfacets);
1694 flush(struct ofproto *ofproto_)
1696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1697 struct subfacet *subfacet, *next_subfacet;
1698 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1702 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1703 &ofproto->backer->subfacets) {
1704 if (subfacet->facet->ofproto != ofproto) {
1708 if (subfacet->path != SF_NOT_INSTALLED) {
1709 batch[n_batch++] = subfacet;
1710 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1711 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1715 subfacet_destroy(subfacet);
1720 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1725 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1726 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1728 *arp_match_ip = true;
1729 *actions = (OFPUTIL_A_OUTPUT |
1730 OFPUTIL_A_SET_VLAN_VID |
1731 OFPUTIL_A_SET_VLAN_PCP |
1732 OFPUTIL_A_STRIP_VLAN |
1733 OFPUTIL_A_SET_DL_SRC |
1734 OFPUTIL_A_SET_DL_DST |
1735 OFPUTIL_A_SET_NW_SRC |
1736 OFPUTIL_A_SET_NW_DST |
1737 OFPUTIL_A_SET_NW_TOS |
1738 OFPUTIL_A_SET_TP_SRC |
1739 OFPUTIL_A_SET_TP_DST |
1744 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1747 struct dpif_dp_stats s;
1748 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1751 strcpy(ots->name, "classifier");
1753 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1754 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1755 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1756 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1758 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1759 ots->lookup_count = htonll(n_lookup);
1760 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1763 static struct ofport *
1766 struct ofport_dpif *port = xmalloc(sizeof *port);
1771 port_dealloc(struct ofport *port_)
1773 struct ofport_dpif *port = ofport_dpif_cast(port_);
1778 port_construct(struct ofport *port_)
1780 struct ofport_dpif *port = ofport_dpif_cast(port_);
1781 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1782 const struct netdev *netdev = port->up.netdev;
1783 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1784 struct dpif_port dpif_port;
1787 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1788 port->bundle = NULL;
1791 port->may_enable = true;
1792 port->stp_port = NULL;
1793 port->stp_state = STP_DISABLED;
1794 port->is_tunnel = false;
1798 port->realdev_ofp_port = 0;
1799 port->vlandev_vid = 0;
1800 port->carrier_seq = netdev_get_carrier_resets(netdev);
1802 if (netdev_vport_is_patch(netdev)) {
1803 /* By bailing out here, we don't submit the port to the sFlow module
1804 * to be considered for counter polling export. This is correct
1805 * because the patch port represents an interface that sFlow considers
1806 * to be "internal" to the switch as a whole, and therefore not an
1807 * candidate for counter polling. */
1808 port->odp_port = ODPP_NONE;
1809 ofport_update_peer(port);
1813 error = dpif_port_query_by_name(ofproto->backer->dpif,
1814 netdev_vport_get_dpif_port(netdev, namebuf,
1821 port->odp_port = dpif_port.port_no;
1823 if (netdev_get_tunnel_config(netdev)) {
1824 tnl_port_add(port, port->up.netdev, port->odp_port);
1825 port->is_tunnel = true;
1827 /* Sanity-check that a mapping doesn't already exist. This
1828 * shouldn't happen for non-tunnel ports. */
1829 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1830 VLOG_ERR("port %s already has an OpenFlow port number",
1832 dpif_port_destroy(&dpif_port);
1836 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1837 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1838 hash_odp_port(port->odp_port));
1839 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1841 dpif_port_destroy(&dpif_port);
1843 if (ofproto->sflow) {
1844 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1851 port_destruct(struct ofport *port_)
1853 struct ofport_dpif *port = ofport_dpif_cast(port_);
1854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1855 const char *devname = netdev_get_name(port->up.netdev);
1856 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1857 const char *dp_port_name;
1859 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1860 ovs_rwlock_wrlock(&xlate_rwlock);
1861 xlate_ofport_remove(port);
1862 ovs_rwlock_unlock(&xlate_rwlock);
1864 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1866 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1867 /* The underlying device is still there, so delete it. This
1868 * happens when the ofproto is being destroyed, since the caller
1869 * assumes that removal of attached ports will happen as part of
1871 if (!port->is_tunnel) {
1872 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1877 port->peer->peer = NULL;
1881 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1882 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1883 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1884 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1888 sset_find_and_delete(&ofproto->ports, devname);
1889 sset_find_and_delete(&ofproto->ghost_ports, devname);
1890 bundle_remove(port_);
1891 set_cfm(port_, NULL);
1892 set_bfd(port_, NULL);
1893 if (ofproto->sflow) {
1894 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1901 port_modified(struct ofport *port_)
1903 struct ofport_dpif *port = ofport_dpif_cast(port_);
1905 if (port->bundle && port->bundle->bond) {
1906 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1910 cfm_set_netdev(port->cfm, port->up.netdev);
1914 bfd_set_netdev(port->bfd, port->up.netdev);
1917 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1919 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1923 ofport_update_peer(port);
1927 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1929 struct ofport_dpif *port = ofport_dpif_cast(port_);
1930 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1931 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1933 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1934 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1935 OFPUTIL_PC_NO_PACKET_IN)) {
1936 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1938 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1939 bundle_update(port->bundle);
1945 set_sflow(struct ofproto *ofproto_,
1946 const struct ofproto_sflow_options *sflow_options)
1948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1949 struct dpif_sflow *ds = ofproto->sflow;
1951 if (sflow_options) {
1953 struct ofport_dpif *ofport;
1955 ds = ofproto->sflow = dpif_sflow_create();
1956 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1957 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1959 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1961 dpif_sflow_set_options(ds, sflow_options);
1964 dpif_sflow_unref(ds);
1965 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1966 ofproto->sflow = NULL;
1974 struct ofproto *ofproto_,
1975 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1976 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1977 size_t n_flow_exporters_options)
1979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1980 struct dpif_ipfix *di = ofproto->ipfix;
1981 bool has_options = bridge_exporter_options || flow_exporters_options;
1983 if (has_options && !di) {
1984 di = ofproto->ipfix = dpif_ipfix_create();
1988 /* Call set_options in any case to cleanly flush the flow
1989 * caches in the last exporters that are to be destroyed. */
1990 dpif_ipfix_set_options(
1991 di, bridge_exporter_options, flow_exporters_options,
1992 n_flow_exporters_options);
1995 dpif_ipfix_unref(di);
1996 ofproto->ipfix = NULL;
2004 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2006 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2013 struct ofproto_dpif *ofproto;
2015 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2016 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2017 ofport->cfm = cfm_create(ofport->up.netdev);
2020 if (cfm_configure(ofport->cfm, s)) {
2026 cfm_unref(ofport->cfm);
2032 get_cfm_status(const struct ofport *ofport_,
2033 struct ofproto_cfm_status *status)
2035 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2038 status->faults = cfm_get_fault(ofport->cfm);
2039 status->remote_opstate = cfm_get_opup(ofport->cfm);
2040 status->health = cfm_get_health(ofport->cfm);
2041 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2049 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2052 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2056 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2057 cfg, ofport->up.netdev);
2058 if (ofport->bfd != old) {
2059 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2066 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2068 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2071 bfd_get_status(ofport->bfd, smap);
2078 /* Spanning Tree. */
2081 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2083 struct ofproto_dpif *ofproto = ofproto_;
2084 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2085 struct ofport_dpif *ofport;
2087 ofport = stp_port_get_aux(sp);
2089 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2090 ofproto->up.name, port_num);
2092 struct eth_header *eth = pkt->l2;
2094 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2095 if (eth_addr_is_zero(eth->eth_src)) {
2096 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2097 "with unknown MAC", ofproto->up.name, port_num);
2099 send_packet(ofport, pkt);
2105 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2107 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2111 /* Only revalidate flows if the configuration changed. */
2112 if (!s != !ofproto->stp) {
2113 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2117 if (!ofproto->stp) {
2118 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2119 send_bpdu_cb, ofproto);
2120 ofproto->stp_last_tick = time_msec();
2123 stp_set_bridge_id(ofproto->stp, s->system_id);
2124 stp_set_bridge_priority(ofproto->stp, s->priority);
2125 stp_set_hello_time(ofproto->stp, s->hello_time);
2126 stp_set_max_age(ofproto->stp, s->max_age);
2127 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2129 struct ofport *ofport;
2131 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2132 set_stp_port(ofport, NULL);
2135 stp_unref(ofproto->stp);
2136 ofproto->stp = NULL;
2143 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2149 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2150 s->designated_root = stp_get_designated_root(ofproto->stp);
2151 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2160 update_stp_port_state(struct ofport_dpif *ofport)
2162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2163 enum stp_state state;
2165 /* Figure out new state. */
2166 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2170 if (ofport->stp_state != state) {
2171 enum ofputil_port_state of_state;
2174 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2175 netdev_get_name(ofport->up.netdev),
2176 stp_state_name(ofport->stp_state),
2177 stp_state_name(state));
2178 if (stp_learn_in_state(ofport->stp_state)
2179 != stp_learn_in_state(state)) {
2180 /* xxx Learning action flows should also be flushed. */
2181 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2182 mac_learning_flush(ofproto->ml);
2183 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2185 fwd_change = stp_forward_in_state(ofport->stp_state)
2186 != stp_forward_in_state(state);
2188 ofproto->backer->need_revalidate = REV_STP;
2189 ofport->stp_state = state;
2190 ofport->stp_state_entered = time_msec();
2192 if (fwd_change && ofport->bundle) {
2193 bundle_update(ofport->bundle);
2196 /* Update the STP state bits in the OpenFlow port description. */
2197 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2198 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2199 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2200 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2201 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2203 ofproto_port_set_state(&ofport->up, of_state);
2207 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2208 * caller is responsible for assigning STP port numbers and ensuring
2209 * there are no duplicates. */
2211 set_stp_port(struct ofport *ofport_,
2212 const struct ofproto_port_stp_settings *s)
2214 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2216 struct stp_port *sp = ofport->stp_port;
2218 if (!s || !s->enable) {
2220 ofport->stp_port = NULL;
2221 stp_port_disable(sp);
2222 update_stp_port_state(ofport);
2225 } else if (sp && stp_port_no(sp) != s->port_num
2226 && ofport == stp_port_get_aux(sp)) {
2227 /* The port-id changed, so disable the old one if it's not
2228 * already in use by another port. */
2229 stp_port_disable(sp);
2232 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2233 stp_port_enable(sp);
2235 stp_port_set_aux(sp, ofport);
2236 stp_port_set_priority(sp, s->priority);
2237 stp_port_set_path_cost(sp, s->path_cost);
2239 update_stp_port_state(ofport);
2245 get_stp_port_status(struct ofport *ofport_,
2246 struct ofproto_port_stp_status *s)
2248 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2250 struct stp_port *sp = ofport->stp_port;
2252 if (!ofproto->stp || !sp) {
2258 s->port_id = stp_port_get_id(sp);
2259 s->state = stp_port_get_state(sp);
2260 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2261 s->role = stp_port_get_role(sp);
2262 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2268 stp_run(struct ofproto_dpif *ofproto)
2271 long long int now = time_msec();
2272 long long int elapsed = now - ofproto->stp_last_tick;
2273 struct stp_port *sp;
2276 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2277 ofproto->stp_last_tick = now;
2279 while (stp_get_changed_port(ofproto->stp, &sp)) {
2280 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2283 update_stp_port_state(ofport);
2287 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2288 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2289 mac_learning_flush(ofproto->ml);
2290 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2296 stp_wait(struct ofproto_dpif *ofproto)
2299 poll_timer_wait(1000);
2304 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2307 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2308 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2310 if (ofport->n_qdscp != n_qdscp
2311 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2312 n_qdscp * sizeof *qdscp))) {
2313 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2314 free(ofport->qdscp);
2315 ofport->qdscp = n_qdscp
2316 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2318 ofport->n_qdscp = n_qdscp;
2326 /* Expires all MAC learning entries associated with 'bundle' and forces its
2327 * ofproto to revalidate every flow.
2329 * Normally MAC learning entries are removed only from the ofproto associated
2330 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2331 * are removed from every ofproto. When patch ports and SLB bonds are in use
2332 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2333 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2334 * with the host from which it migrated. */
2336 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2338 struct ofproto_dpif *ofproto = bundle->ofproto;
2339 struct mac_learning *ml = ofproto->ml;
2340 struct mac_entry *mac, *next_mac;
2342 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2343 ovs_rwlock_wrlock(&ml->rwlock);
2344 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2345 if (mac->port.p == bundle) {
2347 struct ofproto_dpif *o;
2349 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2351 struct mac_entry *e;
2353 ovs_rwlock_wrlock(&o->ml->rwlock);
2354 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2356 mac_learning_expire(o->ml, e);
2358 ovs_rwlock_unlock(&o->ml->rwlock);
2363 mac_learning_expire(ml, mac);
2366 ovs_rwlock_unlock(&ml->rwlock);
2369 static struct ofbundle *
2370 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2372 struct ofbundle *bundle;
2374 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2375 &ofproto->bundles) {
2376 if (bundle->aux == aux) {
2384 bundle_update(struct ofbundle *bundle)
2386 struct ofport_dpif *port;
2388 bundle->floodable = true;
2389 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2390 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2391 || !stp_forward_in_state(port->stp_state)) {
2392 bundle->floodable = false;
2399 bundle_del_port(struct ofport_dpif *port)
2401 struct ofbundle *bundle = port->bundle;
2403 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2405 list_remove(&port->bundle_node);
2406 port->bundle = NULL;
2409 lacp_slave_unregister(bundle->lacp, port);
2412 bond_slave_unregister(bundle->bond, port);
2415 bundle_update(bundle);
2419 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2420 struct lacp_slave_settings *lacp)
2422 struct ofport_dpif *port;
2424 port = get_ofp_port(bundle->ofproto, ofp_port);
2429 if (port->bundle != bundle) {
2430 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2432 bundle_remove(&port->up);
2435 port->bundle = bundle;
2436 list_push_back(&bundle->ports, &port->bundle_node);
2437 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2438 || !stp_forward_in_state(port->stp_state)) {
2439 bundle->floodable = false;
2443 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2444 lacp_slave_register(bundle->lacp, port, lacp);
2451 bundle_destroy(struct ofbundle *bundle)
2453 struct ofproto_dpif *ofproto;
2454 struct ofport_dpif *port, *next_port;
2460 ofproto = bundle->ofproto;
2461 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2463 ovs_rwlock_wrlock(&xlate_rwlock);
2464 xlate_bundle_remove(bundle);
2465 ovs_rwlock_unlock(&xlate_rwlock);
2467 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2468 bundle_del_port(port);
2471 bundle_flush_macs(bundle, true);
2472 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2474 free(bundle->trunks);
2475 lacp_unref(bundle->lacp);
2476 bond_unref(bundle->bond);
2481 bundle_set(struct ofproto *ofproto_, void *aux,
2482 const struct ofproto_bundle_settings *s)
2484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2485 bool need_flush = false;
2486 struct ofport_dpif *port;
2487 struct ofbundle *bundle;
2488 unsigned long *trunks;
2494 bundle_destroy(bundle_lookup(ofproto, aux));
2498 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2499 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2501 bundle = bundle_lookup(ofproto, aux);
2503 bundle = xmalloc(sizeof *bundle);
2505 bundle->ofproto = ofproto;
2506 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2507 hash_pointer(aux, 0));
2509 bundle->name = NULL;
2511 list_init(&bundle->ports);
2512 bundle->vlan_mode = PORT_VLAN_TRUNK;
2514 bundle->trunks = NULL;
2515 bundle->use_priority_tags = s->use_priority_tags;
2516 bundle->lacp = NULL;
2517 bundle->bond = NULL;
2519 bundle->floodable = true;
2520 mbridge_register_bundle(ofproto->mbridge, bundle);
2523 if (!bundle->name || strcmp(s->name, bundle->name)) {
2525 bundle->name = xstrdup(s->name);
2530 if (!bundle->lacp) {
2531 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2532 bundle->lacp = lacp_create();
2534 lacp_configure(bundle->lacp, s->lacp);
2536 lacp_unref(bundle->lacp);
2537 bundle->lacp = NULL;
2540 /* Update set of ports. */
2542 for (i = 0; i < s->n_slaves; i++) {
2543 if (!bundle_add_port(bundle, s->slaves[i],
2544 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2548 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2549 struct ofport_dpif *next_port;
2551 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2552 for (i = 0; i < s->n_slaves; i++) {
2553 if (s->slaves[i] == port->up.ofp_port) {
2558 bundle_del_port(port);
2562 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2564 if (list_is_empty(&bundle->ports)) {
2565 bundle_destroy(bundle);
2569 /* Set VLAN tagging mode */
2570 if (s->vlan_mode != bundle->vlan_mode
2571 || s->use_priority_tags != bundle->use_priority_tags) {
2572 bundle->vlan_mode = s->vlan_mode;
2573 bundle->use_priority_tags = s->use_priority_tags;
2578 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2579 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2581 if (vlan != bundle->vlan) {
2582 bundle->vlan = vlan;
2586 /* Get trunked VLANs. */
2587 switch (s->vlan_mode) {
2588 case PORT_VLAN_ACCESS:
2592 case PORT_VLAN_TRUNK:
2593 trunks = CONST_CAST(unsigned long *, s->trunks);
2596 case PORT_VLAN_NATIVE_UNTAGGED:
2597 case PORT_VLAN_NATIVE_TAGGED:
2598 if (vlan != 0 && (!s->trunks
2599 || !bitmap_is_set(s->trunks, vlan)
2600 || bitmap_is_set(s->trunks, 0))) {
2601 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2603 trunks = bitmap_clone(s->trunks, 4096);
2605 trunks = bitmap_allocate1(4096);
2607 bitmap_set1(trunks, vlan);
2608 bitmap_set0(trunks, 0);
2610 trunks = CONST_CAST(unsigned long *, s->trunks);
2617 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2618 free(bundle->trunks);
2619 if (trunks == s->trunks) {
2620 bundle->trunks = vlan_bitmap_clone(trunks);
2622 bundle->trunks = trunks;
2627 if (trunks != s->trunks) {
2632 if (!list_is_short(&bundle->ports)) {
2633 bundle->ofproto->has_bonded_bundles = true;
2635 if (bond_reconfigure(bundle->bond, s->bond)) {
2636 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2639 bundle->bond = bond_create(s->bond);
2640 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2643 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2644 bond_slave_register(bundle->bond, port, port->up.netdev);
2647 bond_unref(bundle->bond);
2648 bundle->bond = NULL;
2651 /* If we changed something that would affect MAC learning, un-learn
2652 * everything on this port and force flow revalidation. */
2654 bundle_flush_macs(bundle, false);
2661 bundle_remove(struct ofport *port_)
2663 struct ofport_dpif *port = ofport_dpif_cast(port_);
2664 struct ofbundle *bundle = port->bundle;
2667 bundle_del_port(port);
2668 if (list_is_empty(&bundle->ports)) {
2669 bundle_destroy(bundle);
2670 } else if (list_is_short(&bundle->ports)) {
2671 bond_unref(bundle->bond);
2672 bundle->bond = NULL;
2678 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2680 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2681 struct ofport_dpif *port = port_;
2682 uint8_t ea[ETH_ADDR_LEN];
2685 error = netdev_get_etheraddr(port->up.netdev, ea);
2687 struct ofpbuf packet;
2690 ofpbuf_init(&packet, 0);
2691 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2693 memcpy(packet_pdu, pdu, pdu_size);
2695 send_packet(port, &packet);
2696 ofpbuf_uninit(&packet);
2698 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2699 "%s (%s)", port->bundle->name,
2700 netdev_get_name(port->up.netdev), ovs_strerror(error));
2705 bundle_send_learning_packets(struct ofbundle *bundle)
2707 struct ofproto_dpif *ofproto = bundle->ofproto;
2708 int error, n_packets, n_errors;
2709 struct mac_entry *e;
2711 error = n_packets = n_errors = 0;
2712 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2713 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2714 if (e->port.p != bundle) {
2715 struct ofpbuf *learning_packet;
2716 struct ofport_dpif *port;
2720 /* The assignment to "port" is unnecessary but makes "grep"ing for
2721 * struct ofport_dpif more effective. */
2722 learning_packet = bond_compose_learning_packet(bundle->bond,
2726 ret = send_packet(port, learning_packet);
2727 ofpbuf_delete(learning_packet);
2735 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2738 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2739 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2740 "packets, last error was: %s",
2741 bundle->name, n_errors, n_packets, ovs_strerror(error));
2743 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2744 bundle->name, n_packets);
2749 bundle_run(struct ofbundle *bundle)
2752 lacp_run(bundle->lacp, send_pdu_cb);
2755 struct ofport_dpif *port;
2757 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2758 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2761 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2762 bundle->ofproto->backer->need_revalidate = REV_BOND;
2765 if (bond_should_send_learning_packets(bundle->bond)) {
2766 bundle_send_learning_packets(bundle);
2772 bundle_wait(struct ofbundle *bundle)
2775 lacp_wait(bundle->lacp);
2778 bond_wait(bundle->bond);
2785 mirror_set__(struct ofproto *ofproto_, void *aux,
2786 const struct ofproto_mirror_settings *s)
2788 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2789 struct ofbundle **srcs, **dsts;
2794 mirror_destroy(ofproto->mbridge, aux);
2798 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2799 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2801 for (i = 0; i < s->n_srcs; i++) {
2802 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2805 for (i = 0; i < s->n_dsts; i++) {
2806 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2809 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2810 s->n_dsts, s->src_vlans,
2811 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2818 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2819 uint64_t *packets, uint64_t *bytes)
2822 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2827 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2830 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2831 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2832 mac_learning_flush(ofproto->ml);
2834 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2839 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2842 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2843 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2847 forward_bpdu_changed(struct ofproto *ofproto_)
2849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2850 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2854 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2858 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2859 mac_learning_set_idle_time(ofproto->ml, idle_time);
2860 mac_learning_set_max_entries(ofproto->ml, max_entries);
2861 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2866 static struct ofport_dpif *
2867 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2869 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2870 return ofport ? ofport_dpif_cast(ofport) : NULL;
2873 static struct ofport_dpif *
2874 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2876 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2877 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2881 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2882 struct ofproto_port *ofproto_port,
2883 struct dpif_port *dpif_port)
2885 ofproto_port->name = dpif_port->name;
2886 ofproto_port->type = dpif_port->type;
2887 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2891 ofport_update_peer(struct ofport_dpif *ofport)
2893 const struct ofproto_dpif *ofproto;
2894 struct dpif_backer *backer;
2897 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2901 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2902 backer->need_revalidate = REV_RECONFIGURE;
2905 ofport->peer->peer = NULL;
2906 ofport->peer = NULL;
2909 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2915 struct ofport *peer_ofport;
2916 struct ofport_dpif *peer;
2919 if (ofproto->backer != backer) {
2923 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2928 peer = ofport_dpif_cast(peer_ofport);
2929 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2930 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2932 ofport->peer = peer;
2933 ofport->peer->peer = ofport;
2943 port_run_fast(struct ofport_dpif *ofport)
2945 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2946 struct ofpbuf packet;
2948 ofpbuf_init(&packet, 0);
2949 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2950 send_packet(ofport, &packet);
2951 ofpbuf_uninit(&packet);
2954 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2955 struct ofpbuf packet;
2957 ofpbuf_init(&packet, 0);
2958 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2959 send_packet(ofport, &packet);
2960 ofpbuf_uninit(&packet);
2965 port_run(struct ofport_dpif *ofport)
2967 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2968 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2969 bool enable = netdev_get_carrier(ofport->up.netdev);
2970 bool cfm_enable = false;
2971 bool bfd_enable = false;
2973 ofport->carrier_seq = carrier_seq;
2975 port_run_fast(ofport);
2978 int cfm_opup = cfm_get_opup(ofport->cfm);
2980 cfm_run(ofport->cfm);
2981 cfm_enable = !cfm_get_fault(ofport->cfm);
2983 if (cfm_opup >= 0) {
2984 cfm_enable = cfm_enable && cfm_opup;
2989 bfd_run(ofport->bfd);
2990 bfd_enable = bfd_forwarding(ofport->bfd);
2993 if (ofport->bfd || ofport->cfm) {
2994 enable = enable && (cfm_enable || bfd_enable);
2997 if (ofport->bundle) {
2998 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2999 if (carrier_changed) {
3000 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3004 if (ofport->may_enable != enable) {
3005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3006 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3009 ofport->may_enable = enable;
3013 port_wait(struct ofport_dpif *ofport)
3016 cfm_wait(ofport->cfm);
3020 bfd_wait(ofport->bfd);
3025 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3026 struct ofproto_port *ofproto_port)
3028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3029 struct dpif_port dpif_port;
3032 if (sset_contains(&ofproto->ghost_ports, devname)) {
3033 const char *type = netdev_get_type_from_name(devname);
3035 /* We may be called before ofproto->up.port_by_name is populated with
3036 * the appropriate ofport. For this reason, we must get the name and
3037 * type from the netdev layer directly. */
3039 const struct ofport *ofport;
3041 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3042 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3043 ofproto_port->name = xstrdup(devname);
3044 ofproto_port->type = xstrdup(type);
3050 if (!sset_contains(&ofproto->ports, devname)) {
3053 error = dpif_port_query_by_name(ofproto->backer->dpif,
3054 devname, &dpif_port);
3056 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3062 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3065 const char *devname = netdev_get_name(netdev);
3066 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3067 const char *dp_port_name;
3069 if (netdev_vport_is_patch(netdev)) {
3070 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3074 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3075 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3076 odp_port_t port_no = ODPP_NONE;
3079 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3083 if (netdev_get_tunnel_config(netdev)) {
3084 simap_put(&ofproto->backer->tnl_backers,
3085 dp_port_name, odp_to_u32(port_no));
3089 if (netdev_get_tunnel_config(netdev)) {
3090 sset_add(&ofproto->ghost_ports, devname);
3092 sset_add(&ofproto->ports, devname);
3098 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3101 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3108 sset_find_and_delete(&ofproto->ghost_ports,
3109 netdev_get_name(ofport->up.netdev));
3110 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3111 if (!ofport->is_tunnel) {
3112 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3114 /* The caller is going to close ofport->up.netdev. If this is a
3115 * bonded port, then the bond is using that netdev, so remove it
3116 * from the bond. The client will need to reconfigure everything
3117 * after deleting ports, so then the slave will get re-added. */
3118 bundle_remove(&ofport->up);
3125 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3127 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3132 error = netdev_get_stats(ofport->up.netdev, stats);
3134 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3137 /* ofproto->stats.tx_packets represents packets that we created
3138 * internally and sent to some port (e.g. packets sent with
3139 * send_packet()). Account for them as if they had come from
3140 * OFPP_LOCAL and got forwarded. */
3142 if (stats->rx_packets != UINT64_MAX) {
3143 stats->rx_packets += ofproto->stats.tx_packets;
3146 if (stats->rx_bytes != UINT64_MAX) {
3147 stats->rx_bytes += ofproto->stats.tx_bytes;
3150 /* ofproto->stats.rx_packets represents packets that were received on
3151 * some port and we processed internally and dropped (e.g. STP).
3152 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3154 if (stats->tx_packets != UINT64_MAX) {
3155 stats->tx_packets += ofproto->stats.rx_packets;
3158 if (stats->tx_bytes != UINT64_MAX) {
3159 stats->tx_bytes += ofproto->stats.rx_bytes;
3166 struct port_dump_state {
3171 struct ofproto_port port;
3176 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3178 *statep = xzalloc(sizeof(struct port_dump_state));
3183 port_dump_next(const struct ofproto *ofproto_, void *state_,
3184 struct ofproto_port *port)
3186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3187 struct port_dump_state *state = state_;
3188 const struct sset *sset;
3189 struct sset_node *node;
3191 if (state->has_port) {
3192 ofproto_port_destroy(&state->port);
3193 state->has_port = false;
3195 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3196 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3199 error = port_query_by_name(ofproto_, node->name, &state->port);
3201 *port = state->port;
3202 state->has_port = true;
3204 } else if (error != ENODEV) {
3209 if (!state->ghost) {
3210 state->ghost = true;
3213 return port_dump_next(ofproto_, state_, port);
3220 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3222 struct port_dump_state *state = state_;
3224 if (state->has_port) {
3225 ofproto_port_destroy(&state->port);
3232 port_poll(const struct ofproto *ofproto_, char **devnamep)
3234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3236 if (ofproto->port_poll_errno) {
3237 int error = ofproto->port_poll_errno;
3238 ofproto->port_poll_errno = 0;
3242 if (sset_is_empty(&ofproto->port_poll_set)) {
3246 *devnamep = sset_pop(&ofproto->port_poll_set);
3251 port_poll_wait(const struct ofproto *ofproto_)
3253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3254 dpif_port_poll_wait(ofproto->backer->dpif);
3258 port_is_lacp_current(const struct ofport *ofport_)
3260 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3261 return (ofport->bundle && ofport->bundle->lacp
3262 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3266 /* Upcall handling. */
3268 struct flow_miss_op {
3269 struct dpif_op dpif_op;
3271 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3272 struct xlate_out xout;
3273 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3275 struct ofpbuf mask; /* Flow mask for "put" ops. */
3276 struct odputil_keybuf maskbuf;
3278 /* If this is a "put" op, then a pointer to the subfacet that should
3279 * be marked as uninstalled if the operation fails. */
3280 struct subfacet *subfacet;
3283 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3284 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3285 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3286 * return value of true). However, for short flows the cost of bookkeeping is
3287 * much higher than the benefits, so when the datapath holds a large number of
3288 * flows we impose some heuristics to decide which flows are likely to be worth
3291 flow_miss_should_make_facet(struct flow_miss *miss)
3293 struct dpif_backer *backer = miss->ofproto->backer;
3296 switch (flow_miss_model) {
3297 case OFPROTO_HANDLE_MISS_AUTO:
3299 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3301 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3305 if (!backer->governor) {
3308 n_subfacets = hmap_count(&backer->subfacets);
3309 if (n_subfacets * 2 <= flow_eviction_threshold) {
3313 backer->governor = governor_create();
3316 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3317 return governor_should_install_flow(backer->governor, hash,
3318 list_size(&miss->packets));
3321 /* Handles 'miss', which matches 'facet'. May add any required datapath
3322 * operations to 'ops', incrementing '*n_ops' for each new op.
3324 * All of the packets in 'miss' are considered to have arrived at time
3325 * 'miss->stats.used'. This is really important only for new facets: if we
3326 * just called time_msec() here, then the new subfacet or its packets could
3327 * look (occasionally) as though it was used some time after the facet was
3328 * used. That can make a one-packet flow look like it has a nonzero duration,
3329 * which looks odd in e.g. NetFlow statistics. */
3331 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3332 struct flow_miss_op *ops, size_t *n_ops)
3334 enum subfacet_path want_path;
3335 struct subfacet *subfacet;
3337 facet->packet_count += miss->stats.n_packets;
3338 facet->prev_packet_count += miss->stats.n_packets;
3339 facet->byte_count += miss->stats.n_bytes;
3340 facet->prev_byte_count += miss->stats.n_bytes;
3342 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3344 /* Don't install the flow if it's the result of the "userspace"
3345 * action for an already installed facet. This can occur when a
3346 * datapath flow with wildcards has a "userspace" action and flows
3347 * sent to userspace result in a different subfacet, which will then
3348 * be rejected as overlapping by the datapath. */
3349 if (miss->upcall_type == DPIF_UC_ACTION
3350 && !list_is_empty(&facet->subfacets)) {
3354 subfacet = subfacet_create(facet, miss);
3355 if (subfacet->path != want_path) {
3356 struct flow_miss_op *op = &ops[(*n_ops)++];
3357 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3359 subfacet->path = want_path;
3361 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3362 if (enable_megaflows) {
3363 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3364 &miss->flow, UINT32_MAX);
3367 op->xout_garbage = false;
3368 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3369 op->subfacet = subfacet;
3370 put->flags = DPIF_FP_CREATE;
3371 put->key = miss->key;
3372 put->key_len = miss->key_len;
3373 put->mask = op->mask.data;
3374 put->mask_len = op->mask.size;
3376 if (want_path == SF_FAST_PATH) {
3377 put->actions = facet->xout.odp_actions.data;
3378 put->actions_len = facet->xout.odp_actions.size;
3380 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3381 op->slow_stub, sizeof op->slow_stub,
3382 &put->actions, &put->actions_len);
3388 /* Handles flow miss 'miss'. May add any required datapath operations
3389 * to 'ops', incrementing '*n_ops' for each new op. */
3391 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3394 struct facet *facet;
3396 miss->ofproto->n_missed += list_size(&miss->packets);
3398 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3400 /* There does not exist a bijection between 'struct flow' and datapath
3401 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3402 * assumption used throughout the facet and subfacet handling code.
3403 * Since we have to handle these misses in userspace anyway, we simply
3404 * skip facet creation, avoiding the problem altogether. */
3405 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3406 || !flow_miss_should_make_facet(miss)) {
3410 facet = facet_create(miss);
3412 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3415 static struct drop_key *
3416 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3419 struct drop_key *drop_key;
3421 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3422 &backer->drop_keys) {
3423 if (drop_key->key_len == key_len
3424 && !memcmp(drop_key->key, key, key_len)) {
3432 drop_key_clear(struct dpif_backer *backer)
3434 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3435 struct drop_key *drop_key, *next;
3437 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3440 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3442 if (error && !VLOG_DROP_WARN(&rl)) {
3443 struct ds ds = DS_EMPTY_INITIALIZER;
3444 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3445 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3446 ovs_strerror(error), ds_cstr(&ds));
3450 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3451 drop_key_destroy(drop_key);
3454 udpif_drop_key_clear(backer->udpif);
3458 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3460 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3461 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3462 struct flow_miss *miss;
3465 /* Process each element in the to-do list, constructing the set of
3466 * operations to batch. */
3468 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3469 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3471 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3473 /* Execute batch. */
3474 for (i = 0; i < n_ops; i++) {
3475 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3477 dpif_operate(backer->dpif, dpif_ops, n_ops);
3479 for (i = 0; i < n_ops; i++) {
3480 if (dpif_ops[i]->error != 0
3481 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3482 && flow_miss_ops[i].subfacet) {
3483 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3485 COVERAGE_INC(subfacet_install_fail);
3487 /* Zero-out subfacet counters when installation failed, but
3488 * datapath reported hits. This should not happen and
3489 * indicates a bug, since if the datapath flow exists, we
3490 * should not be attempting to create a new subfacet. A
3491 * buggy datapath could trigger this, so just zero out the
3492 * counters and log an error. */
3493 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3494 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3495 "datapath reported hits");
3496 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3499 subfacet->path = SF_NOT_INSTALLED;
3505 handle_sflow_upcall(struct dpif_backer *backer,
3506 const struct dpif_upcall *upcall)
3508 struct ofproto_dpif *ofproto;
3509 union user_action_cookie cookie;
3511 odp_port_t odp_in_port;
3513 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3514 &flow, NULL, &ofproto, &odp_in_port)
3515 || !ofproto->sflow) {
3519 memset(&cookie, 0, sizeof cookie);
3520 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3521 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3522 odp_in_port, &cookie);
3526 handle_flow_sample_upcall(struct dpif_backer *backer,
3527 const struct dpif_upcall *upcall)
3529 struct ofproto_dpif *ofproto;
3530 union user_action_cookie cookie;
3533 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3534 &flow, NULL, &ofproto, NULL)
3535 || !ofproto->ipfix) {
3539 memset(&cookie, 0, sizeof cookie);
3540 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3542 /* The flow reflects exactly the contents of the packet. Sample
3543 * the packet using it. */
3544 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3545 cookie.flow_sample.collector_set_id,
3546 cookie.flow_sample.probability,
3547 cookie.flow_sample.obs_domain_id,
3548 cookie.flow_sample.obs_point_id);
3552 handle_ipfix_upcall(struct dpif_backer *backer,
3553 const struct dpif_upcall *upcall)
3555 struct ofproto_dpif *ofproto;
3558 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3559 &flow, NULL, &ofproto, NULL)
3560 || !ofproto->ipfix) {
3564 /* The flow reflects exactly the contents of the packet. Sample
3565 * the packet using it. */
3566 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3570 handle_upcalls(struct dpif_backer *backer)
3572 struct flow_miss_batch *fmb;
3575 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3576 struct upcall *upcall = upcall_next(backer->udpif);
3582 switch (upcall->type) {
3584 handle_sflow_upcall(backer, &upcall->dpif_upcall);
3587 case FLOW_SAMPLE_UPCALL:
3588 handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3592 handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3602 upcall_destroy(upcall);
3605 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3606 struct drop_key *drop_key = drop_key_next(backer->udpif);
3611 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3612 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3613 hash_bytes(drop_key->key, drop_key->key_len, 0));
3614 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3615 drop_key->key, drop_key->key_len,
3616 NULL, 0, NULL, 0, NULL);
3618 drop_key_destroy(drop_key);
3622 fmb = flow_miss_batch_next(backer->udpif);
3624 handle_flow_misses(backer, fmb);
3625 flow_miss_batch_destroy(fmb);
3629 /* Flow expiration. */
3631 static int subfacet_max_idle(const struct dpif_backer *);
3632 static void update_stats(struct dpif_backer *);
3633 static void rule_expire(struct rule_dpif *);
3634 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3636 /* This function is called periodically by run(). Its job is to collect
3637 * updates for the flows that have been installed into the datapath, most
3638 * importantly when they last were used, and then use that information to
3639 * expire flows that have not been used recently.
3641 * Returns the number of milliseconds after which it should be called again. */
3643 expire(struct dpif_backer *backer)
3645 struct ofproto_dpif *ofproto;
3649 /* Periodically clear out the drop keys in an effort to keep them
3650 * relatively few. */
3651 drop_key_clear(backer);
3653 /* Update stats for each flow in the backer. */
3654 update_stats(backer);
3656 n_subfacets = hmap_count(&backer->subfacets);
3658 struct subfacet *subfacet;
3659 long long int total, now;
3663 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3664 total += now - subfacet->created;
3666 backer->avg_subfacet_life += total / n_subfacets;
3668 backer->avg_subfacet_life /= 2;
3670 backer->avg_n_subfacet += n_subfacets;
3671 backer->avg_n_subfacet /= 2;
3673 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3675 max_idle = subfacet_max_idle(backer);
3676 expire_subfacets(backer, max_idle);
3678 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3679 struct rule *rule, *next_rule;
3681 if (ofproto->backer != backer) {
3685 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3687 ovs_mutex_lock(&ofproto->up.expirable_mutex);
3688 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3689 &ofproto->up.expirable) {
3690 rule_expire(rule_dpif_cast(rule));
3692 ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3694 /* All outstanding data in existing flows has been accounted, so it's a
3695 * good time to do bond rebalancing. */
3696 if (ofproto->has_bonded_bundles) {
3697 struct ofbundle *bundle;
3699 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3701 bond_rebalance(bundle->bond);
3707 return MIN(max_idle, 1000);
3710 /* Updates flow table statistics given that the datapath just reported 'stats'
3711 * as 'subfacet''s statistics. */
3713 update_subfacet_stats(struct subfacet *subfacet,
3714 const struct dpif_flow_stats *stats)
3716 struct facet *facet = subfacet->facet;
3717 struct dpif_flow_stats diff;
3719 diff.tcp_flags = stats->tcp_flags;
3720 diff.used = stats->used;
3722 if (stats->n_packets >= subfacet->dp_packet_count) {
3723 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3725 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3729 if (stats->n_bytes >= subfacet->dp_byte_count) {
3730 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3732 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3736 facet->ofproto->n_hit += diff.n_packets;
3737 subfacet->dp_packet_count = stats->n_packets;
3738 subfacet->dp_byte_count = stats->n_bytes;
3739 subfacet_update_stats(subfacet, &diff);
3741 if (facet->accounted_bytes < facet->byte_count) {
3743 facet_account(facet);
3744 facet->accounted_bytes = facet->byte_count;
3748 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3749 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3751 delete_unexpected_flow(struct dpif_backer *backer,
3752 const struct nlattr *key, size_t key_len)
3754 if (!VLOG_DROP_WARN(&rl)) {
3758 odp_flow_key_format(key, key_len, &s);
3759 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3763 COVERAGE_INC(facet_unexpected);
3764 dpif_flow_del(backer->dpif, key, key_len, NULL);
3767 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3769 * This function also pushes statistics updates to rules which each facet
3770 * resubmits into. Generally these statistics will be accurate. However, if a
3771 * facet changes the rule it resubmits into at some time in between
3772 * update_stats() runs, it is possible that statistics accrued to the
3773 * old rule will be incorrectly attributed to the new rule. This could be
3774 * avoided by calling update_stats() whenever rules are created or
3775 * deleted. However, the performance impact of making so many calls to the
3776 * datapath do not justify the benefit of having perfectly accurate statistics.
3778 * In addition, this function maintains per ofproto flow hit counts. The patch
3779 * port is not treated specially. e.g. A packet ingress from br0 patched into
3780 * br1 will increase the hit count of br0 by 1, however, does not affect
3781 * the hit or miss counts of br1.
3784 update_stats(struct dpif_backer *backer)
3786 const struct dpif_flow_stats *stats;
3787 struct dpif_flow_dump dump;
3788 const struct nlattr *key, *mask;
3789 size_t key_len, mask_len;
3791 dpif_flow_dump_start(&dump, backer->dpif);
3792 while (dpif_flow_dump_next(&dump, &key, &key_len,
3793 &mask, &mask_len, NULL, NULL, &stats)) {
3794 struct subfacet *subfacet;
3797 key_hash = odp_flow_key_hash(key, key_len);
3798 subfacet = subfacet_find(backer, key, key_len, key_hash);
3799 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3801 update_subfacet_stats(subfacet, stats);
3805 /* Stats are updated per-packet. */
3808 case SF_NOT_INSTALLED:
3810 delete_unexpected_flow(backer, key, key_len);
3815 dpif_flow_dump_done(&dump);
3817 update_moving_averages(backer);
3820 /* Calculates and returns the number of milliseconds of idle time after which
3821 * subfacets should expire from the datapath. When a subfacet expires, we fold
3822 * its statistics into its facet, and when a facet's last subfacet expires, we
3823 * fold its statistic into its rule. */
3825 subfacet_max_idle(const struct dpif_backer *backer)
3828 * Idle time histogram.
3830 * Most of the time a switch has a relatively small number of subfacets.
3831 * When this is the case we might as well keep statistics for all of them
3832 * in userspace and to cache them in the kernel datapath for performance as
3835 * As the number of subfacets increases, the memory required to maintain
3836 * statistics about them in userspace and in the kernel becomes
3837 * significant. However, with a large number of subfacets it is likely
3838 * that only a few of them are "heavy hitters" that consume a large amount
3839 * of bandwidth. At this point, only heavy hitters are worth caching in
3840 * the kernel and maintaining in userspaces; other subfacets we can
3843 * The technique used to compute the idle time is to build a histogram with
3844 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3845 * that is installed in the kernel gets dropped in the appropriate bucket.
3846 * After the histogram has been built, we compute the cutoff so that only
3847 * the most-recently-used 1% of subfacets (but at least
3848 * flow_eviction_threshold flows) are kept cached. At least
3849 * the most-recently-used bucket of subfacets is kept, so actually an
3850 * arbitrary number of subfacets can be kept in any given expiration run
3851 * (though the next run will delete most of those unless they receive
3854 * This requires a second pass through the subfacets, in addition to the
3855 * pass made by update_stats(), because the former function never looks at
3856 * uninstallable subfacets.
3858 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3859 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3860 int buckets[N_BUCKETS] = { 0 };
3861 int total, subtotal, bucket;
3862 struct subfacet *subfacet;
3866 total = hmap_count(&backer->subfacets);
3867 if (total <= flow_eviction_threshold) {
3868 return N_BUCKETS * BUCKET_WIDTH;
3871 /* Build histogram. */
3873 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3874 long long int idle = now - subfacet->used;
3875 int bucket = (idle <= 0 ? 0
3876 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3877 : (unsigned int) idle / BUCKET_WIDTH);
3881 /* Find the first bucket whose flows should be expired. */
3882 subtotal = bucket = 0;
3884 subtotal += buckets[bucket++];
3885 } while (bucket < N_BUCKETS &&
3886 subtotal < MAX(flow_eviction_threshold, total / 100));
3888 if (VLOG_IS_DBG_ENABLED()) {
3892 ds_put_cstr(&s, "keep");
3893 for (i = 0; i < N_BUCKETS; i++) {
3895 ds_put_cstr(&s, ", drop");
3898 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3901 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3905 return bucket * BUCKET_WIDTH;
3909 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3911 /* Cutoff time for most flows. */
3912 long long int normal_cutoff = time_msec() - dp_max_idle;
3914 /* We really want to keep flows for special protocols around, so use a more
3915 * conservative cutoff. */
3916 long long int special_cutoff = time_msec() - 10000;
3918 struct subfacet *subfacet, *next_subfacet;
3919 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3923 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3924 &backer->subfacets) {
3925 long long int cutoff;
3927 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3931 if (subfacet->used < cutoff) {
3932 if (subfacet->path != SF_NOT_INSTALLED) {
3933 batch[n_batch++] = subfacet;
3934 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3935 subfacet_destroy_batch(backer, batch, n_batch);
3939 subfacet_destroy(subfacet);
3945 subfacet_destroy_batch(backer, batch, n_batch);
3949 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3950 * then delete it entirely. */
3952 rule_expire(struct rule_dpif *rule)
3954 uint16_t idle_timeout, hard_timeout;
3958 if (rule->up.pending) {
3959 /* We'll have to expire it later. */
3963 ovs_mutex_lock(&rule->up.timeout_mutex);
3964 hard_timeout = rule->up.hard_timeout;
3965 idle_timeout = rule->up.idle_timeout;
3966 ovs_mutex_unlock(&rule->up.timeout_mutex);
3968 /* Has 'rule' expired? */
3970 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3971 reason = OFPRR_HARD_TIMEOUT;
3972 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3973 reason = OFPRR_IDLE_TIMEOUT;
3978 COVERAGE_INC(ofproto_dpif_expired);
3979 ofproto_rule_expire(&rule->up, reason);
3984 /* Creates and returns a new facet based on 'miss'.
3986 * The caller must already have determined that no facet with an identical
3987 * 'miss->flow' exists in 'miss->ofproto'.
3989 * 'rule' and 'xout' must have been created based on 'miss'.
3991 * 'facet'' statistics are initialized based on 'stats'.
3993 * The facet will initially have no subfacets. The caller should create (at
3994 * least) one subfacet with subfacet_create(). */
3995 static struct facet *
3996 facet_create(const struct flow_miss *miss)
3998 struct ofproto_dpif *ofproto = miss->ofproto;
3999 struct facet *facet;
4002 facet = xzalloc(sizeof *facet);
4003 facet->ofproto = miss->ofproto;
4004 facet->used = miss->stats.used;
4005 facet->flow = miss->flow;
4006 facet->learn_rl = time_msec() + 500;
4008 list_init(&facet->subfacets);
4009 netflow_flow_init(&facet->nf_flow);
4010 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4012 xlate_out_copy(&facet->xout, &miss->xout);
4014 match_init(&match, &facet->flow, &facet->xout.wc);
4015 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4016 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4017 classifier_insert(&ofproto->facets, &facet->cr);
4018 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4020 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4025 facet_free(struct facet *facet)
4028 xlate_out_uninit(&facet->xout);
4033 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4034 * 'packet', which arrived on 'in_port'. */
4036 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4037 const struct nlattr *odp_actions, size_t actions_len,
4038 struct ofpbuf *packet)
4040 struct odputil_keybuf keybuf;
4044 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4045 odp_flow_key_from_flow(&key, flow,
4046 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4048 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4049 odp_actions, actions_len, packet);
4053 /* Remove 'facet' from its ofproto and free up the associated memory:
4055 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4056 * rule's statistics, via subfacet_uninstall().
4058 * - Removes 'facet' from its rule and from ofproto->facets.
4061 facet_remove(struct facet *facet)
4063 struct subfacet *subfacet, *next_subfacet;
4065 ovs_assert(!list_is_empty(&facet->subfacets));
4067 /* First uninstall all of the subfacets to get final statistics. */
4068 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4069 subfacet_uninstall(subfacet);
4072 /* Flush the final stats to the rule.
4074 * This might require us to have at least one subfacet around so that we
4075 * can use its actions for accounting in facet_account(), which is why we
4076 * have uninstalled but not yet destroyed the subfacets. */
4077 facet_flush_stats(facet);
4079 /* Now we're really all done so destroy everything. */
4080 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4081 &facet->subfacets) {
4082 subfacet_destroy__(subfacet);
4084 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4085 classifier_remove(&facet->ofproto->facets, &facet->cr);
4086 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4087 cls_rule_destroy(&facet->cr);
4091 /* Feed information from 'facet' back into the learning table to keep it in
4092 * sync with what is actually flowing through the datapath. */
4094 facet_learn(struct facet *facet)
4096 long long int now = time_msec();
4098 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4102 facet->learn_rl = now + 500;
4104 if (!facet->xout.has_learn
4105 && !facet->xout.has_normal
4106 && (!facet->xout.has_fin_timeout
4107 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4111 facet_push_stats(facet, true);
4115 facet_account(struct facet *facet)
4117 const struct nlattr *a;
4122 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4125 n_bytes = facet->byte_count - facet->accounted_bytes;
4127 /* This loop feeds byte counters to bond_account() for rebalancing to use
4128 * as a basis. We also need to track the actual VLAN on which the packet
4129 * is going to be sent to ensure that it matches the one passed to
4130 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4133 * We use the actions from an arbitrary subfacet because they should all
4134 * be equally valid for our purpose. */
4135 vlan_tci = facet->flow.vlan_tci;
4136 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4137 facet->xout.odp_actions.size) {
4138 const struct ovs_action_push_vlan *vlan;
4139 struct ofport_dpif *port;
4141 switch (nl_attr_type(a)) {
4142 case OVS_ACTION_ATTR_OUTPUT:
4143 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4144 if (port && port->bundle && port->bundle->bond) {
4145 bond_account(port->bundle->bond, &facet->flow,
4146 vlan_tci_to_vid(vlan_tci), n_bytes);
4150 case OVS_ACTION_ATTR_POP_VLAN:
4151 vlan_tci = htons(0);
4154 case OVS_ACTION_ATTR_PUSH_VLAN:
4155 vlan = nl_attr_get(a);
4156 vlan_tci = vlan->vlan_tci;
4162 /* Returns true if the only action for 'facet' is to send to the controller.
4163 * (We don't report NetFlow expiration messages for such facets because they
4164 * are just part of the control logic for the network, not real traffic). */
4166 facet_is_controller_flow(struct facet *facet)
4169 struct ofproto_dpif *ofproto = facet->ofproto;
4170 const struct ofpact *ofpacts;
4171 struct rule_dpif *rule;
4175 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4176 ofpacts_len = rule->up.ofpacts_len;
4177 ofpacts = rule->up.ofpacts;
4178 is_controller = ofpacts_len > 0
4179 && ofpacts->type == OFPACT_CONTROLLER
4180 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4182 return is_controller;
4187 /* Folds all of 'facet''s statistics into its rule. Also updates the
4188 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4189 * 'facet''s statistics in the datapath should have been zeroed and folded into
4190 * its packet and byte counts before this function is called. */
4192 facet_flush_stats(struct facet *facet)
4194 struct ofproto_dpif *ofproto = facet->ofproto;
4195 struct subfacet *subfacet;
4197 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4198 ovs_assert(!subfacet->dp_byte_count);
4199 ovs_assert(!subfacet->dp_packet_count);
4202 facet_push_stats(facet, false);
4203 if (facet->accounted_bytes < facet->byte_count) {
4204 facet_account(facet);
4205 facet->accounted_bytes = facet->byte_count;
4208 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4209 struct ofexpired expired;
4210 expired.flow = facet->flow;
4211 expired.packet_count = facet->packet_count;
4212 expired.byte_count = facet->byte_count;
4213 expired.used = facet->used;
4214 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4217 /* Reset counters to prevent double counting if 'facet' ever gets
4219 facet_reset_counters(facet);
4221 netflow_flow_clear(&facet->nf_flow);
4222 facet->tcp_flags = 0;
4225 /* Searches 'ofproto''s table of facets for one which would be responsible for
4226 * 'flow'. Returns it if found, otherwise a null pointer.
4228 * The returned facet might need revalidation; use facet_lookup_valid()
4229 * instead if that is important. */
4230 static struct facet *
4231 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4233 struct cls_rule *cr;
4235 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4236 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4237 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4238 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4241 /* Searches 'ofproto''s table of facets for one capable that covers
4242 * 'flow'. Returns it if found, otherwise a null pointer.
4244 * The returned facet is guaranteed to be valid. */
4245 static struct facet *
4246 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4248 struct facet *facet;
4250 facet = facet_find(ofproto, flow);
4252 && ofproto->backer->need_revalidate
4253 && !facet_revalidate(facet)) {
4261 facet_check_consistency(struct facet *facet)
4263 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4265 struct xlate_out xout;
4266 struct xlate_in xin;
4268 struct rule_dpif *rule;
4271 /* Check the datapath actions for consistency. */
4272 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4273 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4274 xlate_actions(&xin, &xout);
4277 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4278 && facet->xout.slow == xout.slow;
4279 if (!ok && !VLOG_DROP_WARN(&rl)) {
4280 struct ds s = DS_EMPTY_INITIALIZER;
4282 flow_format(&s, &facet->flow);
4283 ds_put_cstr(&s, ": inconsistency in facet");
4285 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4286 ds_put_cstr(&s, " (actions were: ");
4287 format_odp_actions(&s, facet->xout.odp_actions.data,
4288 facet->xout.odp_actions.size);
4289 ds_put_cstr(&s, ") (correct actions: ");
4290 format_odp_actions(&s, xout.odp_actions.data,
4291 xout.odp_actions.size);
4292 ds_put_char(&s, ')');
4295 if (facet->xout.slow != xout.slow) {
4296 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4301 xlate_out_uninit(&xout);
4306 /* Re-searches the classifier for 'facet':
4308 * - If the rule found is different from 'facet''s current rule, moves
4309 * 'facet' to the new rule and recompiles its actions.
4311 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4312 * where it is and recompiles its actions anyway.
4314 * - If any of 'facet''s subfacets correspond to a new flow according to
4315 * xlate_receive(), 'facet' is removed.
4317 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4319 facet_revalidate(struct facet *facet)
4321 struct ofproto_dpif *ofproto = facet->ofproto;
4322 struct rule_dpif *new_rule;
4323 struct subfacet *subfacet;
4324 struct flow_wildcards wc;
4325 struct xlate_out xout;
4326 struct xlate_in xin;
4328 COVERAGE_INC(facet_revalidate);
4330 /* Check that child subfacets still correspond to this facet. Tunnel
4331 * configuration changes could cause a subfacet's OpenFlow in_port to
4333 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4334 struct ofproto_dpif *recv_ofproto;
4335 struct flow recv_flow;
4338 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4339 subfacet->key_len, &recv_flow, NULL,
4340 &recv_ofproto, NULL);
4342 || recv_ofproto != ofproto
4343 || facet != facet_find(ofproto, &recv_flow)) {
4344 facet_remove(facet);
4349 flow_wildcards_init_catchall(&wc);
4350 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4352 /* Calculate new datapath actions.
4354 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4355 * emit a NetFlow expiration and, if so, we need to have the old state
4356 * around to properly compose it. */
4357 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4358 xlate_actions(&xin, &xout);
4359 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4361 /* A facet's slow path reason should only change under dramatic
4362 * circumstances. Rather than try to update everything, it's simpler to
4363 * remove the facet and start over.
4365 * More importantly, if a facet's wildcards change, it will be relatively
4366 * difficult to figure out if its subfacets still belong to it, and if not
4367 * which facet they may belong to. Again, to avoid the complexity, we
4368 * simply give up instead. */
4369 if (facet->xout.slow != xout.slow
4370 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4371 facet_remove(facet);
4372 xlate_out_uninit(&xout);
4373 rule_release(new_rule);
4377 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4378 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4379 if (subfacet->path == SF_FAST_PATH) {
4380 struct dpif_flow_stats stats;
4382 subfacet_install(subfacet, &xout.odp_actions, &stats);
4383 subfacet_update_stats(subfacet, &stats);
4387 facet_flush_stats(facet);
4389 ofpbuf_clear(&facet->xout.odp_actions);
4390 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4391 xout.odp_actions.size);
4394 /* Update 'facet' now that we've taken care of all the old state. */
4395 facet->xout.slow = xout.slow;
4396 facet->xout.has_learn = xout.has_learn;
4397 facet->xout.has_normal = xout.has_normal;
4398 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4399 facet->xout.nf_output_iface = xout.nf_output_iface;
4400 facet->xout.mirrors = xout.mirrors;
4401 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4402 facet->used = MAX(facet->used, new_rule->up.created);
4404 xlate_out_uninit(&xout);
4405 rule_release(new_rule);
4410 facet_reset_counters(struct facet *facet)
4412 facet->packet_count = 0;
4413 facet->byte_count = 0;
4414 facet->prev_packet_count = 0;
4415 facet->prev_byte_count = 0;
4416 facet->accounted_bytes = 0;
4420 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4421 struct dpif_flow_stats *stats, bool may_learn)
4423 struct ofport_dpif *in_port;
4424 struct rule_dpif *rule;
4425 struct xlate_in xin;
4427 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4428 if (in_port && in_port->is_tunnel) {
4429 netdev_vport_inc_rx(in_port->up.netdev, stats);
4432 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4433 rule_credit_stats(rule, stats);
4434 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4435 xin.resubmit_stats = stats;
4436 xin.may_learn = may_learn;
4437 xlate_actions_for_side_effects(&xin);
4442 facet_push_stats(struct facet *facet, bool may_learn)
4444 struct dpif_flow_stats stats;
4446 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4447 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4448 ovs_assert(facet->used >= facet->prev_used);
4450 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4451 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4452 stats.used = facet->used;
4453 stats.tcp_flags = facet->tcp_flags;
4455 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4456 facet->prev_packet_count = facet->packet_count;
4457 facet->prev_byte_count = facet->byte_count;
4458 facet->prev_used = facet->used;
4460 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4462 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4463 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4464 stats.n_packets, stats.n_bytes);
4465 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4470 push_all_stats__(bool run_fast)
4472 static long long int rl = LLONG_MIN;
4473 struct ofproto_dpif *ofproto;
4475 if (time_msec() < rl) {
4479 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4480 struct cls_cursor cursor;
4481 struct facet *facet;
4483 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4484 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4485 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4486 facet_push_stats(facet, false);
4491 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4494 rl = time_msec() + 100;
4498 push_all_stats(void)
4500 push_all_stats__(true);
4504 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4506 ovs_mutex_lock(&rule->stats_mutex);
4507 rule->packet_count += stats->n_packets;
4508 rule->byte_count += stats->n_bytes;
4509 ofproto_rule_update_used(&rule->up, stats->used);
4510 ovs_mutex_unlock(&rule->stats_mutex);
4515 static struct subfacet *
4516 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4517 size_t key_len, uint32_t key_hash)
4519 struct subfacet *subfacet;
4521 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4522 &backer->subfacets) {
4523 if (subfacet->key_len == key_len
4524 && !memcmp(key, subfacet->key, key_len)) {
4532 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4533 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4534 * existing subfacet if there is one, otherwise creates and returns a
4536 static struct subfacet *
4537 subfacet_create(struct facet *facet, struct flow_miss *miss)
4539 struct dpif_backer *backer = miss->ofproto->backer;
4540 enum odp_key_fitness key_fitness = miss->key_fitness;
4541 const struct nlattr *key = miss->key;
4542 size_t key_len = miss->key_len;
4544 struct subfacet *subfacet;
4546 key_hash = odp_flow_key_hash(key, key_len);
4548 if (list_is_empty(&facet->subfacets)) {
4549 subfacet = &facet->one_subfacet;
4551 subfacet = subfacet_find(backer, key, key_len, key_hash);
4553 if (subfacet->facet == facet) {
4557 /* This shouldn't happen. */
4558 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4559 subfacet_destroy(subfacet);
4562 subfacet = xmalloc(sizeof *subfacet);
4565 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4566 list_push_back(&facet->subfacets, &subfacet->list_node);
4567 subfacet->facet = facet;
4568 subfacet->key_fitness = key_fitness;
4569 subfacet->key = xmemdup(key, key_len);
4570 subfacet->key_len = key_len;
4571 subfacet->used = miss->stats.used;
4572 subfacet->created = subfacet->used;
4573 subfacet->dp_packet_count = 0;
4574 subfacet->dp_byte_count = 0;
4575 subfacet->path = SF_NOT_INSTALLED;
4576 subfacet->backer = backer;
4578 backer->subfacet_add_count++;
4582 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4583 * its facet within 'ofproto', and frees it. */
4585 subfacet_destroy__(struct subfacet *subfacet)
4587 struct facet *facet = subfacet->facet;
4588 struct ofproto_dpif *ofproto = facet->ofproto;
4590 /* Update ofproto stats before uninstall the subfacet. */
4591 ofproto->backer->subfacet_del_count++;
4593 subfacet_uninstall(subfacet);
4594 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4595 list_remove(&subfacet->list_node);
4596 free(subfacet->key);
4597 if (subfacet != &facet->one_subfacet) {
4602 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4603 * last remaining subfacet in its facet destroys the facet too. */
4605 subfacet_destroy(struct subfacet *subfacet)
4607 struct facet *facet = subfacet->facet;
4609 if (list_is_singleton(&facet->subfacets)) {
4610 /* facet_remove() needs at least one subfacet (it will remove it). */
4611 facet_remove(facet);
4613 subfacet_destroy__(subfacet);
4618 subfacet_destroy_batch(struct dpif_backer *backer,
4619 struct subfacet **subfacets, int n)
4621 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4622 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4623 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4626 for (i = 0; i < n; i++) {
4627 ops[i].type = DPIF_OP_FLOW_DEL;
4628 ops[i].u.flow_del.key = subfacets[i]->key;
4629 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4630 ops[i].u.flow_del.stats = &stats[i];
4634 dpif_operate(backer->dpif, opsp, n);
4635 for (i = 0; i < n; i++) {
4636 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4637 subfacets[i]->path = SF_NOT_INSTALLED;
4638 subfacet_destroy(subfacets[i]);
4643 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4644 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4645 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4646 * since 'subfacet' was last updated.
4648 * Returns 0 if successful, otherwise a positive errno value. */
4650 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4651 struct dpif_flow_stats *stats)
4653 struct facet *facet = subfacet->facet;
4654 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4655 const struct nlattr *actions = odp_actions->data;
4656 size_t actions_len = odp_actions->size;
4657 struct odputil_keybuf maskbuf;
4660 uint64_t slow_path_stub[128 / 8];
4661 enum dpif_flow_put_flags flags;
4664 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4667 flags |= DPIF_FP_ZERO_STATS;
4670 if (path == SF_SLOW_PATH) {
4671 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4672 slow_path_stub, sizeof slow_path_stub,
4673 &actions, &actions_len);
4676 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4677 if (enable_megaflows) {
4678 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4679 &facet->flow, UINT32_MAX);
4682 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4683 subfacet->key_len, mask.data, mask.size,
4684 actions, actions_len, stats);
4687 subfacet_reset_dp_stats(subfacet, stats);
4691 COVERAGE_INC(subfacet_install_fail);
4693 subfacet->path = path;
4698 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4700 subfacet_uninstall(struct subfacet *subfacet)
4702 if (subfacet->path != SF_NOT_INSTALLED) {
4703 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4704 struct dpif_flow_stats stats;
4707 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4708 subfacet->key_len, &stats);
4709 subfacet_reset_dp_stats(subfacet, &stats);
4711 subfacet_update_stats(subfacet, &stats);
4713 subfacet->path = SF_NOT_INSTALLED;
4715 ovs_assert(subfacet->dp_packet_count == 0);
4716 ovs_assert(subfacet->dp_byte_count == 0);
4720 /* Resets 'subfacet''s datapath statistics counters. This should be called
4721 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4722 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4723 * was reset in the datapath. 'stats' will be modified to include only
4724 * statistics new since 'subfacet' was last updated. */
4726 subfacet_reset_dp_stats(struct subfacet *subfacet,
4727 struct dpif_flow_stats *stats)
4730 && subfacet->dp_packet_count <= stats->n_packets
4731 && subfacet->dp_byte_count <= stats->n_bytes) {
4732 stats->n_packets -= subfacet->dp_packet_count;
4733 stats->n_bytes -= subfacet->dp_byte_count;
4736 subfacet->dp_packet_count = 0;
4737 subfacet->dp_byte_count = 0;
4740 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4742 * Because of the meaning of a subfacet's counters, it only makes sense to do
4743 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4744 * represents a packet that was sent by hand or if it represents statistics
4745 * that have been cleared out of the datapath. */
4747 subfacet_update_stats(struct subfacet *subfacet,
4748 const struct dpif_flow_stats *stats)
4750 if (stats->n_packets || stats->used > subfacet->used) {
4751 struct facet *facet = subfacet->facet;
4753 subfacet->used = MAX(subfacet->used, stats->used);
4754 facet->used = MAX(facet->used, stats->used);
4755 facet->packet_count += stats->n_packets;
4756 facet->byte_count += stats->n_bytes;
4757 facet->tcp_flags |= stats->tcp_flags;
4763 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4764 * the fields that were relevant as part of the lookup. */
4766 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4767 struct flow_wildcards *wc, struct rule_dpif **rule)
4769 struct ofport_dpif *port;
4771 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4774 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4776 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4777 flow->in_port.ofp_port);
4780 *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4781 ofproto->no_packet_in_rule);
4782 ovs_rwlock_rdlock(&(*rule)->up.evict);
4786 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4787 const struct flow *flow, struct flow_wildcards *wc,
4788 uint8_t table_id, struct rule_dpif **rule)
4789 OVS_TRY_RDLOCK(true, (*rule)->up.evict)
4791 struct cls_rule *cls_rule;
4792 struct classifier *cls;
4796 if (table_id >= N_TABLES) {
4801 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4802 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4805 cls = &ofproto->up.tables[table_id].cls;
4806 ovs_rwlock_rdlock(&cls->rwlock);
4807 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4808 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4809 /* We must pretend that transport ports are unavailable. */
4810 struct flow ofpc_normal_flow = *flow;
4811 ofpc_normal_flow.tp_src = htons(0);
4812 ofpc_normal_flow.tp_dst = htons(0);
4813 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4814 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4815 cls_rule = &ofproto->drop_frags_rule->up.cr;
4817 flow_wildcards_init_exact(wc);
4820 cls_rule = classifier_lookup(cls, flow, wc);
4823 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4824 if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4825 /* The rule is in the process of being removed. Best we can do is
4826 * pretend it isn't there. */
4829 ovs_rwlock_unlock(&cls->rwlock);
4831 return *rule != NULL;
4834 /* Given a port configuration (specified as zero if there's no port), chooses
4835 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4836 * flow table miss. */
4838 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4839 struct rule_dpif *no_packet_in_rule)
4841 return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4845 rule_release(struct rule_dpif *rule)
4846 OVS_NO_THREAD_SAFETY_ANALYSIS
4849 ovs_rwlock_unlock(&rule->up.evict);
4854 complete_operation(struct rule_dpif *rule)
4856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4858 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4860 struct dpif_completion *c = xmalloc(sizeof *c);
4861 c->op = rule->up.pending;
4862 list_push_back(&ofproto->completions, &c->list_node);
4864 ofoperation_complete(rule->up.pending, 0);
4868 static struct rule *
4871 struct rule_dpif *rule = xmalloc(sizeof *rule);
4876 rule_dealloc(struct rule *rule_)
4878 struct rule_dpif *rule = rule_dpif_cast(rule_);
4883 rule_construct(struct rule *rule_)
4885 struct rule_dpif *rule = rule_dpif_cast(rule_);
4886 ovs_mutex_init(&rule->stats_mutex);
4887 ovs_mutex_lock(&rule->stats_mutex);
4888 rule->packet_count = 0;
4889 rule->byte_count = 0;
4890 ovs_mutex_unlock(&rule->stats_mutex);
4895 rule_insert(struct rule *rule_)
4897 struct rule_dpif *rule = rule_dpif_cast(rule_);
4898 complete_operation(rule);
4902 rule_delete(struct rule *rule_)
4904 struct rule_dpif *rule = rule_dpif_cast(rule_);
4905 complete_operation(rule);
4909 rule_destruct(struct rule *rule_)
4911 struct rule_dpif *rule = rule_dpif_cast(rule_);
4912 ovs_mutex_destroy(&rule->stats_mutex);
4916 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4918 struct rule_dpif *rule = rule_dpif_cast(rule_);
4920 /* push_all_stats() can handle flow misses which, when using the learn
4921 * action, can cause rules to be added and deleted. This can corrupt our
4922 * caller's datastructures which assume that rule_get_stats() doesn't have
4923 * an impact on the flow table. To be safe, we disable miss handling. */
4924 push_all_stats__(false);
4926 /* Start from historical data for 'rule' itself that are no longer tracked
4927 * in facets. This counts, for example, facets that have expired. */
4928 ovs_mutex_lock(&rule->stats_mutex);
4929 *packets = rule->packet_count;
4930 *bytes = rule->byte_count;
4931 ovs_mutex_unlock(&rule->stats_mutex);
4935 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4936 struct ofpbuf *packet)
4938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4939 struct dpif_flow_stats stats;
4940 struct xlate_out xout;
4941 struct xlate_in xin;
4943 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4944 rule_credit_stats(rule, &stats);
4946 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4947 xin.resubmit_stats = &stats;
4948 xlate_actions(&xin, &xout);
4950 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4951 xout.odp_actions.size, packet);
4953 xlate_out_uninit(&xout);
4957 rule_execute(struct rule *rule, const struct flow *flow,
4958 struct ofpbuf *packet)
4960 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4961 ofpbuf_delete(packet);
4966 rule_modify_actions(struct rule *rule_, bool reset_counters)
4968 struct rule_dpif *rule = rule_dpif_cast(rule_);
4970 if (reset_counters) {
4971 ovs_mutex_lock(&rule->stats_mutex);
4972 rule->packet_count = 0;
4973 rule->byte_count = 0;
4974 ovs_mutex_unlock(&rule->stats_mutex);
4977 complete_operation(rule);
4980 /* Sends 'packet' out 'ofport'.
4981 * May modify 'packet'.
4982 * Returns 0 if successful, otherwise a positive errno value. */
4984 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4986 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4987 uint64_t odp_actions_stub[1024 / 8];
4988 struct ofpbuf key, odp_actions;
4989 struct dpif_flow_stats stats;
4990 struct odputil_keybuf keybuf;
4991 struct ofpact_output output;
4992 struct xlate_out xout;
4993 struct xlate_in xin;
4995 union flow_in_port in_port_;
4998 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4999 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5001 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5002 in_port_.ofp_port = OFPP_NONE;
5003 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5004 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5006 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5008 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5009 output.port = ofport->up.ofp_port;
5012 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5013 xin.ofpacts_len = sizeof output;
5014 xin.ofpacts = &output.ofpact;
5015 xin.resubmit_stats = &stats;
5016 xlate_actions(&xin, &xout);
5018 error = dpif_execute(ofproto->backer->dpif,
5020 xout.odp_actions.data, xout.odp_actions.size,
5022 xlate_out_uninit(&xout);
5025 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5026 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5027 ovs_strerror(error));
5030 ofproto->stats.tx_packets++;
5031 ofproto->stats.tx_bytes += packet->size;
5035 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5036 * The action will state 'slow' as the reason that the action is in the slow
5037 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5038 * dump-flows" output to see why a flow is in the slow path.)
5040 * The 'stub_size' bytes in 'stub' will be used to store the action.
5041 * 'stub_size' must be large enough for the action.
5043 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5046 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5047 enum slow_path_reason slow,
5048 uint64_t *stub, size_t stub_size,
5049 const struct nlattr **actionsp, size_t *actions_lenp)
5051 union user_action_cookie cookie;
5054 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5055 cookie.slow_path.unused = 0;
5056 cookie.slow_path.reason = slow;
5058 ofpbuf_use_stack(&buf, stub, stub_size);
5059 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5060 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5062 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5064 odp_port_t odp_port;
5067 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5068 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5069 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5071 *actionsp = buf.data;
5072 *actions_lenp = buf.size;
5076 set_frag_handling(struct ofproto *ofproto_,
5077 enum ofp_config_flags frag_handling)
5079 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5080 if (frag_handling != OFPC_FRAG_REASM) {
5081 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5089 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5090 const struct flow *flow,
5091 const struct ofpact *ofpacts, size_t ofpacts_len)
5093 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5094 struct odputil_keybuf keybuf;
5095 struct dpif_flow_stats stats;
5096 struct xlate_out xout;
5097 struct xlate_in xin;
5101 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5102 odp_flow_key_from_flow(&key, flow,
5103 ofp_port_to_odp_port(ofproto,
5104 flow->in_port.ofp_port));
5106 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5108 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5109 xin.resubmit_stats = &stats;
5110 xin.ofpacts_len = ofpacts_len;
5111 xin.ofpacts = ofpacts;
5113 xlate_actions(&xin, &xout);
5114 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5115 xout.odp_actions.data, xout.odp_actions.size, packet);
5116 xlate_out_uninit(&xout);
5124 set_netflow(struct ofproto *ofproto_,
5125 const struct netflow_options *netflow_options)
5127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5129 if (netflow_options) {
5130 if (!ofproto->netflow) {
5131 ofproto->netflow = netflow_create();
5132 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5134 return netflow_set_options(ofproto->netflow, netflow_options);
5135 } else if (ofproto->netflow) {
5136 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5137 netflow_destroy(ofproto->netflow);
5138 ofproto->netflow = NULL;
5145 get_netflow_ids(const struct ofproto *ofproto_,
5146 uint8_t *engine_type, uint8_t *engine_id)
5148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5150 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5154 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5156 if (!facet_is_controller_flow(facet) &&
5157 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5158 struct subfacet *subfacet;
5159 struct ofexpired expired;
5161 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5162 if (subfacet->path == SF_FAST_PATH) {
5163 struct dpif_flow_stats stats;
5165 subfacet_install(subfacet, &facet->xout.odp_actions,
5167 subfacet_update_stats(subfacet, &stats);
5171 expired.flow = facet->flow;
5172 expired.packet_count = facet->packet_count;
5173 expired.byte_count = facet->byte_count;
5174 expired.used = facet->used;
5175 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5180 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5182 struct cls_cursor cursor;
5183 struct facet *facet;
5185 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5186 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5187 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5188 send_active_timeout(ofproto, facet);
5190 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5193 static struct ofproto_dpif *
5194 ofproto_dpif_lookup(const char *name)
5196 struct ofproto_dpif *ofproto;
5198 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5199 hash_string(name, 0), &all_ofproto_dpifs) {
5200 if (!strcmp(ofproto->up.name, name)) {
5208 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5209 const char *argv[], void *aux OVS_UNUSED)
5211 struct ofproto_dpif *ofproto;
5214 ofproto = ofproto_dpif_lookup(argv[1]);
5216 unixctl_command_reply_error(conn, "no such bridge");
5219 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5220 mac_learning_flush(ofproto->ml);
5221 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5223 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5224 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5225 mac_learning_flush(ofproto->ml);
5226 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5230 unixctl_command_reply(conn, "table successfully flushed");
5233 static struct ofport_dpif *
5234 ofbundle_get_a_port(const struct ofbundle *bundle)
5236 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5241 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5242 const char *argv[], void *aux OVS_UNUSED)
5244 struct ds ds = DS_EMPTY_INITIALIZER;
5245 const struct ofproto_dpif *ofproto;
5246 const struct mac_entry *e;
5248 ofproto = ofproto_dpif_lookup(argv[1]);
5250 unixctl_command_reply_error(conn, "no such bridge");
5254 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5255 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5256 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5257 struct ofbundle *bundle = e->port.p;
5258 char name[OFP_MAX_PORT_NAME_LEN];
5260 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5262 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5263 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5264 mac_entry_age(ofproto->ml, e));
5266 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5267 unixctl_command_reply(conn, ds_cstr(&ds));
5272 struct xlate_out xout;
5273 struct xlate_in xin;
5279 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5281 ds_put_char_multiple(result, '\t', level);
5283 ds_put_cstr(result, "No match\n");
5287 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5288 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5289 cls_rule_format(&rule->up.cr, result);
5290 ds_put_char(result, '\n');
5292 ds_put_char_multiple(result, '\t', level);
5293 ds_put_cstr(result, "OpenFlow ");
5294 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5295 ds_put_char(result, '\n');
5299 trace_format_flow(struct ds *result, int level, const char *title,
5300 struct trace_ctx *trace)
5302 ds_put_char_multiple(result, '\t', level);
5303 ds_put_format(result, "%s: ", title);
5304 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5305 ds_put_cstr(result, "unchanged");
5307 flow_format(result, &trace->xin.flow);
5308 trace->flow = trace->xin.flow;
5310 ds_put_char(result, '\n');
5314 trace_format_regs(struct ds *result, int level, const char *title,
5315 struct trace_ctx *trace)
5319 ds_put_char_multiple(result, '\t', level);
5320 ds_put_format(result, "%s:", title);
5321 for (i = 0; i < FLOW_N_REGS; i++) {
5322 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5324 ds_put_char(result, '\n');
5328 trace_format_odp(struct ds *result, int level, const char *title,
5329 struct trace_ctx *trace)
5331 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5333 ds_put_char_multiple(result, '\t', level);
5334 ds_put_format(result, "%s: ", title);
5335 format_odp_actions(result, odp_actions->data, odp_actions->size);
5336 ds_put_char(result, '\n');
5340 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5342 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5343 struct ds *result = trace->result;
5345 ds_put_char(result, '\n');
5346 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5347 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5348 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5349 trace_format_rule(result, recurse + 1, rule);
5353 trace_report(struct xlate_in *xin, const char *s, int recurse)
5355 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5356 struct ds *result = trace->result;
5358 ds_put_char_multiple(result, '\t', recurse);
5359 ds_put_cstr(result, s);
5360 ds_put_char(result, '\n');
5364 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5365 void *aux OVS_UNUSED)
5367 const struct dpif_backer *backer;
5368 struct ofproto_dpif *ofproto;
5369 struct ofpbuf odp_key, odp_mask;
5370 struct ofpbuf *packet;
5378 ofpbuf_init(&odp_key, 0);
5379 ofpbuf_init(&odp_mask, 0);
5381 /* Handle "-generate" or a hex string as the last argument. */
5382 if (!strcmp(argv[argc - 1], "-generate")) {
5383 packet = ofpbuf_new(0);
5386 const char *error = eth_from_hex(argv[argc - 1], &packet);
5389 } else if (argc == 4) {
5390 /* The 3-argument form must end in "-generate' or a hex string. */
5391 unixctl_command_reply_error(conn, error);
5396 /* Parse the flow and determine whether a datapath or
5397 * bridge is specified. If function odp_flow_key_from_string()
5398 * returns 0, the flow is a odp_flow. If function
5399 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5400 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5401 /* If the odp_flow is the second argument,
5402 * the datapath name is the first argument. */
5404 const char *dp_type;
5405 if (!strncmp(argv[1], "ovs-", 4)) {
5406 dp_type = argv[1] + 4;
5410 backer = shash_find_data(&all_dpif_backers, dp_type);
5412 unixctl_command_reply_error(conn, "Cannot find datapath "
5417 /* No datapath name specified, so there should be only one
5419 struct shash_node *node;
5420 if (shash_count(&all_dpif_backers) != 1) {
5421 unixctl_command_reply_error(conn, "Must specify datapath "
5422 "name, there is more than one type of datapath");
5425 node = shash_first(&all_dpif_backers);
5426 backer = node->data;
5429 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5430 NULL, &ofproto, NULL)) {
5431 unixctl_command_reply_error(conn, "Invalid datapath flow");
5434 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5435 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5437 unixctl_command_reply_error(conn, "Must specify bridge name");
5441 ofproto = ofproto_dpif_lookup(argv[1]);
5443 unixctl_command_reply_error(conn, "Unknown bridge name");
5447 unixctl_command_reply_error(conn, "Bad flow syntax");
5451 /* Generate a packet, if requested. */
5453 if (!packet->size) {
5454 flow_compose(packet, &flow);
5456 union flow_in_port in_port_;
5458 in_port_ = flow.in_port;
5459 ds_put_cstr(&result, "Packet: ");
5460 s = ofp_packet_to_string(packet->data, packet->size);
5461 ds_put_cstr(&result, s);
5464 /* Use the metadata from the flow and the packet argument
5465 * to reconstruct the flow. */
5466 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5471 ofproto_trace(ofproto, &flow, packet, &result);
5472 unixctl_command_reply(conn, ds_cstr(&result));
5475 ds_destroy(&result);
5476 ofpbuf_delete(packet);
5477 ofpbuf_uninit(&odp_key);
5478 ofpbuf_uninit(&odp_mask);
5482 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5483 const struct ofpbuf *packet, struct ds *ds)
5485 struct rule_dpif *rule;
5486 struct flow_wildcards wc;
5488 ds_put_cstr(ds, "Flow: ");
5489 flow_format(ds, flow);
5490 ds_put_char(ds, '\n');
5492 flow_wildcards_init_catchall(&wc);
5493 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5495 trace_format_rule(ds, 0, rule);
5496 if (rule == ofproto->miss_rule) {
5497 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5498 } else if (rule == ofproto->no_packet_in_rule) {
5499 ds_put_cstr(ds, "\nNo match, packets dropped because "
5500 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5501 } else if (rule == ofproto->drop_frags_rule) {
5502 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5503 "and the fragment handling mode is \"drop\".\n");
5507 uint64_t odp_actions_stub[1024 / 8];
5508 struct ofpbuf odp_actions;
5509 struct trace_ctx trace;
5513 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5516 ofpbuf_use_stub(&odp_actions,
5517 odp_actions_stub, sizeof odp_actions_stub);
5518 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5519 trace.xin.resubmit_hook = trace_resubmit;
5520 trace.xin.report_hook = trace_report;
5522 xlate_actions(&trace.xin, &trace.xout);
5523 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5525 ds_put_char(ds, '\n');
5526 trace_format_flow(ds, 0, "Final flow", &trace);
5528 match_init(&match, flow, &trace.xout.wc);
5529 ds_put_cstr(ds, "Relevant fields: ");
5530 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5531 ds_put_char(ds, '\n');
5533 ds_put_cstr(ds, "Datapath actions: ");
5534 format_odp_actions(ds, trace.xout.odp_actions.data,
5535 trace.xout.odp_actions.size);
5537 if (trace.xout.slow) {
5538 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5539 "slow path because it:");
5540 switch (trace.xout.slow) {
5542 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5545 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5548 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5551 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5553 case SLOW_CONTROLLER:
5554 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5555 "to the OpenFlow controller.");
5562 xlate_out_uninit(&trace.xout);
5569 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5570 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5573 unixctl_command_reply(conn, NULL);
5577 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5578 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5581 unixctl_command_reply(conn, NULL);
5584 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5585 * 'reply' describing the results. */
5587 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5589 struct cls_cursor cursor;
5590 struct facet *facet;
5594 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5595 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5596 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5597 if (!facet_check_consistency(facet)) {
5601 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5603 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5607 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5608 ofproto->up.name, errors);
5610 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5615 ofproto_dpif_self_check(struct unixctl_conn *conn,
5616 int argc, const char *argv[], void *aux OVS_UNUSED)
5618 struct ds reply = DS_EMPTY_INITIALIZER;
5619 struct ofproto_dpif *ofproto;
5622 ofproto = ofproto_dpif_lookup(argv[1]);
5624 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5625 "ofproto/list for help)");
5628 ofproto_dpif_self_check__(ofproto, &reply);
5630 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5631 ofproto_dpif_self_check__(ofproto, &reply);
5635 unixctl_command_reply(conn, ds_cstr(&reply));
5639 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5640 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5641 * to destroy 'ofproto_shash' and free the returned value. */
5642 static const struct shash_node **
5643 get_ofprotos(struct shash *ofproto_shash)
5645 const struct ofproto_dpif *ofproto;
5647 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5648 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5649 shash_add_nocopy(ofproto_shash, name, ofproto);
5652 return shash_sort(ofproto_shash);
5656 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5657 const char *argv[] OVS_UNUSED,
5658 void *aux OVS_UNUSED)
5660 struct ds ds = DS_EMPTY_INITIALIZER;
5661 struct shash ofproto_shash;
5662 const struct shash_node **sorted_ofprotos;
5665 shash_init(&ofproto_shash);
5666 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5667 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5668 const struct shash_node *node = sorted_ofprotos[i];
5669 ds_put_format(&ds, "%s\n", node->name);
5672 shash_destroy(&ofproto_shash);
5673 free(sorted_ofprotos);
5675 unixctl_command_reply(conn, ds_cstr(&ds));
5680 show_dp_rates(struct ds *ds, const char *heading,
5681 const struct avg_subfacet_rates *rates)
5683 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5684 heading, rates->add_rate, rates->del_rate);
5688 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5690 const struct shash_node **ofprotos;
5691 struct ofproto_dpif *ofproto;
5692 struct shash ofproto_shash;
5693 uint64_t n_hit, n_missed;
5694 long long int minutes;
5697 n_hit = n_missed = 0;
5698 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5699 if (ofproto->backer == backer) {
5700 n_missed += ofproto->n_missed;
5701 n_hit += ofproto->n_hit;
5705 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5706 dpif_name(backer->dpif), n_hit, n_missed);
5707 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5708 " life span: %lldms\n", hmap_count(&backer->subfacets),
5709 backer->avg_n_subfacet, backer->max_n_subfacet,
5710 backer->avg_subfacet_life);
5712 minutes = (time_msec() - backer->created) / (1000 * 60);
5713 if (minutes >= 60) {
5714 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5716 if (minutes >= 60 * 24) {
5717 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5719 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5721 shash_init(&ofproto_shash);
5722 ofprotos = get_ofprotos(&ofproto_shash);
5723 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5724 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5725 const struct shash_node **ports;
5728 if (ofproto->backer != backer) {
5732 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5733 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5735 ports = shash_sort(&ofproto->up.port_by_name);
5736 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5737 const struct shash_node *node = ports[j];
5738 struct ofport *ofport = node->data;
5740 odp_port_t odp_port;
5742 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5745 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5746 if (odp_port != ODPP_NONE) {
5747 ds_put_format(ds, "%"PRIu32":", odp_port);
5749 ds_put_cstr(ds, "none:");
5752 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5755 if (!netdev_get_config(ofport->netdev, &config)) {
5756 const struct smap_node **nodes;
5759 nodes = smap_sort(&config);
5760 for (i = 0; i < smap_count(&config); i++) {
5761 const struct smap_node *node = nodes[i];
5762 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5763 node->key, node->value);
5767 smap_destroy(&config);
5769 ds_put_char(ds, ')');
5770 ds_put_char(ds, '\n');
5774 shash_destroy(&ofproto_shash);
5779 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5780 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5782 struct ds ds = DS_EMPTY_INITIALIZER;
5783 const struct shash_node **backers;
5786 backers = shash_sort(&all_dpif_backers);
5787 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5788 dpif_show_backer(backers[i]->data, &ds);
5792 unixctl_command_reply(conn, ds_cstr(&ds));
5796 /* Dump the megaflow (facet) cache. This is useful to check the
5797 * correctness of flow wildcarding, since the same mechanism is used for
5798 * both xlate caching and kernel wildcarding.
5800 * It's important to note that in the output the flow description uses
5801 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5803 * This command is only needed for advanced debugging, so it's not
5804 * documented in the man page. */
5806 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5807 int argc OVS_UNUSED, const char *argv[],
5808 void *aux OVS_UNUSED)
5810 struct ds ds = DS_EMPTY_INITIALIZER;
5811 const struct ofproto_dpif *ofproto;
5812 long long int now = time_msec();
5813 struct cls_cursor cursor;
5814 struct facet *facet;
5816 ofproto = ofproto_dpif_lookup(argv[1]);
5818 unixctl_command_reply_error(conn, "no such bridge");
5822 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5823 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5824 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5825 cls_rule_format(&facet->cr, &ds);
5826 ds_put_cstr(&ds, ", ");
5827 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5828 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5829 ds_put_cstr(&ds, "Datapath actions: ");
5830 if (facet->xout.slow) {
5831 uint64_t slow_path_stub[128 / 8];
5832 const struct nlattr *actions;
5835 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5836 slow_path_stub, sizeof slow_path_stub,
5837 &actions, &actions_len);
5838 format_odp_actions(&ds, actions, actions_len);
5840 format_odp_actions(&ds, facet->xout.odp_actions.data,
5841 facet->xout.odp_actions.size);
5843 ds_put_cstr(&ds, "\n");
5845 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5847 ds_chomp(&ds, '\n');
5848 unixctl_command_reply(conn, ds_cstr(&ds));
5852 /* Disable using the megaflows.
5854 * This command is only needed for advanced debugging, so it's not
5855 * documented in the man page. */
5857 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5858 int argc OVS_UNUSED,
5859 const char *argv[] OVS_UNUSED,
5860 void *aux OVS_UNUSED)
5862 struct ofproto_dpif *ofproto;
5864 enable_megaflows = false;
5866 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5867 flush(&ofproto->up);
5870 unixctl_command_reply(conn, "megaflows disabled");
5873 /* Re-enable using megaflows.
5875 * This command is only needed for advanced debugging, so it's not
5876 * documented in the man page. */
5878 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5879 int argc OVS_UNUSED,
5880 const char *argv[] OVS_UNUSED,
5881 void *aux OVS_UNUSED)
5883 struct ofproto_dpif *ofproto;
5885 enable_megaflows = true;
5887 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5888 flush(&ofproto->up);
5891 unixctl_command_reply(conn, "megaflows enabled");
5895 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5896 int argc OVS_UNUSED, const char *argv[],
5897 void *aux OVS_UNUSED)
5899 struct ds ds = DS_EMPTY_INITIALIZER;
5900 const struct ofproto_dpif *ofproto;
5901 struct subfacet *subfacet;
5903 ofproto = ofproto_dpif_lookup(argv[1]);
5905 unixctl_command_reply_error(conn, "no such bridge");
5909 update_stats(ofproto->backer);
5911 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5912 struct facet *facet = subfacet->facet;
5913 struct odputil_keybuf maskbuf;
5916 if (facet->ofproto != ofproto) {
5920 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5921 if (enable_megaflows) {
5922 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5923 &facet->flow, UINT32_MAX);
5926 odp_flow_format(subfacet->key, subfacet->key_len,
5927 mask.data, mask.size, &ds, false);
5929 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5930 subfacet->dp_packet_count, subfacet->dp_byte_count);
5931 if (subfacet->used) {
5932 ds_put_format(&ds, "%.3fs",
5933 (time_msec() - subfacet->used) / 1000.0);
5935 ds_put_format(&ds, "never");
5937 if (subfacet->facet->tcp_flags) {
5938 ds_put_cstr(&ds, ", flags:");
5939 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5942 ds_put_cstr(&ds, ", actions:");
5943 if (facet->xout.slow) {
5944 uint64_t slow_path_stub[128 / 8];
5945 const struct nlattr *actions;
5948 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5949 slow_path_stub, sizeof slow_path_stub,
5950 &actions, &actions_len);
5951 format_odp_actions(&ds, actions, actions_len);
5953 format_odp_actions(&ds, facet->xout.odp_actions.data,
5954 facet->xout.odp_actions.size);
5956 ds_put_char(&ds, '\n');
5959 unixctl_command_reply(conn, ds_cstr(&ds));
5964 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5965 int argc OVS_UNUSED, const char *argv[],
5966 void *aux OVS_UNUSED)
5968 struct ds ds = DS_EMPTY_INITIALIZER;
5969 struct ofproto_dpif *ofproto;
5971 ofproto = ofproto_dpif_lookup(argv[1]);
5973 unixctl_command_reply_error(conn, "no such bridge");
5977 flush(&ofproto->up);
5979 unixctl_command_reply(conn, ds_cstr(&ds));
5984 ofproto_dpif_unixctl_init(void)
5986 static bool registered;
5992 unixctl_command_register(
5994 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5995 1, 3, ofproto_unixctl_trace, NULL);
5996 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5997 ofproto_unixctl_fdb_flush, NULL);
5998 unixctl_command_register("fdb/show", "bridge", 1, 1,
5999 ofproto_unixctl_fdb_show, NULL);
6000 unixctl_command_register("ofproto/clog", "", 0, 0,
6001 ofproto_dpif_clog, NULL);
6002 unixctl_command_register("ofproto/unclog", "", 0, 0,
6003 ofproto_dpif_unclog, NULL);
6004 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6005 ofproto_dpif_self_check, NULL);
6006 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6007 ofproto_unixctl_dpif_dump_dps, NULL);
6008 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6010 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6011 ofproto_unixctl_dpif_dump_flows, NULL);
6012 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6013 ofproto_unixctl_dpif_del_flows, NULL);
6014 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6015 ofproto_unixctl_dpif_dump_megaflows, NULL);
6016 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6017 ofproto_unixctl_dpif_disable_megaflows, NULL);
6018 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6019 ofproto_unixctl_dpif_enable_megaflows, NULL);
6022 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6024 * This is deprecated. It is only for compatibility with broken device drivers
6025 * in old versions of Linux that do not properly support VLANs when VLAN
6026 * devices are not used. When broken device drivers are no longer in
6027 * widespread use, we will delete these interfaces. */
6030 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6033 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6035 if (realdev_ofp_port == ofport->realdev_ofp_port
6036 && vid == ofport->vlandev_vid) {
6040 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6042 if (ofport->realdev_ofp_port) {
6045 if (realdev_ofp_port && ofport->bundle) {
6046 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6047 * themselves be part of a bundle. */
6048 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6051 ofport->realdev_ofp_port = realdev_ofp_port;
6052 ofport->vlandev_vid = vid;
6054 if (realdev_ofp_port) {
6055 vsp_add(ofport, realdev_ofp_port, vid);
6062 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6064 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6068 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6069 OVS_EXCLUDED(ofproto->vsp_mutex)
6073 ovs_mutex_lock(&ofproto->vsp_mutex);
6074 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6075 ovs_mutex_unlock(&ofproto->vsp_mutex);
6080 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6081 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6082 OVS_REQUIRES(ofproto->vsp_mutex)
6084 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6085 int vid = vlan_tci_to_vid(vlan_tci);
6086 const struct vlan_splinter *vsp;
6088 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6089 hash_realdev_vid(realdev_ofp_port, vid),
6090 &ofproto->realdev_vid_map) {
6091 if (vsp->realdev_ofp_port == realdev_ofp_port
6092 && vsp->vid == vid) {
6093 return vsp->vlandev_ofp_port;
6097 return realdev_ofp_port;
6100 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6101 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6102 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6103 * 'vlan_tci' 9, it would return the port number of eth0.9.
6105 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6106 * function just returns its 'realdev_ofp_port' argument. */
6108 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6109 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6110 OVS_EXCLUDED(ofproto->vsp_mutex)
6114 ovs_mutex_lock(&ofproto->vsp_mutex);
6115 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6116 ovs_mutex_unlock(&ofproto->vsp_mutex);
6120 static struct vlan_splinter *
6121 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6123 struct vlan_splinter *vsp;
6125 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6126 hash_ofp_port(vlandev_ofp_port),
6127 &ofproto->vlandev_map) {
6128 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6136 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6137 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6138 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6139 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6140 * eth0 and store 9 in '*vid'.
6142 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6143 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6146 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6147 ofp_port_t vlandev_ofp_port, int *vid)
6148 OVS_REQUIRES(ofproto->vsp_mutex)
6150 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6151 const struct vlan_splinter *vsp;
6153 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6158 return vsp->realdev_ofp_port;
6164 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6165 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6166 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6167 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6168 * always the case unless VLAN splinters are enabled), returns false without
6169 * making any changes. */
6171 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6172 OVS_EXCLUDED(ofproto->vsp_mutex)
6177 ovs_mutex_lock(&ofproto->vsp_mutex);
6178 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6179 ovs_mutex_unlock(&ofproto->vsp_mutex);
6184 /* Cause the flow to be processed as if it came in on the real device with
6185 * the VLAN device's VLAN ID. */
6186 flow->in_port.ofp_port = realdev;
6187 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6192 vsp_remove(struct ofport_dpif *port)
6194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6195 struct vlan_splinter *vsp;
6197 ovs_mutex_lock(&ofproto->vsp_mutex);
6198 vsp = vlandev_find(ofproto, port->up.ofp_port);
6200 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6201 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6204 port->realdev_ofp_port = 0;
6206 VLOG_ERR("missing vlan device record");
6208 ovs_mutex_unlock(&ofproto->vsp_mutex);
6212 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6216 ovs_mutex_lock(&ofproto->vsp_mutex);
6217 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6218 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6219 == realdev_ofp_port)) {
6220 struct vlan_splinter *vsp;
6222 vsp = xmalloc(sizeof *vsp);
6223 vsp->realdev_ofp_port = realdev_ofp_port;
6224 vsp->vlandev_ofp_port = port->up.ofp_port;
6227 port->realdev_ofp_port = realdev_ofp_port;
6229 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6230 hash_ofp_port(port->up.ofp_port));
6231 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6232 hash_realdev_vid(realdev_ofp_port, vid));
6234 VLOG_ERR("duplicate vlan device record");
6236 ovs_mutex_unlock(&ofproto->vsp_mutex);
6240 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6242 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6243 return ofport ? ofport->odp_port : ODPP_NONE;
6246 struct ofport_dpif *
6247 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6249 struct ofport_dpif *port;
6251 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6252 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6253 &backer->odp_to_ofport_map) {
6254 if (port->odp_port == odp_port) {
6255 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6260 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6265 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6267 struct ofport_dpif *port;
6269 port = odp_port_to_ofport(ofproto->backer, odp_port);
6270 if (port && &ofproto->up == port->up.ofproto) {
6271 return port->up.ofp_port;
6277 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6278 * most heavily weighted element. 'base' designates the rate of decay: after
6279 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6282 exp_mavg(double *avg, int base, double new)
6284 *avg = (*avg * (base - 1) + new) / base;
6288 update_moving_averages(struct dpif_backer *backer)
6290 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6291 long long int minutes = (time_msec() - backer->created) / min_ms;
6294 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6296 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6299 backer->lifetime.add_rate = 0.0;
6300 backer->lifetime.del_rate = 0.0;
6303 /* Update hourly averages on the minute boundaries. */
6304 if (time_msec() - backer->last_minute >= min_ms) {
6305 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6306 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6308 /* Update daily averages on the hour boundaries. */
6309 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6310 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6311 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6314 backer->total_subfacet_add_count += backer->subfacet_add_count;
6315 backer->total_subfacet_del_count += backer->subfacet_del_count;
6316 backer->subfacet_add_count = 0;
6317 backer->subfacet_del_count = 0;
6318 backer->last_minute += min_ms;
6322 const struct ofproto_class ofproto_dpif_class = {
6357 port_is_lacp_current,
6358 NULL, /* rule_choose_table */
6367 rule_modify_actions,
6381 get_stp_port_status,
6388 is_mirror_output_bundle,
6389 forward_bpdu_changed,
6390 set_mac_table_config,
6392 NULL, /* meter_get_features */
6393 NULL, /* meter_set */
6394 NULL, /* meter_get */
6395 NULL, /* meter_del */
6396 NULL, /* group_alloc */
6397 NULL, /* group_construct */
6398 NULL, /* group_destruct */
6399 NULL, /* group_dealloc */
6400 NULL, /* group_modify */
6401 NULL, /* group_get_stats */