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 if (ofproto->n_flow_mods) {
1513 flow_mods = ofproto->flow_mods;
1514 list_moved(&flow_mods);
1515 list_init(&ofproto->flow_mods);
1516 ofproto->n_flow_mods = 0;
1518 list_init(&flow_mods);
1520 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1522 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1523 int error = ofproto_flow_mod(&ofproto->up, fm);
1524 if (error && !VLOG_DROP_WARN(&rl)) {
1525 VLOG_WARN("learning action failed to modify flow table (%s)",
1526 ofperr_get_name(error));
1529 list_remove(&fm->list_node);
1534 ovs_mutex_lock(&ofproto->pin_mutex);
1535 if (ofproto->n_pins) {
1536 pins = ofproto->pins;
1538 list_init(&ofproto->pins);
1539 ofproto->n_pins = 0;
1543 ovs_mutex_unlock(&ofproto->pin_mutex);
1545 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1546 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1547 list_remove(&pin->list_node);
1548 free(CONST_CAST(void *, pin->packet));
1552 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1553 port_run_fast(ofport);
1560 run(struct ofproto *ofproto_)
1562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1563 struct ofport_dpif *ofport;
1564 struct ofbundle *bundle;
1568 complete_operations(ofproto);
1571 if (mbridge_need_revalidate(ofproto->mbridge)) {
1572 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1573 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1574 mac_learning_flush(ofproto->ml);
1575 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1578 /* Do not perform any periodic activity below required by 'ofproto' while
1579 * waiting for flow restore to complete. */
1580 if (ofproto_get_flow_restore_wait()) {
1584 error = run_fast(ofproto_);
1589 if (ofproto->netflow) {
1590 if (netflow_run(ofproto->netflow)) {
1591 send_netflow_active_timeouts(ofproto);
1594 if (ofproto->sflow) {
1595 dpif_sflow_run(ofproto->sflow);
1597 if (ofproto->ipfix) {
1598 dpif_ipfix_run(ofproto->ipfix);
1601 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1604 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1609 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1610 if (mac_learning_run(ofproto->ml)) {
1611 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1613 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1615 /* Check the consistency of a random facet, to aid debugging. */
1616 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1617 if (time_msec() >= ofproto->consistency_rl
1618 && !classifier_is_empty(&ofproto->facets)
1619 && !ofproto->backer->need_revalidate) {
1620 struct cls_table *table;
1621 struct cls_rule *cr;
1622 struct facet *facet;
1624 ofproto->consistency_rl = time_msec() + 250;
1626 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1627 struct cls_table, hmap_node);
1628 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1630 facet = CONTAINER_OF(cr, struct facet, cr);
1632 if (!facet_check_consistency(facet)) {
1633 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1636 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1642 wait(struct ofproto *ofproto_)
1644 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1645 struct ofport_dpif *ofport;
1646 struct ofbundle *bundle;
1648 if (!clogged && !list_is_empty(&ofproto->completions)) {
1649 poll_immediate_wake();
1652 if (ofproto_get_flow_restore_wait()) {
1656 if (ofproto->sflow) {
1657 dpif_sflow_wait(ofproto->sflow);
1659 if (ofproto->ipfix) {
1660 dpif_ipfix_wait(ofproto->ipfix);
1662 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1665 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1666 bundle_wait(bundle);
1668 if (ofproto->netflow) {
1669 netflow_wait(ofproto->netflow);
1671 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1672 mac_learning_wait(ofproto->ml);
1673 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1675 if (ofproto->backer->need_revalidate) {
1676 /* Shouldn't happen, but if it does just go around again. */
1677 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1678 poll_immediate_wake();
1683 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1685 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1686 struct cls_cursor cursor;
1687 size_t n_subfacets = 0;
1688 struct facet *facet;
1690 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1691 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1692 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1694 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1695 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1696 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1697 n_subfacets += list_size(&facet->subfacets);
1699 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1700 simap_increase(usage, "subfacets", n_subfacets);
1704 flush(struct ofproto *ofproto_)
1706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1707 struct subfacet *subfacet, *next_subfacet;
1708 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1712 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1713 &ofproto->backer->subfacets) {
1714 if (subfacet->facet->ofproto != ofproto) {
1718 if (subfacet->path != SF_NOT_INSTALLED) {
1719 batch[n_batch++] = subfacet;
1720 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1721 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1725 subfacet_destroy(subfacet);
1730 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1735 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1736 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1738 *arp_match_ip = true;
1739 *actions = (OFPUTIL_A_OUTPUT |
1740 OFPUTIL_A_SET_VLAN_VID |
1741 OFPUTIL_A_SET_VLAN_PCP |
1742 OFPUTIL_A_STRIP_VLAN |
1743 OFPUTIL_A_SET_DL_SRC |
1744 OFPUTIL_A_SET_DL_DST |
1745 OFPUTIL_A_SET_NW_SRC |
1746 OFPUTIL_A_SET_NW_DST |
1747 OFPUTIL_A_SET_NW_TOS |
1748 OFPUTIL_A_SET_TP_SRC |
1749 OFPUTIL_A_SET_TP_DST |
1754 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1757 struct dpif_dp_stats s;
1758 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1761 strcpy(ots->name, "classifier");
1763 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1764 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1765 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1766 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1768 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1769 ots->lookup_count = htonll(n_lookup);
1770 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1773 static struct ofport *
1776 struct ofport_dpif *port = xmalloc(sizeof *port);
1781 port_dealloc(struct ofport *port_)
1783 struct ofport_dpif *port = ofport_dpif_cast(port_);
1788 port_construct(struct ofport *port_)
1790 struct ofport_dpif *port = ofport_dpif_cast(port_);
1791 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1792 const struct netdev *netdev = port->up.netdev;
1793 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1794 struct dpif_port dpif_port;
1797 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1798 port->bundle = NULL;
1801 port->may_enable = true;
1802 port->stp_port = NULL;
1803 port->stp_state = STP_DISABLED;
1804 port->is_tunnel = false;
1808 port->realdev_ofp_port = 0;
1809 port->vlandev_vid = 0;
1810 port->carrier_seq = netdev_get_carrier_resets(netdev);
1812 if (netdev_vport_is_patch(netdev)) {
1813 /* By bailing out here, we don't submit the port to the sFlow module
1814 * to be considered for counter polling export. This is correct
1815 * because the patch port represents an interface that sFlow considers
1816 * to be "internal" to the switch as a whole, and therefore not an
1817 * candidate for counter polling. */
1818 port->odp_port = ODPP_NONE;
1819 ofport_update_peer(port);
1823 error = dpif_port_query_by_name(ofproto->backer->dpif,
1824 netdev_vport_get_dpif_port(netdev, namebuf,
1831 port->odp_port = dpif_port.port_no;
1833 if (netdev_get_tunnel_config(netdev)) {
1834 tnl_port_add(port, port->up.netdev, port->odp_port);
1835 port->is_tunnel = true;
1837 /* Sanity-check that a mapping doesn't already exist. This
1838 * shouldn't happen for non-tunnel ports. */
1839 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1840 VLOG_ERR("port %s already has an OpenFlow port number",
1842 dpif_port_destroy(&dpif_port);
1846 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1847 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1848 hash_odp_port(port->odp_port));
1849 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1851 dpif_port_destroy(&dpif_port);
1853 if (ofproto->sflow) {
1854 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1861 port_destruct(struct ofport *port_)
1863 struct ofport_dpif *port = ofport_dpif_cast(port_);
1864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1865 const char *devname = netdev_get_name(port->up.netdev);
1866 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1867 const char *dp_port_name;
1869 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1870 ovs_rwlock_wrlock(&xlate_rwlock);
1871 xlate_ofport_remove(port);
1872 ovs_rwlock_unlock(&xlate_rwlock);
1874 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1876 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1877 /* The underlying device is still there, so delete it. This
1878 * happens when the ofproto is being destroyed, since the caller
1879 * assumes that removal of attached ports will happen as part of
1881 if (!port->is_tunnel) {
1882 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1887 port->peer->peer = NULL;
1891 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1892 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1893 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1894 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1898 sset_find_and_delete(&ofproto->ports, devname);
1899 sset_find_and_delete(&ofproto->ghost_ports, devname);
1900 bundle_remove(port_);
1901 set_cfm(port_, NULL);
1902 set_bfd(port_, NULL);
1903 if (ofproto->sflow) {
1904 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1911 port_modified(struct ofport *port_)
1913 struct ofport_dpif *port = ofport_dpif_cast(port_);
1915 if (port->bundle && port->bundle->bond) {
1916 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1920 cfm_set_netdev(port->cfm, port->up.netdev);
1924 bfd_set_netdev(port->bfd, port->up.netdev);
1927 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1929 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1933 ofport_update_peer(port);
1937 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1939 struct ofport_dpif *port = ofport_dpif_cast(port_);
1940 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1941 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1943 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1944 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1945 OFPUTIL_PC_NO_PACKET_IN)) {
1946 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1948 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1949 bundle_update(port->bundle);
1955 set_sflow(struct ofproto *ofproto_,
1956 const struct ofproto_sflow_options *sflow_options)
1958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1959 struct dpif_sflow *ds = ofproto->sflow;
1961 if (sflow_options) {
1963 struct ofport_dpif *ofport;
1965 ds = ofproto->sflow = dpif_sflow_create();
1966 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1967 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1969 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1971 dpif_sflow_set_options(ds, sflow_options);
1974 dpif_sflow_unref(ds);
1975 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1976 ofproto->sflow = NULL;
1984 struct ofproto *ofproto_,
1985 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1986 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1987 size_t n_flow_exporters_options)
1989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1990 struct dpif_ipfix *di = ofproto->ipfix;
1991 bool has_options = bridge_exporter_options || flow_exporters_options;
1993 if (has_options && !di) {
1994 di = ofproto->ipfix = dpif_ipfix_create();
1998 /* Call set_options in any case to cleanly flush the flow
1999 * caches in the last exporters that are to be destroyed. */
2000 dpif_ipfix_set_options(
2001 di, bridge_exporter_options, flow_exporters_options,
2002 n_flow_exporters_options);
2005 dpif_ipfix_unref(di);
2006 ofproto->ipfix = NULL;
2014 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2016 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2023 struct ofproto_dpif *ofproto;
2025 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2026 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2027 ofport->cfm = cfm_create(ofport->up.netdev);
2030 if (cfm_configure(ofport->cfm, s)) {
2036 cfm_unref(ofport->cfm);
2042 get_cfm_status(const struct ofport *ofport_,
2043 struct ofproto_cfm_status *status)
2045 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2048 status->faults = cfm_get_fault(ofport->cfm);
2049 status->remote_opstate = cfm_get_opup(ofport->cfm);
2050 status->health = cfm_get_health(ofport->cfm);
2051 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2059 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2061 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2062 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2066 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2067 cfg, ofport->up.netdev);
2068 if (ofport->bfd != old) {
2069 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2076 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2078 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2081 bfd_get_status(ofport->bfd, smap);
2088 /* Spanning Tree. */
2091 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2093 struct ofproto_dpif *ofproto = ofproto_;
2094 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2095 struct ofport_dpif *ofport;
2097 ofport = stp_port_get_aux(sp);
2099 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2100 ofproto->up.name, port_num);
2102 struct eth_header *eth = pkt->l2;
2104 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2105 if (eth_addr_is_zero(eth->eth_src)) {
2106 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2107 "with unknown MAC", ofproto->up.name, port_num);
2109 send_packet(ofport, pkt);
2115 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2117 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2121 /* Only revalidate flows if the configuration changed. */
2122 if (!s != !ofproto->stp) {
2123 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2127 if (!ofproto->stp) {
2128 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2129 send_bpdu_cb, ofproto);
2130 ofproto->stp_last_tick = time_msec();
2133 stp_set_bridge_id(ofproto->stp, s->system_id);
2134 stp_set_bridge_priority(ofproto->stp, s->priority);
2135 stp_set_hello_time(ofproto->stp, s->hello_time);
2136 stp_set_max_age(ofproto->stp, s->max_age);
2137 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2139 struct ofport *ofport;
2141 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2142 set_stp_port(ofport, NULL);
2145 stp_unref(ofproto->stp);
2146 ofproto->stp = NULL;
2153 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2159 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2160 s->designated_root = stp_get_designated_root(ofproto->stp);
2161 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2170 update_stp_port_state(struct ofport_dpif *ofport)
2172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2173 enum stp_state state;
2175 /* Figure out new state. */
2176 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2180 if (ofport->stp_state != state) {
2181 enum ofputil_port_state of_state;
2184 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2185 netdev_get_name(ofport->up.netdev),
2186 stp_state_name(ofport->stp_state),
2187 stp_state_name(state));
2188 if (stp_learn_in_state(ofport->stp_state)
2189 != stp_learn_in_state(state)) {
2190 /* xxx Learning action flows should also be flushed. */
2191 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2192 mac_learning_flush(ofproto->ml);
2193 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2195 fwd_change = stp_forward_in_state(ofport->stp_state)
2196 != stp_forward_in_state(state);
2198 ofproto->backer->need_revalidate = REV_STP;
2199 ofport->stp_state = state;
2200 ofport->stp_state_entered = time_msec();
2202 if (fwd_change && ofport->bundle) {
2203 bundle_update(ofport->bundle);
2206 /* Update the STP state bits in the OpenFlow port description. */
2207 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2208 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2209 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2210 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2211 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2213 ofproto_port_set_state(&ofport->up, of_state);
2217 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2218 * caller is responsible for assigning STP port numbers and ensuring
2219 * there are no duplicates. */
2221 set_stp_port(struct ofport *ofport_,
2222 const struct ofproto_port_stp_settings *s)
2224 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2225 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2226 struct stp_port *sp = ofport->stp_port;
2228 if (!s || !s->enable) {
2230 ofport->stp_port = NULL;
2231 stp_port_disable(sp);
2232 update_stp_port_state(ofport);
2235 } else if (sp && stp_port_no(sp) != s->port_num
2236 && ofport == stp_port_get_aux(sp)) {
2237 /* The port-id changed, so disable the old one if it's not
2238 * already in use by another port. */
2239 stp_port_disable(sp);
2242 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2243 stp_port_enable(sp);
2245 stp_port_set_aux(sp, ofport);
2246 stp_port_set_priority(sp, s->priority);
2247 stp_port_set_path_cost(sp, s->path_cost);
2249 update_stp_port_state(ofport);
2255 get_stp_port_status(struct ofport *ofport_,
2256 struct ofproto_port_stp_status *s)
2258 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2260 struct stp_port *sp = ofport->stp_port;
2262 if (!ofproto->stp || !sp) {
2268 s->port_id = stp_port_get_id(sp);
2269 s->state = stp_port_get_state(sp);
2270 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2271 s->role = stp_port_get_role(sp);
2272 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2278 stp_run(struct ofproto_dpif *ofproto)
2281 long long int now = time_msec();
2282 long long int elapsed = now - ofproto->stp_last_tick;
2283 struct stp_port *sp;
2286 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2287 ofproto->stp_last_tick = now;
2289 while (stp_get_changed_port(ofproto->stp, &sp)) {
2290 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2293 update_stp_port_state(ofport);
2297 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2298 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2299 mac_learning_flush(ofproto->ml);
2300 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2306 stp_wait(struct ofproto_dpif *ofproto)
2309 poll_timer_wait(1000);
2314 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2317 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2318 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2320 if (ofport->n_qdscp != n_qdscp
2321 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2322 n_qdscp * sizeof *qdscp))) {
2323 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2324 free(ofport->qdscp);
2325 ofport->qdscp = n_qdscp
2326 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2328 ofport->n_qdscp = n_qdscp;
2336 /* Expires all MAC learning entries associated with 'bundle' and forces its
2337 * ofproto to revalidate every flow.
2339 * Normally MAC learning entries are removed only from the ofproto associated
2340 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2341 * are removed from every ofproto. When patch ports and SLB bonds are in use
2342 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2343 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2344 * with the host from which it migrated. */
2346 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2348 struct ofproto_dpif *ofproto = bundle->ofproto;
2349 struct mac_learning *ml = ofproto->ml;
2350 struct mac_entry *mac, *next_mac;
2352 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2353 ovs_rwlock_wrlock(&ml->rwlock);
2354 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2355 if (mac->port.p == bundle) {
2357 struct ofproto_dpif *o;
2359 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2361 struct mac_entry *e;
2363 ovs_rwlock_wrlock(&o->ml->rwlock);
2364 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2366 mac_learning_expire(o->ml, e);
2368 ovs_rwlock_unlock(&o->ml->rwlock);
2373 mac_learning_expire(ml, mac);
2376 ovs_rwlock_unlock(&ml->rwlock);
2379 static struct ofbundle *
2380 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2382 struct ofbundle *bundle;
2384 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2385 &ofproto->bundles) {
2386 if (bundle->aux == aux) {
2394 bundle_update(struct ofbundle *bundle)
2396 struct ofport_dpif *port;
2398 bundle->floodable = true;
2399 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2400 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2401 || !stp_forward_in_state(port->stp_state)) {
2402 bundle->floodable = false;
2409 bundle_del_port(struct ofport_dpif *port)
2411 struct ofbundle *bundle = port->bundle;
2413 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2415 list_remove(&port->bundle_node);
2416 port->bundle = NULL;
2419 lacp_slave_unregister(bundle->lacp, port);
2422 bond_slave_unregister(bundle->bond, port);
2425 bundle_update(bundle);
2429 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2430 struct lacp_slave_settings *lacp)
2432 struct ofport_dpif *port;
2434 port = get_ofp_port(bundle->ofproto, ofp_port);
2439 if (port->bundle != bundle) {
2440 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2442 bundle_remove(&port->up);
2445 port->bundle = bundle;
2446 list_push_back(&bundle->ports, &port->bundle_node);
2447 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2448 || !stp_forward_in_state(port->stp_state)) {
2449 bundle->floodable = false;
2453 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2454 lacp_slave_register(bundle->lacp, port, lacp);
2461 bundle_destroy(struct ofbundle *bundle)
2463 struct ofproto_dpif *ofproto;
2464 struct ofport_dpif *port, *next_port;
2470 ofproto = bundle->ofproto;
2471 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2473 ovs_rwlock_wrlock(&xlate_rwlock);
2474 xlate_bundle_remove(bundle);
2475 ovs_rwlock_unlock(&xlate_rwlock);
2477 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2478 bundle_del_port(port);
2481 bundle_flush_macs(bundle, true);
2482 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2484 free(bundle->trunks);
2485 lacp_unref(bundle->lacp);
2486 bond_unref(bundle->bond);
2491 bundle_set(struct ofproto *ofproto_, void *aux,
2492 const struct ofproto_bundle_settings *s)
2494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2495 bool need_flush = false;
2496 struct ofport_dpif *port;
2497 struct ofbundle *bundle;
2498 unsigned long *trunks;
2504 bundle_destroy(bundle_lookup(ofproto, aux));
2508 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2509 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2511 bundle = bundle_lookup(ofproto, aux);
2513 bundle = xmalloc(sizeof *bundle);
2515 bundle->ofproto = ofproto;
2516 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2517 hash_pointer(aux, 0));
2519 bundle->name = NULL;
2521 list_init(&bundle->ports);
2522 bundle->vlan_mode = PORT_VLAN_TRUNK;
2524 bundle->trunks = NULL;
2525 bundle->use_priority_tags = s->use_priority_tags;
2526 bundle->lacp = NULL;
2527 bundle->bond = NULL;
2529 bundle->floodable = true;
2530 mbridge_register_bundle(ofproto->mbridge, bundle);
2533 if (!bundle->name || strcmp(s->name, bundle->name)) {
2535 bundle->name = xstrdup(s->name);
2540 if (!bundle->lacp) {
2541 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2542 bundle->lacp = lacp_create();
2544 lacp_configure(bundle->lacp, s->lacp);
2546 lacp_unref(bundle->lacp);
2547 bundle->lacp = NULL;
2550 /* Update set of ports. */
2552 for (i = 0; i < s->n_slaves; i++) {
2553 if (!bundle_add_port(bundle, s->slaves[i],
2554 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2558 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2559 struct ofport_dpif *next_port;
2561 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2562 for (i = 0; i < s->n_slaves; i++) {
2563 if (s->slaves[i] == port->up.ofp_port) {
2568 bundle_del_port(port);
2572 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2574 if (list_is_empty(&bundle->ports)) {
2575 bundle_destroy(bundle);
2579 /* Set VLAN tagging mode */
2580 if (s->vlan_mode != bundle->vlan_mode
2581 || s->use_priority_tags != bundle->use_priority_tags) {
2582 bundle->vlan_mode = s->vlan_mode;
2583 bundle->use_priority_tags = s->use_priority_tags;
2588 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2589 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2591 if (vlan != bundle->vlan) {
2592 bundle->vlan = vlan;
2596 /* Get trunked VLANs. */
2597 switch (s->vlan_mode) {
2598 case PORT_VLAN_ACCESS:
2602 case PORT_VLAN_TRUNK:
2603 trunks = CONST_CAST(unsigned long *, s->trunks);
2606 case PORT_VLAN_NATIVE_UNTAGGED:
2607 case PORT_VLAN_NATIVE_TAGGED:
2608 if (vlan != 0 && (!s->trunks
2609 || !bitmap_is_set(s->trunks, vlan)
2610 || bitmap_is_set(s->trunks, 0))) {
2611 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2613 trunks = bitmap_clone(s->trunks, 4096);
2615 trunks = bitmap_allocate1(4096);
2617 bitmap_set1(trunks, vlan);
2618 bitmap_set0(trunks, 0);
2620 trunks = CONST_CAST(unsigned long *, s->trunks);
2627 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2628 free(bundle->trunks);
2629 if (trunks == s->trunks) {
2630 bundle->trunks = vlan_bitmap_clone(trunks);
2632 bundle->trunks = trunks;
2637 if (trunks != s->trunks) {
2642 if (!list_is_short(&bundle->ports)) {
2643 bundle->ofproto->has_bonded_bundles = true;
2645 if (bond_reconfigure(bundle->bond, s->bond)) {
2646 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2649 bundle->bond = bond_create(s->bond);
2650 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2653 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2654 bond_slave_register(bundle->bond, port, port->up.netdev);
2657 bond_unref(bundle->bond);
2658 bundle->bond = NULL;
2661 /* If we changed something that would affect MAC learning, un-learn
2662 * everything on this port and force flow revalidation. */
2664 bundle_flush_macs(bundle, false);
2671 bundle_remove(struct ofport *port_)
2673 struct ofport_dpif *port = ofport_dpif_cast(port_);
2674 struct ofbundle *bundle = port->bundle;
2677 bundle_del_port(port);
2678 if (list_is_empty(&bundle->ports)) {
2679 bundle_destroy(bundle);
2680 } else if (list_is_short(&bundle->ports)) {
2681 bond_unref(bundle->bond);
2682 bundle->bond = NULL;
2688 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2690 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2691 struct ofport_dpif *port = port_;
2692 uint8_t ea[ETH_ADDR_LEN];
2695 error = netdev_get_etheraddr(port->up.netdev, ea);
2697 struct ofpbuf packet;
2700 ofpbuf_init(&packet, 0);
2701 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2703 memcpy(packet_pdu, pdu, pdu_size);
2705 send_packet(port, &packet);
2706 ofpbuf_uninit(&packet);
2708 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2709 "%s (%s)", port->bundle->name,
2710 netdev_get_name(port->up.netdev), ovs_strerror(error));
2715 bundle_send_learning_packets(struct ofbundle *bundle)
2717 struct ofproto_dpif *ofproto = bundle->ofproto;
2718 int error, n_packets, n_errors;
2719 struct mac_entry *e;
2721 error = n_packets = n_errors = 0;
2722 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2723 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2724 if (e->port.p != bundle) {
2725 struct ofpbuf *learning_packet;
2726 struct ofport_dpif *port;
2730 /* The assignment to "port" is unnecessary but makes "grep"ing for
2731 * struct ofport_dpif more effective. */
2732 learning_packet = bond_compose_learning_packet(bundle->bond,
2736 ret = send_packet(port, learning_packet);
2737 ofpbuf_delete(learning_packet);
2745 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2748 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2749 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2750 "packets, last error was: %s",
2751 bundle->name, n_errors, n_packets, ovs_strerror(error));
2753 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2754 bundle->name, n_packets);
2759 bundle_run(struct ofbundle *bundle)
2762 lacp_run(bundle->lacp, send_pdu_cb);
2765 struct ofport_dpif *port;
2767 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2768 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2771 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2772 bundle->ofproto->backer->need_revalidate = REV_BOND;
2775 if (bond_should_send_learning_packets(bundle->bond)) {
2776 bundle_send_learning_packets(bundle);
2782 bundle_wait(struct ofbundle *bundle)
2785 lacp_wait(bundle->lacp);
2788 bond_wait(bundle->bond);
2795 mirror_set__(struct ofproto *ofproto_, void *aux,
2796 const struct ofproto_mirror_settings *s)
2798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2799 struct ofbundle **srcs, **dsts;
2804 mirror_destroy(ofproto->mbridge, aux);
2808 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2809 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2811 for (i = 0; i < s->n_srcs; i++) {
2812 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2815 for (i = 0; i < s->n_dsts; i++) {
2816 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2819 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2820 s->n_dsts, s->src_vlans,
2821 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2828 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2829 uint64_t *packets, uint64_t *bytes)
2832 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2837 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2840 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2841 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2842 mac_learning_flush(ofproto->ml);
2844 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2849 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2852 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2853 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2857 forward_bpdu_changed(struct ofproto *ofproto_)
2859 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2860 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2864 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2868 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2869 mac_learning_set_idle_time(ofproto->ml, idle_time);
2870 mac_learning_set_max_entries(ofproto->ml, max_entries);
2871 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2876 static struct ofport_dpif *
2877 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2879 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2880 return ofport ? ofport_dpif_cast(ofport) : NULL;
2883 static struct ofport_dpif *
2884 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2886 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2887 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2891 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2892 struct ofproto_port *ofproto_port,
2893 struct dpif_port *dpif_port)
2895 ofproto_port->name = dpif_port->name;
2896 ofproto_port->type = dpif_port->type;
2897 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2901 ofport_update_peer(struct ofport_dpif *ofport)
2903 const struct ofproto_dpif *ofproto;
2904 struct dpif_backer *backer;
2907 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2911 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2912 backer->need_revalidate = REV_RECONFIGURE;
2915 ofport->peer->peer = NULL;
2916 ofport->peer = NULL;
2919 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2924 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2925 struct ofport *peer_ofport;
2926 struct ofport_dpif *peer;
2929 if (ofproto->backer != backer) {
2933 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2938 peer = ofport_dpif_cast(peer_ofport);
2939 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2940 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2942 ofport->peer = peer;
2943 ofport->peer->peer = ofport;
2953 port_run_fast(struct ofport_dpif *ofport)
2955 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2956 struct ofpbuf packet;
2958 ofpbuf_init(&packet, 0);
2959 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2960 send_packet(ofport, &packet);
2961 ofpbuf_uninit(&packet);
2964 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2965 struct ofpbuf packet;
2967 ofpbuf_init(&packet, 0);
2968 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2969 send_packet(ofport, &packet);
2970 ofpbuf_uninit(&packet);
2975 port_run(struct ofport_dpif *ofport)
2977 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2978 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2979 bool enable = netdev_get_carrier(ofport->up.netdev);
2980 bool cfm_enable = false;
2981 bool bfd_enable = false;
2983 ofport->carrier_seq = carrier_seq;
2985 port_run_fast(ofport);
2988 int cfm_opup = cfm_get_opup(ofport->cfm);
2990 cfm_run(ofport->cfm);
2991 cfm_enable = !cfm_get_fault(ofport->cfm);
2993 if (cfm_opup >= 0) {
2994 cfm_enable = cfm_enable && cfm_opup;
2999 bfd_run(ofport->bfd);
3000 bfd_enable = bfd_forwarding(ofport->bfd);
3003 if (ofport->bfd || ofport->cfm) {
3004 enable = enable && (cfm_enable || bfd_enable);
3007 if (ofport->bundle) {
3008 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3009 if (carrier_changed) {
3010 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3014 if (ofport->may_enable != enable) {
3015 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3016 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3019 ofport->may_enable = enable;
3023 port_wait(struct ofport_dpif *ofport)
3026 cfm_wait(ofport->cfm);
3030 bfd_wait(ofport->bfd);
3035 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3036 struct ofproto_port *ofproto_port)
3038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3039 struct dpif_port dpif_port;
3042 if (sset_contains(&ofproto->ghost_ports, devname)) {
3043 const char *type = netdev_get_type_from_name(devname);
3045 /* We may be called before ofproto->up.port_by_name is populated with
3046 * the appropriate ofport. For this reason, we must get the name and
3047 * type from the netdev layer directly. */
3049 const struct ofport *ofport;
3051 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3052 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3053 ofproto_port->name = xstrdup(devname);
3054 ofproto_port->type = xstrdup(type);
3060 if (!sset_contains(&ofproto->ports, devname)) {
3063 error = dpif_port_query_by_name(ofproto->backer->dpif,
3064 devname, &dpif_port);
3066 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3072 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3075 const char *devname = netdev_get_name(netdev);
3076 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3077 const char *dp_port_name;
3079 if (netdev_vport_is_patch(netdev)) {
3080 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3084 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3085 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3086 odp_port_t port_no = ODPP_NONE;
3089 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3093 if (netdev_get_tunnel_config(netdev)) {
3094 simap_put(&ofproto->backer->tnl_backers,
3095 dp_port_name, odp_to_u32(port_no));
3099 if (netdev_get_tunnel_config(netdev)) {
3100 sset_add(&ofproto->ghost_ports, devname);
3102 sset_add(&ofproto->ports, devname);
3108 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3111 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3118 sset_find_and_delete(&ofproto->ghost_ports,
3119 netdev_get_name(ofport->up.netdev));
3120 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3121 if (!ofport->is_tunnel) {
3122 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3124 /* The caller is going to close ofport->up.netdev. If this is a
3125 * bonded port, then the bond is using that netdev, so remove it
3126 * from the bond. The client will need to reconfigure everything
3127 * after deleting ports, so then the slave will get re-added. */
3128 bundle_remove(&ofport->up);
3135 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3137 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3142 error = netdev_get_stats(ofport->up.netdev, stats);
3144 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3147 /* ofproto->stats.tx_packets represents packets that we created
3148 * internally and sent to some port (e.g. packets sent with
3149 * send_packet()). Account for them as if they had come from
3150 * OFPP_LOCAL and got forwarded. */
3152 if (stats->rx_packets != UINT64_MAX) {
3153 stats->rx_packets += ofproto->stats.tx_packets;
3156 if (stats->rx_bytes != UINT64_MAX) {
3157 stats->rx_bytes += ofproto->stats.tx_bytes;
3160 /* ofproto->stats.rx_packets represents packets that were received on
3161 * some port and we processed internally and dropped (e.g. STP).
3162 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3164 if (stats->tx_packets != UINT64_MAX) {
3165 stats->tx_packets += ofproto->stats.rx_packets;
3168 if (stats->tx_bytes != UINT64_MAX) {
3169 stats->tx_bytes += ofproto->stats.rx_bytes;
3176 struct port_dump_state {
3181 struct ofproto_port port;
3186 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3188 *statep = xzalloc(sizeof(struct port_dump_state));
3193 port_dump_next(const struct ofproto *ofproto_, void *state_,
3194 struct ofproto_port *port)
3196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3197 struct port_dump_state *state = state_;
3198 const struct sset *sset;
3199 struct sset_node *node;
3201 if (state->has_port) {
3202 ofproto_port_destroy(&state->port);
3203 state->has_port = false;
3205 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3206 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3209 error = port_query_by_name(ofproto_, node->name, &state->port);
3211 *port = state->port;
3212 state->has_port = true;
3214 } else if (error != ENODEV) {
3219 if (!state->ghost) {
3220 state->ghost = true;
3223 return port_dump_next(ofproto_, state_, port);
3230 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3232 struct port_dump_state *state = state_;
3234 if (state->has_port) {
3235 ofproto_port_destroy(&state->port);
3242 port_poll(const struct ofproto *ofproto_, char **devnamep)
3244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3246 if (ofproto->port_poll_errno) {
3247 int error = ofproto->port_poll_errno;
3248 ofproto->port_poll_errno = 0;
3252 if (sset_is_empty(&ofproto->port_poll_set)) {
3256 *devnamep = sset_pop(&ofproto->port_poll_set);
3261 port_poll_wait(const struct ofproto *ofproto_)
3263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3264 dpif_port_poll_wait(ofproto->backer->dpif);
3268 port_is_lacp_current(const struct ofport *ofport_)
3270 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3271 return (ofport->bundle && ofport->bundle->lacp
3272 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3276 /* Upcall handling. */
3278 struct flow_miss_op {
3279 struct dpif_op dpif_op;
3281 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3282 struct xlate_out xout;
3283 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3285 struct ofpbuf mask; /* Flow mask for "put" ops. */
3286 struct odputil_keybuf maskbuf;
3288 /* If this is a "put" op, then a pointer to the subfacet that should
3289 * be marked as uninstalled if the operation fails. */
3290 struct subfacet *subfacet;
3293 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3294 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3295 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3296 * return value of true). However, for short flows the cost of bookkeeping is
3297 * much higher than the benefits, so when the datapath holds a large number of
3298 * flows we impose some heuristics to decide which flows are likely to be worth
3301 flow_miss_should_make_facet(struct flow_miss *miss)
3303 struct dpif_backer *backer = miss->ofproto->backer;
3306 switch (flow_miss_model) {
3307 case OFPROTO_HANDLE_MISS_AUTO:
3309 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3311 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3315 if (!backer->governor) {
3318 n_subfacets = hmap_count(&backer->subfacets);
3319 if (n_subfacets * 2 <= flow_eviction_threshold) {
3323 backer->governor = governor_create();
3326 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3327 return governor_should_install_flow(backer->governor, hash,
3328 list_size(&miss->packets));
3331 /* Handles 'miss', which matches 'facet'. May add any required datapath
3332 * operations to 'ops', incrementing '*n_ops' for each new op.
3334 * All of the packets in 'miss' are considered to have arrived at time
3335 * 'miss->stats.used'. This is really important only for new facets: if we
3336 * just called time_msec() here, then the new subfacet or its packets could
3337 * look (occasionally) as though it was used some time after the facet was
3338 * used. That can make a one-packet flow look like it has a nonzero duration,
3339 * which looks odd in e.g. NetFlow statistics. */
3341 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3342 struct flow_miss_op *ops, size_t *n_ops)
3344 enum subfacet_path want_path;
3345 struct subfacet *subfacet;
3347 facet->packet_count += miss->stats.n_packets;
3348 facet->prev_packet_count += miss->stats.n_packets;
3349 facet->byte_count += miss->stats.n_bytes;
3350 facet->prev_byte_count += miss->stats.n_bytes;
3352 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3354 /* Don't install the flow if it's the result of the "userspace"
3355 * action for an already installed facet. This can occur when a
3356 * datapath flow with wildcards has a "userspace" action and flows
3357 * sent to userspace result in a different subfacet, which will then
3358 * be rejected as overlapping by the datapath. */
3359 if (miss->upcall_type == DPIF_UC_ACTION
3360 && !list_is_empty(&facet->subfacets)) {
3364 subfacet = subfacet_create(facet, miss);
3365 if (subfacet->path != want_path) {
3366 struct flow_miss_op *op = &ops[(*n_ops)++];
3367 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3369 subfacet->path = want_path;
3371 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3372 if (enable_megaflows) {
3373 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3374 &miss->flow, UINT32_MAX);
3377 op->xout_garbage = false;
3378 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3379 op->subfacet = subfacet;
3380 put->flags = DPIF_FP_CREATE;
3381 put->key = miss->key;
3382 put->key_len = miss->key_len;
3383 put->mask = op->mask.data;
3384 put->mask_len = op->mask.size;
3386 if (want_path == SF_FAST_PATH) {
3387 put->actions = facet->xout.odp_actions.data;
3388 put->actions_len = facet->xout.odp_actions.size;
3390 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3391 op->slow_stub, sizeof op->slow_stub,
3392 &put->actions, &put->actions_len);
3398 /* Handles flow miss 'miss'. May add any required datapath operations
3399 * to 'ops', incrementing '*n_ops' for each new op. */
3401 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3404 struct facet *facet;
3406 miss->ofproto->n_missed += list_size(&miss->packets);
3408 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3410 /* There does not exist a bijection between 'struct flow' and datapath
3411 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3412 * assumption used throughout the facet and subfacet handling code.
3413 * Since we have to handle these misses in userspace anyway, we simply
3414 * skip facet creation, avoiding the problem altogether. */
3415 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3416 || !flow_miss_should_make_facet(miss)) {
3420 facet = facet_create(miss);
3422 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3425 static struct drop_key *
3426 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3429 struct drop_key *drop_key;
3431 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3432 &backer->drop_keys) {
3433 if (drop_key->key_len == key_len
3434 && !memcmp(drop_key->key, key, key_len)) {
3442 drop_key_clear(struct dpif_backer *backer)
3444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3445 struct drop_key *drop_key, *next;
3447 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3450 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3452 if (error && !VLOG_DROP_WARN(&rl)) {
3453 struct ds ds = DS_EMPTY_INITIALIZER;
3454 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3455 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3456 ovs_strerror(error), ds_cstr(&ds));
3460 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3461 drop_key_destroy(drop_key);
3464 udpif_drop_key_clear(backer->udpif);
3468 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3470 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3471 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3472 struct flow_miss *miss;
3475 /* Process each element in the to-do list, constructing the set of
3476 * operations to batch. */
3478 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3479 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3481 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3483 /* Execute batch. */
3484 for (i = 0; i < n_ops; i++) {
3485 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3487 dpif_operate(backer->dpif, dpif_ops, n_ops);
3489 for (i = 0; i < n_ops; i++) {
3490 if (dpif_ops[i]->error != 0
3491 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3492 && flow_miss_ops[i].subfacet) {
3493 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3495 COVERAGE_INC(subfacet_install_fail);
3497 /* Zero-out subfacet counters when installation failed, but
3498 * datapath reported hits. This should not happen and
3499 * indicates a bug, since if the datapath flow exists, we
3500 * should not be attempting to create a new subfacet. A
3501 * buggy datapath could trigger this, so just zero out the
3502 * counters and log an error. */
3503 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3504 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3505 "datapath reported hits");
3506 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3509 subfacet->path = SF_NOT_INSTALLED;
3515 handle_sflow_upcall(struct dpif_backer *backer,
3516 const struct dpif_upcall *upcall)
3518 struct ofproto_dpif *ofproto;
3519 union user_action_cookie cookie;
3521 odp_port_t odp_in_port;
3523 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3524 &flow, NULL, &ofproto, &odp_in_port)
3525 || !ofproto->sflow) {
3529 memset(&cookie, 0, sizeof cookie);
3530 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3531 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3532 odp_in_port, &cookie);
3536 handle_flow_sample_upcall(struct dpif_backer *backer,
3537 const struct dpif_upcall *upcall)
3539 struct ofproto_dpif *ofproto;
3540 union user_action_cookie cookie;
3543 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3544 &flow, NULL, &ofproto, NULL)
3545 || !ofproto->ipfix) {
3549 memset(&cookie, 0, sizeof cookie);
3550 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3552 /* The flow reflects exactly the contents of the packet. Sample
3553 * the packet using it. */
3554 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3555 cookie.flow_sample.collector_set_id,
3556 cookie.flow_sample.probability,
3557 cookie.flow_sample.obs_domain_id,
3558 cookie.flow_sample.obs_point_id);
3562 handle_ipfix_upcall(struct dpif_backer *backer,
3563 const struct dpif_upcall *upcall)
3565 struct ofproto_dpif *ofproto;
3568 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3569 &flow, NULL, &ofproto, NULL)
3570 || !ofproto->ipfix) {
3574 /* The flow reflects exactly the contents of the packet. Sample
3575 * the packet using it. */
3576 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3580 handle_upcalls(struct dpif_backer *backer)
3582 struct flow_miss_batch *fmb;
3585 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3586 struct upcall *upcall = upcall_next(backer->udpif);
3592 switch (upcall->type) {
3594 handle_sflow_upcall(backer, &upcall->dpif_upcall);
3597 case FLOW_SAMPLE_UPCALL:
3598 handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3602 handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3612 upcall_destroy(upcall);
3615 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3616 struct drop_key *drop_key = drop_key_next(backer->udpif);
3621 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3622 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3623 hash_bytes(drop_key->key, drop_key->key_len, 0));
3624 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3625 drop_key->key, drop_key->key_len,
3626 NULL, 0, NULL, 0, NULL);
3628 drop_key_destroy(drop_key);
3632 fmb = flow_miss_batch_next(backer->udpif);
3634 handle_flow_misses(backer, fmb);
3635 flow_miss_batch_destroy(fmb);
3639 /* Flow expiration. */
3641 static int subfacet_max_idle(const struct dpif_backer *);
3642 static void update_stats(struct dpif_backer *);
3643 static void rule_expire(struct rule_dpif *);
3644 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3646 /* This function is called periodically by run(). Its job is to collect
3647 * updates for the flows that have been installed into the datapath, most
3648 * importantly when they last were used, and then use that information to
3649 * expire flows that have not been used recently.
3651 * Returns the number of milliseconds after which it should be called again. */
3653 expire(struct dpif_backer *backer)
3655 struct ofproto_dpif *ofproto;
3659 /* Periodically clear out the drop keys in an effort to keep them
3660 * relatively few. */
3661 drop_key_clear(backer);
3663 /* Update stats for each flow in the backer. */
3664 update_stats(backer);
3666 n_subfacets = hmap_count(&backer->subfacets);
3668 struct subfacet *subfacet;
3669 long long int total, now;
3673 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3674 total += now - subfacet->created;
3676 backer->avg_subfacet_life += total / n_subfacets;
3678 backer->avg_subfacet_life /= 2;
3680 backer->avg_n_subfacet += n_subfacets;
3681 backer->avg_n_subfacet /= 2;
3683 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3685 max_idle = subfacet_max_idle(backer);
3686 expire_subfacets(backer, max_idle);
3688 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3689 struct rule *rule, *next_rule;
3691 if (ofproto->backer != backer) {
3695 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3697 ovs_mutex_lock(&ofproto->up.expirable_mutex);
3698 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3699 &ofproto->up.expirable) {
3700 rule_expire(rule_dpif_cast(rule));
3702 ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3704 /* All outstanding data in existing flows has been accounted, so it's a
3705 * good time to do bond rebalancing. */
3706 if (ofproto->has_bonded_bundles) {
3707 struct ofbundle *bundle;
3709 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3711 bond_rebalance(bundle->bond);
3717 return MIN(max_idle, 1000);
3720 /* Updates flow table statistics given that the datapath just reported 'stats'
3721 * as 'subfacet''s statistics. */
3723 update_subfacet_stats(struct subfacet *subfacet,
3724 const struct dpif_flow_stats *stats)
3726 struct facet *facet = subfacet->facet;
3727 struct dpif_flow_stats diff;
3729 diff.tcp_flags = stats->tcp_flags;
3730 diff.used = stats->used;
3732 if (stats->n_packets >= subfacet->dp_packet_count) {
3733 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3735 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3739 if (stats->n_bytes >= subfacet->dp_byte_count) {
3740 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3742 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3746 facet->ofproto->n_hit += diff.n_packets;
3747 subfacet->dp_packet_count = stats->n_packets;
3748 subfacet->dp_byte_count = stats->n_bytes;
3749 subfacet_update_stats(subfacet, &diff);
3751 if (facet->accounted_bytes < facet->byte_count) {
3753 facet_account(facet);
3754 facet->accounted_bytes = facet->byte_count;
3758 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3759 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3761 delete_unexpected_flow(struct dpif_backer *backer,
3762 const struct nlattr *key, size_t key_len)
3764 if (!VLOG_DROP_WARN(&rl)) {
3768 odp_flow_key_format(key, key_len, &s);
3769 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3773 COVERAGE_INC(facet_unexpected);
3774 dpif_flow_del(backer->dpif, key, key_len, NULL);
3777 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3779 * This function also pushes statistics updates to rules which each facet
3780 * resubmits into. Generally these statistics will be accurate. However, if a
3781 * facet changes the rule it resubmits into at some time in between
3782 * update_stats() runs, it is possible that statistics accrued to the
3783 * old rule will be incorrectly attributed to the new rule. This could be
3784 * avoided by calling update_stats() whenever rules are created or
3785 * deleted. However, the performance impact of making so many calls to the
3786 * datapath do not justify the benefit of having perfectly accurate statistics.
3788 * In addition, this function maintains per ofproto flow hit counts. The patch
3789 * port is not treated specially. e.g. A packet ingress from br0 patched into
3790 * br1 will increase the hit count of br0 by 1, however, does not affect
3791 * the hit or miss counts of br1.
3794 update_stats(struct dpif_backer *backer)
3796 const struct dpif_flow_stats *stats;
3797 struct dpif_flow_dump dump;
3798 const struct nlattr *key, *mask;
3799 size_t key_len, mask_len;
3801 dpif_flow_dump_start(&dump, backer->dpif);
3802 while (dpif_flow_dump_next(&dump, &key, &key_len,
3803 &mask, &mask_len, NULL, NULL, &stats)) {
3804 struct subfacet *subfacet;
3807 key_hash = odp_flow_key_hash(key, key_len);
3808 subfacet = subfacet_find(backer, key, key_len, key_hash);
3809 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3811 update_subfacet_stats(subfacet, stats);
3815 /* Stats are updated per-packet. */
3818 case SF_NOT_INSTALLED:
3820 delete_unexpected_flow(backer, key, key_len);
3825 dpif_flow_dump_done(&dump);
3827 update_moving_averages(backer);
3830 /* Calculates and returns the number of milliseconds of idle time after which
3831 * subfacets should expire from the datapath. When a subfacet expires, we fold
3832 * its statistics into its facet, and when a facet's last subfacet expires, we
3833 * fold its statistic into its rule. */
3835 subfacet_max_idle(const struct dpif_backer *backer)
3838 * Idle time histogram.
3840 * Most of the time a switch has a relatively small number of subfacets.
3841 * When this is the case we might as well keep statistics for all of them
3842 * in userspace and to cache them in the kernel datapath for performance as
3845 * As the number of subfacets increases, the memory required to maintain
3846 * statistics about them in userspace and in the kernel becomes
3847 * significant. However, with a large number of subfacets it is likely
3848 * that only a few of them are "heavy hitters" that consume a large amount
3849 * of bandwidth. At this point, only heavy hitters are worth caching in
3850 * the kernel and maintaining in userspaces; other subfacets we can
3853 * The technique used to compute the idle time is to build a histogram with
3854 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3855 * that is installed in the kernel gets dropped in the appropriate bucket.
3856 * After the histogram has been built, we compute the cutoff so that only
3857 * the most-recently-used 1% of subfacets (but at least
3858 * flow_eviction_threshold flows) are kept cached. At least
3859 * the most-recently-used bucket of subfacets is kept, so actually an
3860 * arbitrary number of subfacets can be kept in any given expiration run
3861 * (though the next run will delete most of those unless they receive
3864 * This requires a second pass through the subfacets, in addition to the
3865 * pass made by update_stats(), because the former function never looks at
3866 * uninstallable subfacets.
3868 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3869 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3870 int buckets[N_BUCKETS] = { 0 };
3871 int total, subtotal, bucket;
3872 struct subfacet *subfacet;
3876 total = hmap_count(&backer->subfacets);
3877 if (total <= flow_eviction_threshold) {
3878 return N_BUCKETS * BUCKET_WIDTH;
3881 /* Build histogram. */
3883 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3884 long long int idle = now - subfacet->used;
3885 int bucket = (idle <= 0 ? 0
3886 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3887 : (unsigned int) idle / BUCKET_WIDTH);
3891 /* Find the first bucket whose flows should be expired. */
3892 subtotal = bucket = 0;
3894 subtotal += buckets[bucket++];
3895 } while (bucket < N_BUCKETS &&
3896 subtotal < MAX(flow_eviction_threshold, total / 100));
3898 if (VLOG_IS_DBG_ENABLED()) {
3902 ds_put_cstr(&s, "keep");
3903 for (i = 0; i < N_BUCKETS; i++) {
3905 ds_put_cstr(&s, ", drop");
3908 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3911 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3915 return bucket * BUCKET_WIDTH;
3919 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3921 /* Cutoff time for most flows. */
3922 long long int normal_cutoff = time_msec() - dp_max_idle;
3924 /* We really want to keep flows for special protocols around, so use a more
3925 * conservative cutoff. */
3926 long long int special_cutoff = time_msec() - 10000;
3928 struct subfacet *subfacet, *next_subfacet;
3929 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3933 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3934 &backer->subfacets) {
3935 long long int cutoff;
3937 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3941 if (subfacet->used < cutoff) {
3942 if (subfacet->path != SF_NOT_INSTALLED) {
3943 batch[n_batch++] = subfacet;
3944 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3945 subfacet_destroy_batch(backer, batch, n_batch);
3949 subfacet_destroy(subfacet);
3955 subfacet_destroy_batch(backer, batch, n_batch);
3959 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3960 * then delete it entirely. */
3962 rule_expire(struct rule_dpif *rule)
3964 uint16_t idle_timeout, hard_timeout;
3968 if (rule->up.pending) {
3969 /* We'll have to expire it later. */
3973 ovs_mutex_lock(&rule->up.timeout_mutex);
3974 hard_timeout = rule->up.hard_timeout;
3975 idle_timeout = rule->up.idle_timeout;
3976 ovs_mutex_unlock(&rule->up.timeout_mutex);
3978 /* Has 'rule' expired? */
3980 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3981 reason = OFPRR_HARD_TIMEOUT;
3982 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3983 reason = OFPRR_IDLE_TIMEOUT;
3988 COVERAGE_INC(ofproto_dpif_expired);
3989 ofproto_rule_expire(&rule->up, reason);
3994 /* Creates and returns a new facet based on 'miss'.
3996 * The caller must already have determined that no facet with an identical
3997 * 'miss->flow' exists in 'miss->ofproto'.
3999 * 'rule' and 'xout' must have been created based on 'miss'.
4001 * 'facet'' statistics are initialized based on 'stats'.
4003 * The facet will initially have no subfacets. The caller should create (at
4004 * least) one subfacet with subfacet_create(). */
4005 static struct facet *
4006 facet_create(const struct flow_miss *miss)
4008 struct ofproto_dpif *ofproto = miss->ofproto;
4009 struct facet *facet;
4012 facet = xzalloc(sizeof *facet);
4013 facet->ofproto = miss->ofproto;
4014 facet->used = miss->stats.used;
4015 facet->flow = miss->flow;
4016 facet->learn_rl = time_msec() + 500;
4018 list_init(&facet->subfacets);
4019 netflow_flow_init(&facet->nf_flow);
4020 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4022 xlate_out_copy(&facet->xout, &miss->xout);
4024 match_init(&match, &facet->flow, &facet->xout.wc);
4025 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4026 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4027 classifier_insert(&ofproto->facets, &facet->cr);
4028 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4030 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4035 facet_free(struct facet *facet)
4038 xlate_out_uninit(&facet->xout);
4043 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4044 * 'packet', which arrived on 'in_port'. */
4046 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4047 const struct nlattr *odp_actions, size_t actions_len,
4048 struct ofpbuf *packet)
4050 struct odputil_keybuf keybuf;
4054 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4055 odp_flow_key_from_flow(&key, flow,
4056 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4058 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4059 odp_actions, actions_len, packet);
4063 /* Remove 'facet' from its ofproto and free up the associated memory:
4065 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4066 * rule's statistics, via subfacet_uninstall().
4068 * - Removes 'facet' from its rule and from ofproto->facets.
4071 facet_remove(struct facet *facet)
4073 struct subfacet *subfacet, *next_subfacet;
4075 ovs_assert(!list_is_empty(&facet->subfacets));
4077 /* First uninstall all of the subfacets to get final statistics. */
4078 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4079 subfacet_uninstall(subfacet);
4082 /* Flush the final stats to the rule.
4084 * This might require us to have at least one subfacet around so that we
4085 * can use its actions for accounting in facet_account(), which is why we
4086 * have uninstalled but not yet destroyed the subfacets. */
4087 facet_flush_stats(facet);
4089 /* Now we're really all done so destroy everything. */
4090 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4091 &facet->subfacets) {
4092 subfacet_destroy__(subfacet);
4094 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4095 classifier_remove(&facet->ofproto->facets, &facet->cr);
4096 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4097 cls_rule_destroy(&facet->cr);
4101 /* Feed information from 'facet' back into the learning table to keep it in
4102 * sync with what is actually flowing through the datapath. */
4104 facet_learn(struct facet *facet)
4106 long long int now = time_msec();
4108 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4112 facet->learn_rl = now + 500;
4114 if (!facet->xout.has_learn
4115 && !facet->xout.has_normal
4116 && (!facet->xout.has_fin_timeout
4117 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4121 facet_push_stats(facet, true);
4125 facet_account(struct facet *facet)
4127 const struct nlattr *a;
4132 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4135 n_bytes = facet->byte_count - facet->accounted_bytes;
4137 /* This loop feeds byte counters to bond_account() for rebalancing to use
4138 * as a basis. We also need to track the actual VLAN on which the packet
4139 * is going to be sent to ensure that it matches the one passed to
4140 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4143 * We use the actions from an arbitrary subfacet because they should all
4144 * be equally valid for our purpose. */
4145 vlan_tci = facet->flow.vlan_tci;
4146 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4147 facet->xout.odp_actions.size) {
4148 const struct ovs_action_push_vlan *vlan;
4149 struct ofport_dpif *port;
4151 switch (nl_attr_type(a)) {
4152 case OVS_ACTION_ATTR_OUTPUT:
4153 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4154 if (port && port->bundle && port->bundle->bond) {
4155 bond_account(port->bundle->bond, &facet->flow,
4156 vlan_tci_to_vid(vlan_tci), n_bytes);
4160 case OVS_ACTION_ATTR_POP_VLAN:
4161 vlan_tci = htons(0);
4164 case OVS_ACTION_ATTR_PUSH_VLAN:
4165 vlan = nl_attr_get(a);
4166 vlan_tci = vlan->vlan_tci;
4172 /* Returns true if the only action for 'facet' is to send to the controller.
4173 * (We don't report NetFlow expiration messages for such facets because they
4174 * are just part of the control logic for the network, not real traffic). */
4176 facet_is_controller_flow(struct facet *facet)
4179 struct ofproto_dpif *ofproto = facet->ofproto;
4180 const struct ofpact *ofpacts;
4181 struct rule_dpif *rule;
4185 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4186 ofpacts_len = rule->up.ofpacts_len;
4187 ofpacts = rule->up.ofpacts;
4188 is_controller = ofpacts_len > 0
4189 && ofpacts->type == OFPACT_CONTROLLER
4190 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4192 return is_controller;
4197 /* Folds all of 'facet''s statistics into its rule. Also updates the
4198 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4199 * 'facet''s statistics in the datapath should have been zeroed and folded into
4200 * its packet and byte counts before this function is called. */
4202 facet_flush_stats(struct facet *facet)
4204 struct ofproto_dpif *ofproto = facet->ofproto;
4205 struct subfacet *subfacet;
4207 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4208 ovs_assert(!subfacet->dp_byte_count);
4209 ovs_assert(!subfacet->dp_packet_count);
4212 facet_push_stats(facet, false);
4213 if (facet->accounted_bytes < facet->byte_count) {
4214 facet_account(facet);
4215 facet->accounted_bytes = facet->byte_count;
4218 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4219 struct ofexpired expired;
4220 expired.flow = facet->flow;
4221 expired.packet_count = facet->packet_count;
4222 expired.byte_count = facet->byte_count;
4223 expired.used = facet->used;
4224 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4227 /* Reset counters to prevent double counting if 'facet' ever gets
4229 facet_reset_counters(facet);
4231 netflow_flow_clear(&facet->nf_flow);
4232 facet->tcp_flags = 0;
4235 /* Searches 'ofproto''s table of facets for one which would be responsible for
4236 * 'flow'. Returns it if found, otherwise a null pointer.
4238 * The returned facet might need revalidation; use facet_lookup_valid()
4239 * instead if that is important. */
4240 static struct facet *
4241 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4243 struct cls_rule *cr;
4245 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4246 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4247 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4248 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4251 /* Searches 'ofproto''s table of facets for one capable that covers
4252 * 'flow'. Returns it if found, otherwise a null pointer.
4254 * The returned facet is guaranteed to be valid. */
4255 static struct facet *
4256 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4258 struct facet *facet;
4260 facet = facet_find(ofproto, flow);
4262 && ofproto->backer->need_revalidate
4263 && !facet_revalidate(facet)) {
4271 facet_check_consistency(struct facet *facet)
4273 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4275 struct xlate_out xout;
4276 struct xlate_in xin;
4278 struct rule_dpif *rule;
4281 /* Check the datapath actions for consistency. */
4282 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4283 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4284 xlate_actions(&xin, &xout);
4287 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4288 && facet->xout.slow == xout.slow;
4289 if (!ok && !VLOG_DROP_WARN(&rl)) {
4290 struct ds s = DS_EMPTY_INITIALIZER;
4292 flow_format(&s, &facet->flow);
4293 ds_put_cstr(&s, ": inconsistency in facet");
4295 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4296 ds_put_cstr(&s, " (actions were: ");
4297 format_odp_actions(&s, facet->xout.odp_actions.data,
4298 facet->xout.odp_actions.size);
4299 ds_put_cstr(&s, ") (correct actions: ");
4300 format_odp_actions(&s, xout.odp_actions.data,
4301 xout.odp_actions.size);
4302 ds_put_char(&s, ')');
4305 if (facet->xout.slow != xout.slow) {
4306 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4311 xlate_out_uninit(&xout);
4316 /* Re-searches the classifier for 'facet':
4318 * - If the rule found is different from 'facet''s current rule, moves
4319 * 'facet' to the new rule and recompiles its actions.
4321 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4322 * where it is and recompiles its actions anyway.
4324 * - If any of 'facet''s subfacets correspond to a new flow according to
4325 * xlate_receive(), 'facet' is removed.
4327 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4329 facet_revalidate(struct facet *facet)
4331 struct ofproto_dpif *ofproto = facet->ofproto;
4332 struct rule_dpif *new_rule;
4333 struct subfacet *subfacet;
4334 struct flow_wildcards wc;
4335 struct xlate_out xout;
4336 struct xlate_in xin;
4338 COVERAGE_INC(facet_revalidate);
4340 /* Check that child subfacets still correspond to this facet. Tunnel
4341 * configuration changes could cause a subfacet's OpenFlow in_port to
4343 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4344 struct ofproto_dpif *recv_ofproto;
4345 struct flow recv_flow;
4348 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4349 subfacet->key_len, &recv_flow, NULL,
4350 &recv_ofproto, NULL);
4352 || recv_ofproto != ofproto
4353 || facet != facet_find(ofproto, &recv_flow)) {
4354 facet_remove(facet);
4359 flow_wildcards_init_catchall(&wc);
4360 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4362 /* Calculate new datapath actions.
4364 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4365 * emit a NetFlow expiration and, if so, we need to have the old state
4366 * around to properly compose it. */
4367 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4368 xlate_actions(&xin, &xout);
4369 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4371 /* A facet's slow path reason should only change under dramatic
4372 * circumstances. Rather than try to update everything, it's simpler to
4373 * remove the facet and start over.
4375 * More importantly, if a facet's wildcards change, it will be relatively
4376 * difficult to figure out if its subfacets still belong to it, and if not
4377 * which facet they may belong to. Again, to avoid the complexity, we
4378 * simply give up instead. */
4379 if (facet->xout.slow != xout.slow
4380 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4381 facet_remove(facet);
4382 xlate_out_uninit(&xout);
4383 rule_release(new_rule);
4387 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4388 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4389 if (subfacet->path == SF_FAST_PATH) {
4390 struct dpif_flow_stats stats;
4392 subfacet_install(subfacet, &xout.odp_actions, &stats);
4393 subfacet_update_stats(subfacet, &stats);
4397 facet_flush_stats(facet);
4399 ofpbuf_clear(&facet->xout.odp_actions);
4400 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4401 xout.odp_actions.size);
4404 /* Update 'facet' now that we've taken care of all the old state. */
4405 facet->xout.slow = xout.slow;
4406 facet->xout.has_learn = xout.has_learn;
4407 facet->xout.has_normal = xout.has_normal;
4408 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4409 facet->xout.nf_output_iface = xout.nf_output_iface;
4410 facet->xout.mirrors = xout.mirrors;
4411 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4412 facet->used = MAX(facet->used, new_rule->up.created);
4414 xlate_out_uninit(&xout);
4415 rule_release(new_rule);
4420 facet_reset_counters(struct facet *facet)
4422 facet->packet_count = 0;
4423 facet->byte_count = 0;
4424 facet->prev_packet_count = 0;
4425 facet->prev_byte_count = 0;
4426 facet->accounted_bytes = 0;
4430 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4431 struct dpif_flow_stats *stats, bool may_learn)
4433 struct ofport_dpif *in_port;
4434 struct rule_dpif *rule;
4435 struct xlate_in xin;
4437 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4438 if (in_port && in_port->is_tunnel) {
4439 netdev_vport_inc_rx(in_port->up.netdev, stats);
4442 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4443 rule_credit_stats(rule, stats);
4444 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4445 xin.resubmit_stats = stats;
4446 xin.may_learn = may_learn;
4447 xlate_actions_for_side_effects(&xin);
4452 facet_push_stats(struct facet *facet, bool may_learn)
4454 struct dpif_flow_stats stats;
4456 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4457 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4458 ovs_assert(facet->used >= facet->prev_used);
4460 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4461 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4462 stats.used = facet->used;
4463 stats.tcp_flags = facet->tcp_flags;
4465 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4466 facet->prev_packet_count = facet->packet_count;
4467 facet->prev_byte_count = facet->byte_count;
4468 facet->prev_used = facet->used;
4470 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4472 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4473 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4474 stats.n_packets, stats.n_bytes);
4475 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4480 push_all_stats__(bool run_fast)
4482 static long long int rl = LLONG_MIN;
4483 struct ofproto_dpif *ofproto;
4485 if (time_msec() < rl) {
4489 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4490 struct cls_cursor cursor;
4491 struct facet *facet;
4493 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4494 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4495 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4496 facet_push_stats(facet, false);
4501 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4504 rl = time_msec() + 100;
4508 push_all_stats(void)
4510 push_all_stats__(true);
4514 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4516 ovs_mutex_lock(&rule->stats_mutex);
4517 rule->packet_count += stats->n_packets;
4518 rule->byte_count += stats->n_bytes;
4519 ofproto_rule_update_used(&rule->up, stats->used);
4520 ovs_mutex_unlock(&rule->stats_mutex);
4525 static struct subfacet *
4526 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4527 size_t key_len, uint32_t key_hash)
4529 struct subfacet *subfacet;
4531 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4532 &backer->subfacets) {
4533 if (subfacet->key_len == key_len
4534 && !memcmp(key, subfacet->key, key_len)) {
4542 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4543 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4544 * existing subfacet if there is one, otherwise creates and returns a
4546 static struct subfacet *
4547 subfacet_create(struct facet *facet, struct flow_miss *miss)
4549 struct dpif_backer *backer = miss->ofproto->backer;
4550 enum odp_key_fitness key_fitness = miss->key_fitness;
4551 const struct nlattr *key = miss->key;
4552 size_t key_len = miss->key_len;
4554 struct subfacet *subfacet;
4556 key_hash = odp_flow_key_hash(key, key_len);
4558 if (list_is_empty(&facet->subfacets)) {
4559 subfacet = &facet->one_subfacet;
4561 subfacet = subfacet_find(backer, key, key_len, key_hash);
4563 if (subfacet->facet == facet) {
4567 /* This shouldn't happen. */
4568 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4569 subfacet_destroy(subfacet);
4572 subfacet = xmalloc(sizeof *subfacet);
4575 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4576 list_push_back(&facet->subfacets, &subfacet->list_node);
4577 subfacet->facet = facet;
4578 subfacet->key_fitness = key_fitness;
4579 subfacet->key = xmemdup(key, key_len);
4580 subfacet->key_len = key_len;
4581 subfacet->used = miss->stats.used;
4582 subfacet->created = subfacet->used;
4583 subfacet->dp_packet_count = 0;
4584 subfacet->dp_byte_count = 0;
4585 subfacet->path = SF_NOT_INSTALLED;
4586 subfacet->backer = backer;
4588 backer->subfacet_add_count++;
4592 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4593 * its facet within 'ofproto', and frees it. */
4595 subfacet_destroy__(struct subfacet *subfacet)
4597 struct facet *facet = subfacet->facet;
4598 struct ofproto_dpif *ofproto = facet->ofproto;
4600 /* Update ofproto stats before uninstall the subfacet. */
4601 ofproto->backer->subfacet_del_count++;
4603 subfacet_uninstall(subfacet);
4604 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4605 list_remove(&subfacet->list_node);
4606 free(subfacet->key);
4607 if (subfacet != &facet->one_subfacet) {
4612 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4613 * last remaining subfacet in its facet destroys the facet too. */
4615 subfacet_destroy(struct subfacet *subfacet)
4617 struct facet *facet = subfacet->facet;
4619 if (list_is_singleton(&facet->subfacets)) {
4620 /* facet_remove() needs at least one subfacet (it will remove it). */
4621 facet_remove(facet);
4623 subfacet_destroy__(subfacet);
4628 subfacet_destroy_batch(struct dpif_backer *backer,
4629 struct subfacet **subfacets, int n)
4631 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4632 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4633 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4636 for (i = 0; i < n; i++) {
4637 ops[i].type = DPIF_OP_FLOW_DEL;
4638 ops[i].u.flow_del.key = subfacets[i]->key;
4639 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4640 ops[i].u.flow_del.stats = &stats[i];
4644 dpif_operate(backer->dpif, opsp, n);
4645 for (i = 0; i < n; i++) {
4646 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4647 subfacets[i]->path = SF_NOT_INSTALLED;
4648 subfacet_destroy(subfacets[i]);
4653 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4654 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4655 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4656 * since 'subfacet' was last updated.
4658 * Returns 0 if successful, otherwise a positive errno value. */
4660 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4661 struct dpif_flow_stats *stats)
4663 struct facet *facet = subfacet->facet;
4664 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4665 const struct nlattr *actions = odp_actions->data;
4666 size_t actions_len = odp_actions->size;
4667 struct odputil_keybuf maskbuf;
4670 uint64_t slow_path_stub[128 / 8];
4671 enum dpif_flow_put_flags flags;
4674 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4677 flags |= DPIF_FP_ZERO_STATS;
4680 if (path == SF_SLOW_PATH) {
4681 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4682 slow_path_stub, sizeof slow_path_stub,
4683 &actions, &actions_len);
4686 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4687 if (enable_megaflows) {
4688 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4689 &facet->flow, UINT32_MAX);
4692 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4693 subfacet->key_len, mask.data, mask.size,
4694 actions, actions_len, stats);
4697 subfacet_reset_dp_stats(subfacet, stats);
4701 COVERAGE_INC(subfacet_install_fail);
4703 subfacet->path = path;
4708 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4710 subfacet_uninstall(struct subfacet *subfacet)
4712 if (subfacet->path != SF_NOT_INSTALLED) {
4713 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4714 struct dpif_flow_stats stats;
4717 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4718 subfacet->key_len, &stats);
4719 subfacet_reset_dp_stats(subfacet, &stats);
4721 subfacet_update_stats(subfacet, &stats);
4723 subfacet->path = SF_NOT_INSTALLED;
4725 ovs_assert(subfacet->dp_packet_count == 0);
4726 ovs_assert(subfacet->dp_byte_count == 0);
4730 /* Resets 'subfacet''s datapath statistics counters. This should be called
4731 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4732 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4733 * was reset in the datapath. 'stats' will be modified to include only
4734 * statistics new since 'subfacet' was last updated. */
4736 subfacet_reset_dp_stats(struct subfacet *subfacet,
4737 struct dpif_flow_stats *stats)
4740 && subfacet->dp_packet_count <= stats->n_packets
4741 && subfacet->dp_byte_count <= stats->n_bytes) {
4742 stats->n_packets -= subfacet->dp_packet_count;
4743 stats->n_bytes -= subfacet->dp_byte_count;
4746 subfacet->dp_packet_count = 0;
4747 subfacet->dp_byte_count = 0;
4750 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4752 * Because of the meaning of a subfacet's counters, it only makes sense to do
4753 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4754 * represents a packet that was sent by hand or if it represents statistics
4755 * that have been cleared out of the datapath. */
4757 subfacet_update_stats(struct subfacet *subfacet,
4758 const struct dpif_flow_stats *stats)
4760 if (stats->n_packets || stats->used > subfacet->used) {
4761 struct facet *facet = subfacet->facet;
4763 subfacet->used = MAX(subfacet->used, stats->used);
4764 facet->used = MAX(facet->used, stats->used);
4765 facet->packet_count += stats->n_packets;
4766 facet->byte_count += stats->n_bytes;
4767 facet->tcp_flags |= stats->tcp_flags;
4773 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4774 * the fields that were relevant as part of the lookup. */
4776 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4777 struct flow_wildcards *wc, struct rule_dpif **rule)
4779 struct ofport_dpif *port;
4781 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4784 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4786 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4787 flow->in_port.ofp_port);
4790 *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4791 ofproto->no_packet_in_rule);
4792 ovs_rwlock_rdlock(&(*rule)->up.evict);
4796 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4797 const struct flow *flow, struct flow_wildcards *wc,
4798 uint8_t table_id, struct rule_dpif **rule)
4799 OVS_TRY_RDLOCK(true, (*rule)->up.evict)
4801 struct cls_rule *cls_rule;
4802 struct classifier *cls;
4806 if (table_id >= N_TABLES) {
4811 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4812 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4815 cls = &ofproto->up.tables[table_id].cls;
4816 ovs_rwlock_rdlock(&cls->rwlock);
4817 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4818 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4819 /* We must pretend that transport ports are unavailable. */
4820 struct flow ofpc_normal_flow = *flow;
4821 ofpc_normal_flow.tp_src = htons(0);
4822 ofpc_normal_flow.tp_dst = htons(0);
4823 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4824 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4825 cls_rule = &ofproto->drop_frags_rule->up.cr;
4827 flow_wildcards_init_exact(wc);
4830 cls_rule = classifier_lookup(cls, flow, wc);
4833 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4834 if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4835 /* The rule is in the process of being removed. Best we can do is
4836 * pretend it isn't there. */
4839 ovs_rwlock_unlock(&cls->rwlock);
4841 return *rule != NULL;
4844 /* Given a port configuration (specified as zero if there's no port), chooses
4845 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4846 * flow table miss. */
4848 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4849 struct rule_dpif *no_packet_in_rule)
4851 return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4855 rule_release(struct rule_dpif *rule)
4856 OVS_NO_THREAD_SAFETY_ANALYSIS
4859 ovs_rwlock_unlock(&rule->up.evict);
4864 complete_operation(struct rule_dpif *rule)
4866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4868 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4870 struct dpif_completion *c = xmalloc(sizeof *c);
4871 c->op = rule->up.pending;
4872 list_push_back(&ofproto->completions, &c->list_node);
4874 ofoperation_complete(rule->up.pending, 0);
4878 static struct rule *
4881 struct rule_dpif *rule = xmalloc(sizeof *rule);
4886 rule_dealloc(struct rule *rule_)
4888 struct rule_dpif *rule = rule_dpif_cast(rule_);
4893 rule_construct(struct rule *rule_)
4895 struct rule_dpif *rule = rule_dpif_cast(rule_);
4896 ovs_mutex_init(&rule->stats_mutex);
4897 ovs_mutex_lock(&rule->stats_mutex);
4898 rule->packet_count = 0;
4899 rule->byte_count = 0;
4900 ovs_mutex_unlock(&rule->stats_mutex);
4905 rule_insert(struct rule *rule_)
4907 struct rule_dpif *rule = rule_dpif_cast(rule_);
4908 complete_operation(rule);
4912 rule_delete(struct rule *rule_)
4914 struct rule_dpif *rule = rule_dpif_cast(rule_);
4915 complete_operation(rule);
4919 rule_destruct(struct rule *rule_)
4921 struct rule_dpif *rule = rule_dpif_cast(rule_);
4922 ovs_mutex_destroy(&rule->stats_mutex);
4926 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4928 struct rule_dpif *rule = rule_dpif_cast(rule_);
4930 /* push_all_stats() can handle flow misses which, when using the learn
4931 * action, can cause rules to be added and deleted. This can corrupt our
4932 * caller's datastructures which assume that rule_get_stats() doesn't have
4933 * an impact on the flow table. To be safe, we disable miss handling. */
4934 push_all_stats__(false);
4936 /* Start from historical data for 'rule' itself that are no longer tracked
4937 * in facets. This counts, for example, facets that have expired. */
4938 ovs_mutex_lock(&rule->stats_mutex);
4939 *packets = rule->packet_count;
4940 *bytes = rule->byte_count;
4941 ovs_mutex_unlock(&rule->stats_mutex);
4945 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4946 struct ofpbuf *packet)
4948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4949 struct dpif_flow_stats stats;
4950 struct xlate_out xout;
4951 struct xlate_in xin;
4953 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4954 rule_credit_stats(rule, &stats);
4956 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4957 xin.resubmit_stats = &stats;
4958 xlate_actions(&xin, &xout);
4960 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4961 xout.odp_actions.size, packet);
4963 xlate_out_uninit(&xout);
4967 rule_execute(struct rule *rule, const struct flow *flow,
4968 struct ofpbuf *packet)
4970 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4971 ofpbuf_delete(packet);
4976 rule_modify_actions(struct rule *rule_, bool reset_counters)
4978 struct rule_dpif *rule = rule_dpif_cast(rule_);
4980 if (reset_counters) {
4981 ovs_mutex_lock(&rule->stats_mutex);
4982 rule->packet_count = 0;
4983 rule->byte_count = 0;
4984 ovs_mutex_unlock(&rule->stats_mutex);
4987 complete_operation(rule);
4990 /* Sends 'packet' out 'ofport'.
4991 * May modify 'packet'.
4992 * Returns 0 if successful, otherwise a positive errno value. */
4994 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4997 uint64_t odp_actions_stub[1024 / 8];
4998 struct ofpbuf key, odp_actions;
4999 struct dpif_flow_stats stats;
5000 struct odputil_keybuf keybuf;
5001 struct ofpact_output output;
5002 struct xlate_out xout;
5003 struct xlate_in xin;
5005 union flow_in_port in_port_;
5008 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5009 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5011 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5012 in_port_.ofp_port = OFPP_NONE;
5013 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5014 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5016 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5018 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5019 output.port = ofport->up.ofp_port;
5022 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5023 xin.ofpacts_len = sizeof output;
5024 xin.ofpacts = &output.ofpact;
5025 xin.resubmit_stats = &stats;
5026 xlate_actions(&xin, &xout);
5028 error = dpif_execute(ofproto->backer->dpif,
5030 xout.odp_actions.data, xout.odp_actions.size,
5032 xlate_out_uninit(&xout);
5035 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5036 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5037 ovs_strerror(error));
5040 ofproto->stats.tx_packets++;
5041 ofproto->stats.tx_bytes += packet->size;
5045 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5046 * The action will state 'slow' as the reason that the action is in the slow
5047 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5048 * dump-flows" output to see why a flow is in the slow path.)
5050 * The 'stub_size' bytes in 'stub' will be used to store the action.
5051 * 'stub_size' must be large enough for the action.
5053 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5056 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5057 enum slow_path_reason slow,
5058 uint64_t *stub, size_t stub_size,
5059 const struct nlattr **actionsp, size_t *actions_lenp)
5061 union user_action_cookie cookie;
5064 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5065 cookie.slow_path.unused = 0;
5066 cookie.slow_path.reason = slow;
5068 ofpbuf_use_stack(&buf, stub, stub_size);
5069 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5070 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5072 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5074 odp_port_t odp_port;
5077 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5078 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5079 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5081 *actionsp = buf.data;
5082 *actions_lenp = buf.size;
5086 set_frag_handling(struct ofproto *ofproto_,
5087 enum ofp_config_flags frag_handling)
5089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5090 if (frag_handling != OFPC_FRAG_REASM) {
5091 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5099 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5100 const struct flow *flow,
5101 const struct ofpact *ofpacts, size_t ofpacts_len)
5103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5104 struct odputil_keybuf keybuf;
5105 struct dpif_flow_stats stats;
5106 struct xlate_out xout;
5107 struct xlate_in xin;
5111 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5112 odp_flow_key_from_flow(&key, flow,
5113 ofp_port_to_odp_port(ofproto,
5114 flow->in_port.ofp_port));
5116 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5118 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5119 xin.resubmit_stats = &stats;
5120 xin.ofpacts_len = ofpacts_len;
5121 xin.ofpacts = ofpacts;
5123 xlate_actions(&xin, &xout);
5124 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5125 xout.odp_actions.data, xout.odp_actions.size, packet);
5126 xlate_out_uninit(&xout);
5134 set_netflow(struct ofproto *ofproto_,
5135 const struct netflow_options *netflow_options)
5137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5139 if (netflow_options) {
5140 if (!ofproto->netflow) {
5141 ofproto->netflow = netflow_create();
5142 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5144 return netflow_set_options(ofproto->netflow, netflow_options);
5145 } else if (ofproto->netflow) {
5146 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5147 netflow_destroy(ofproto->netflow);
5148 ofproto->netflow = NULL;
5155 get_netflow_ids(const struct ofproto *ofproto_,
5156 uint8_t *engine_type, uint8_t *engine_id)
5158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5160 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5164 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5166 if (!facet_is_controller_flow(facet) &&
5167 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5168 struct subfacet *subfacet;
5169 struct ofexpired expired;
5171 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5172 if (subfacet->path == SF_FAST_PATH) {
5173 struct dpif_flow_stats stats;
5175 subfacet_install(subfacet, &facet->xout.odp_actions,
5177 subfacet_update_stats(subfacet, &stats);
5181 expired.flow = facet->flow;
5182 expired.packet_count = facet->packet_count;
5183 expired.byte_count = facet->byte_count;
5184 expired.used = facet->used;
5185 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5190 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5192 struct cls_cursor cursor;
5193 struct facet *facet;
5195 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5196 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5197 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5198 send_active_timeout(ofproto, facet);
5200 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5203 static struct ofproto_dpif *
5204 ofproto_dpif_lookup(const char *name)
5206 struct ofproto_dpif *ofproto;
5208 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5209 hash_string(name, 0), &all_ofproto_dpifs) {
5210 if (!strcmp(ofproto->up.name, name)) {
5218 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5219 const char *argv[], void *aux OVS_UNUSED)
5221 struct ofproto_dpif *ofproto;
5224 ofproto = ofproto_dpif_lookup(argv[1]);
5226 unixctl_command_reply_error(conn, "no such bridge");
5229 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5230 mac_learning_flush(ofproto->ml);
5231 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5233 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5234 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5235 mac_learning_flush(ofproto->ml);
5236 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5240 unixctl_command_reply(conn, "table successfully flushed");
5243 static struct ofport_dpif *
5244 ofbundle_get_a_port(const struct ofbundle *bundle)
5246 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5251 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5252 const char *argv[], void *aux OVS_UNUSED)
5254 struct ds ds = DS_EMPTY_INITIALIZER;
5255 const struct ofproto_dpif *ofproto;
5256 const struct mac_entry *e;
5258 ofproto = ofproto_dpif_lookup(argv[1]);
5260 unixctl_command_reply_error(conn, "no such bridge");
5264 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5265 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5266 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5267 struct ofbundle *bundle = e->port.p;
5268 char name[OFP_MAX_PORT_NAME_LEN];
5270 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5272 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5273 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5274 mac_entry_age(ofproto->ml, e));
5276 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5277 unixctl_command_reply(conn, ds_cstr(&ds));
5282 struct xlate_out xout;
5283 struct xlate_in xin;
5289 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5291 ds_put_char_multiple(result, '\t', level);
5293 ds_put_cstr(result, "No match\n");
5297 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5298 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5299 cls_rule_format(&rule->up.cr, result);
5300 ds_put_char(result, '\n');
5302 ds_put_char_multiple(result, '\t', level);
5303 ds_put_cstr(result, "OpenFlow ");
5304 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5305 ds_put_char(result, '\n');
5309 trace_format_flow(struct ds *result, int level, const char *title,
5310 struct trace_ctx *trace)
5312 ds_put_char_multiple(result, '\t', level);
5313 ds_put_format(result, "%s: ", title);
5314 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5315 ds_put_cstr(result, "unchanged");
5317 flow_format(result, &trace->xin.flow);
5318 trace->flow = trace->xin.flow;
5320 ds_put_char(result, '\n');
5324 trace_format_regs(struct ds *result, int level, const char *title,
5325 struct trace_ctx *trace)
5329 ds_put_char_multiple(result, '\t', level);
5330 ds_put_format(result, "%s:", title);
5331 for (i = 0; i < FLOW_N_REGS; i++) {
5332 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5334 ds_put_char(result, '\n');
5338 trace_format_odp(struct ds *result, int level, const char *title,
5339 struct trace_ctx *trace)
5341 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5343 ds_put_char_multiple(result, '\t', level);
5344 ds_put_format(result, "%s: ", title);
5345 format_odp_actions(result, odp_actions->data, odp_actions->size);
5346 ds_put_char(result, '\n');
5350 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5352 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5353 struct ds *result = trace->result;
5355 ds_put_char(result, '\n');
5356 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5357 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5358 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5359 trace_format_rule(result, recurse + 1, rule);
5363 trace_report(struct xlate_in *xin, const char *s, int recurse)
5365 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5366 struct ds *result = trace->result;
5368 ds_put_char_multiple(result, '\t', recurse);
5369 ds_put_cstr(result, s);
5370 ds_put_char(result, '\n');
5374 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5375 void *aux OVS_UNUSED)
5377 const struct dpif_backer *backer;
5378 struct ofproto_dpif *ofproto;
5379 struct ofpbuf odp_key, odp_mask;
5380 struct ofpbuf *packet;
5388 ofpbuf_init(&odp_key, 0);
5389 ofpbuf_init(&odp_mask, 0);
5391 /* Handle "-generate" or a hex string as the last argument. */
5392 if (!strcmp(argv[argc - 1], "-generate")) {
5393 packet = ofpbuf_new(0);
5396 const char *error = eth_from_hex(argv[argc - 1], &packet);
5399 } else if (argc == 4) {
5400 /* The 3-argument form must end in "-generate' or a hex string. */
5401 unixctl_command_reply_error(conn, error);
5406 /* Parse the flow and determine whether a datapath or
5407 * bridge is specified. If function odp_flow_key_from_string()
5408 * returns 0, the flow is a odp_flow. If function
5409 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5410 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5411 /* If the odp_flow is the second argument,
5412 * the datapath name is the first argument. */
5414 const char *dp_type;
5415 if (!strncmp(argv[1], "ovs-", 4)) {
5416 dp_type = argv[1] + 4;
5420 backer = shash_find_data(&all_dpif_backers, dp_type);
5422 unixctl_command_reply_error(conn, "Cannot find datapath "
5427 /* No datapath name specified, so there should be only one
5429 struct shash_node *node;
5430 if (shash_count(&all_dpif_backers) != 1) {
5431 unixctl_command_reply_error(conn, "Must specify datapath "
5432 "name, there is more than one type of datapath");
5435 node = shash_first(&all_dpif_backers);
5436 backer = node->data;
5439 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5440 NULL, &ofproto, NULL)) {
5441 unixctl_command_reply_error(conn, "Invalid datapath flow");
5444 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5445 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5447 unixctl_command_reply_error(conn, "Must specify bridge name");
5451 ofproto = ofproto_dpif_lookup(argv[1]);
5453 unixctl_command_reply_error(conn, "Unknown bridge name");
5457 unixctl_command_reply_error(conn, "Bad flow syntax");
5461 /* Generate a packet, if requested. */
5463 if (!packet->size) {
5464 flow_compose(packet, &flow);
5466 union flow_in_port in_port_;
5468 in_port_ = flow.in_port;
5469 ds_put_cstr(&result, "Packet: ");
5470 s = ofp_packet_to_string(packet->data, packet->size);
5471 ds_put_cstr(&result, s);
5474 /* Use the metadata from the flow and the packet argument
5475 * to reconstruct the flow. */
5476 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5481 ofproto_trace(ofproto, &flow, packet, &result);
5482 unixctl_command_reply(conn, ds_cstr(&result));
5485 ds_destroy(&result);
5486 ofpbuf_delete(packet);
5487 ofpbuf_uninit(&odp_key);
5488 ofpbuf_uninit(&odp_mask);
5492 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5493 const struct ofpbuf *packet, struct ds *ds)
5495 struct rule_dpif *rule;
5496 struct flow_wildcards wc;
5498 ds_put_cstr(ds, "Flow: ");
5499 flow_format(ds, flow);
5500 ds_put_char(ds, '\n');
5502 flow_wildcards_init_catchall(&wc);
5503 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5505 trace_format_rule(ds, 0, rule);
5506 if (rule == ofproto->miss_rule) {
5507 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5508 } else if (rule == ofproto->no_packet_in_rule) {
5509 ds_put_cstr(ds, "\nNo match, packets dropped because "
5510 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5511 } else if (rule == ofproto->drop_frags_rule) {
5512 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5513 "and the fragment handling mode is \"drop\".\n");
5517 uint64_t odp_actions_stub[1024 / 8];
5518 struct ofpbuf odp_actions;
5519 struct trace_ctx trace;
5523 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5526 ofpbuf_use_stub(&odp_actions,
5527 odp_actions_stub, sizeof odp_actions_stub);
5528 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5529 trace.xin.resubmit_hook = trace_resubmit;
5530 trace.xin.report_hook = trace_report;
5532 xlate_actions(&trace.xin, &trace.xout);
5533 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5535 ds_put_char(ds, '\n');
5536 trace_format_flow(ds, 0, "Final flow", &trace);
5538 match_init(&match, flow, &trace.xout.wc);
5539 ds_put_cstr(ds, "Relevant fields: ");
5540 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5541 ds_put_char(ds, '\n');
5543 ds_put_cstr(ds, "Datapath actions: ");
5544 format_odp_actions(ds, trace.xout.odp_actions.data,
5545 trace.xout.odp_actions.size);
5547 if (trace.xout.slow) {
5548 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5549 "slow path because it:");
5550 switch (trace.xout.slow) {
5552 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5555 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5558 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5561 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5563 case SLOW_CONTROLLER:
5564 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5565 "to the OpenFlow controller.");
5572 xlate_out_uninit(&trace.xout);
5579 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5580 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5583 unixctl_command_reply(conn, NULL);
5587 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5588 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5591 unixctl_command_reply(conn, NULL);
5594 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5595 * 'reply' describing the results. */
5597 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5599 struct cls_cursor cursor;
5600 struct facet *facet;
5604 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5605 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5606 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5607 if (!facet_check_consistency(facet)) {
5611 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5613 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5617 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5618 ofproto->up.name, errors);
5620 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5625 ofproto_dpif_self_check(struct unixctl_conn *conn,
5626 int argc, const char *argv[], void *aux OVS_UNUSED)
5628 struct ds reply = DS_EMPTY_INITIALIZER;
5629 struct ofproto_dpif *ofproto;
5632 ofproto = ofproto_dpif_lookup(argv[1]);
5634 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5635 "ofproto/list for help)");
5638 ofproto_dpif_self_check__(ofproto, &reply);
5640 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5641 ofproto_dpif_self_check__(ofproto, &reply);
5645 unixctl_command_reply(conn, ds_cstr(&reply));
5649 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5650 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5651 * to destroy 'ofproto_shash' and free the returned value. */
5652 static const struct shash_node **
5653 get_ofprotos(struct shash *ofproto_shash)
5655 const struct ofproto_dpif *ofproto;
5657 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5658 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5659 shash_add_nocopy(ofproto_shash, name, ofproto);
5662 return shash_sort(ofproto_shash);
5666 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5667 const char *argv[] OVS_UNUSED,
5668 void *aux OVS_UNUSED)
5670 struct ds ds = DS_EMPTY_INITIALIZER;
5671 struct shash ofproto_shash;
5672 const struct shash_node **sorted_ofprotos;
5675 shash_init(&ofproto_shash);
5676 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5677 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5678 const struct shash_node *node = sorted_ofprotos[i];
5679 ds_put_format(&ds, "%s\n", node->name);
5682 shash_destroy(&ofproto_shash);
5683 free(sorted_ofprotos);
5685 unixctl_command_reply(conn, ds_cstr(&ds));
5690 show_dp_rates(struct ds *ds, const char *heading,
5691 const struct avg_subfacet_rates *rates)
5693 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5694 heading, rates->add_rate, rates->del_rate);
5698 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5700 const struct shash_node **ofprotos;
5701 struct ofproto_dpif *ofproto;
5702 struct shash ofproto_shash;
5703 uint64_t n_hit, n_missed;
5704 long long int minutes;
5707 n_hit = n_missed = 0;
5708 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5709 if (ofproto->backer == backer) {
5710 n_missed += ofproto->n_missed;
5711 n_hit += ofproto->n_hit;
5715 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5716 dpif_name(backer->dpif), n_hit, n_missed);
5717 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5718 " life span: %lldms\n", hmap_count(&backer->subfacets),
5719 backer->avg_n_subfacet, backer->max_n_subfacet,
5720 backer->avg_subfacet_life);
5722 minutes = (time_msec() - backer->created) / (1000 * 60);
5723 if (minutes >= 60) {
5724 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5726 if (minutes >= 60 * 24) {
5727 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5729 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5731 shash_init(&ofproto_shash);
5732 ofprotos = get_ofprotos(&ofproto_shash);
5733 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5734 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5735 const struct shash_node **ports;
5738 if (ofproto->backer != backer) {
5742 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5743 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5745 ports = shash_sort(&ofproto->up.port_by_name);
5746 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5747 const struct shash_node *node = ports[j];
5748 struct ofport *ofport = node->data;
5750 odp_port_t odp_port;
5752 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5755 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5756 if (odp_port != ODPP_NONE) {
5757 ds_put_format(ds, "%"PRIu32":", odp_port);
5759 ds_put_cstr(ds, "none:");
5762 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5765 if (!netdev_get_config(ofport->netdev, &config)) {
5766 const struct smap_node **nodes;
5769 nodes = smap_sort(&config);
5770 for (i = 0; i < smap_count(&config); i++) {
5771 const struct smap_node *node = nodes[i];
5772 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5773 node->key, node->value);
5777 smap_destroy(&config);
5779 ds_put_char(ds, ')');
5780 ds_put_char(ds, '\n');
5784 shash_destroy(&ofproto_shash);
5789 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5790 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5792 struct ds ds = DS_EMPTY_INITIALIZER;
5793 const struct shash_node **backers;
5796 backers = shash_sort(&all_dpif_backers);
5797 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5798 dpif_show_backer(backers[i]->data, &ds);
5802 unixctl_command_reply(conn, ds_cstr(&ds));
5806 /* Dump the megaflow (facet) cache. This is useful to check the
5807 * correctness of flow wildcarding, since the same mechanism is used for
5808 * both xlate caching and kernel wildcarding.
5810 * It's important to note that in the output the flow description uses
5811 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5813 * This command is only needed for advanced debugging, so it's not
5814 * documented in the man page. */
5816 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5817 int argc OVS_UNUSED, const char *argv[],
5818 void *aux OVS_UNUSED)
5820 struct ds ds = DS_EMPTY_INITIALIZER;
5821 const struct ofproto_dpif *ofproto;
5822 long long int now = time_msec();
5823 struct cls_cursor cursor;
5824 struct facet *facet;
5826 ofproto = ofproto_dpif_lookup(argv[1]);
5828 unixctl_command_reply_error(conn, "no such bridge");
5832 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5833 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5834 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5835 cls_rule_format(&facet->cr, &ds);
5836 ds_put_cstr(&ds, ", ");
5837 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5838 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5839 ds_put_cstr(&ds, "Datapath actions: ");
5840 if (facet->xout.slow) {
5841 uint64_t slow_path_stub[128 / 8];
5842 const struct nlattr *actions;
5845 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5846 slow_path_stub, sizeof slow_path_stub,
5847 &actions, &actions_len);
5848 format_odp_actions(&ds, actions, actions_len);
5850 format_odp_actions(&ds, facet->xout.odp_actions.data,
5851 facet->xout.odp_actions.size);
5853 ds_put_cstr(&ds, "\n");
5855 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5857 ds_chomp(&ds, '\n');
5858 unixctl_command_reply(conn, ds_cstr(&ds));
5862 /* Disable using the megaflows.
5864 * This command is only needed for advanced debugging, so it's not
5865 * documented in the man page. */
5867 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5868 int argc OVS_UNUSED,
5869 const char *argv[] OVS_UNUSED,
5870 void *aux OVS_UNUSED)
5872 struct ofproto_dpif *ofproto;
5874 enable_megaflows = false;
5876 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5877 flush(&ofproto->up);
5880 unixctl_command_reply(conn, "megaflows disabled");
5883 /* Re-enable using megaflows.
5885 * This command is only needed for advanced debugging, so it's not
5886 * documented in the man page. */
5888 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5889 int argc OVS_UNUSED,
5890 const char *argv[] OVS_UNUSED,
5891 void *aux OVS_UNUSED)
5893 struct ofproto_dpif *ofproto;
5895 enable_megaflows = true;
5897 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5898 flush(&ofproto->up);
5901 unixctl_command_reply(conn, "megaflows enabled");
5905 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5906 int argc OVS_UNUSED, const char *argv[],
5907 void *aux OVS_UNUSED)
5909 struct ds ds = DS_EMPTY_INITIALIZER;
5910 const struct ofproto_dpif *ofproto;
5911 struct subfacet *subfacet;
5913 ofproto = ofproto_dpif_lookup(argv[1]);
5915 unixctl_command_reply_error(conn, "no such bridge");
5919 update_stats(ofproto->backer);
5921 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5922 struct facet *facet = subfacet->facet;
5923 struct odputil_keybuf maskbuf;
5926 if (facet->ofproto != ofproto) {
5930 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5931 if (enable_megaflows) {
5932 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5933 &facet->flow, UINT32_MAX);
5936 odp_flow_format(subfacet->key, subfacet->key_len,
5937 mask.data, mask.size, &ds, false);
5939 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5940 subfacet->dp_packet_count, subfacet->dp_byte_count);
5941 if (subfacet->used) {
5942 ds_put_format(&ds, "%.3fs",
5943 (time_msec() - subfacet->used) / 1000.0);
5945 ds_put_format(&ds, "never");
5947 if (subfacet->facet->tcp_flags) {
5948 ds_put_cstr(&ds, ", flags:");
5949 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5952 ds_put_cstr(&ds, ", actions:");
5953 if (facet->xout.slow) {
5954 uint64_t slow_path_stub[128 / 8];
5955 const struct nlattr *actions;
5958 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5959 slow_path_stub, sizeof slow_path_stub,
5960 &actions, &actions_len);
5961 format_odp_actions(&ds, actions, actions_len);
5963 format_odp_actions(&ds, facet->xout.odp_actions.data,
5964 facet->xout.odp_actions.size);
5966 ds_put_char(&ds, '\n');
5969 unixctl_command_reply(conn, ds_cstr(&ds));
5974 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5975 int argc OVS_UNUSED, const char *argv[],
5976 void *aux OVS_UNUSED)
5978 struct ds ds = DS_EMPTY_INITIALIZER;
5979 struct ofproto_dpif *ofproto;
5981 ofproto = ofproto_dpif_lookup(argv[1]);
5983 unixctl_command_reply_error(conn, "no such bridge");
5987 flush(&ofproto->up);
5989 unixctl_command_reply(conn, ds_cstr(&ds));
5994 ofproto_dpif_unixctl_init(void)
5996 static bool registered;
6002 unixctl_command_register(
6004 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6005 1, 3, ofproto_unixctl_trace, NULL);
6006 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6007 ofproto_unixctl_fdb_flush, NULL);
6008 unixctl_command_register("fdb/show", "bridge", 1, 1,
6009 ofproto_unixctl_fdb_show, NULL);
6010 unixctl_command_register("ofproto/clog", "", 0, 0,
6011 ofproto_dpif_clog, NULL);
6012 unixctl_command_register("ofproto/unclog", "", 0, 0,
6013 ofproto_dpif_unclog, NULL);
6014 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6015 ofproto_dpif_self_check, NULL);
6016 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6017 ofproto_unixctl_dpif_dump_dps, NULL);
6018 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6020 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6021 ofproto_unixctl_dpif_dump_flows, NULL);
6022 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6023 ofproto_unixctl_dpif_del_flows, NULL);
6024 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6025 ofproto_unixctl_dpif_dump_megaflows, NULL);
6026 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6027 ofproto_unixctl_dpif_disable_megaflows, NULL);
6028 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6029 ofproto_unixctl_dpif_enable_megaflows, NULL);
6032 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6034 * This is deprecated. It is only for compatibility with broken device drivers
6035 * in old versions of Linux that do not properly support VLANs when VLAN
6036 * devices are not used. When broken device drivers are no longer in
6037 * widespread use, we will delete these interfaces. */
6040 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6042 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6043 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6045 if (realdev_ofp_port == ofport->realdev_ofp_port
6046 && vid == ofport->vlandev_vid) {
6050 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6052 if (ofport->realdev_ofp_port) {
6055 if (realdev_ofp_port && ofport->bundle) {
6056 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6057 * themselves be part of a bundle. */
6058 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6061 ofport->realdev_ofp_port = realdev_ofp_port;
6062 ofport->vlandev_vid = vid;
6064 if (realdev_ofp_port) {
6065 vsp_add(ofport, realdev_ofp_port, vid);
6072 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6074 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6078 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6079 OVS_EXCLUDED(ofproto->vsp_mutex)
6083 ovs_mutex_lock(&ofproto->vsp_mutex);
6084 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6085 ovs_mutex_unlock(&ofproto->vsp_mutex);
6090 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6091 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6092 OVS_REQUIRES(ofproto->vsp_mutex)
6094 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6095 int vid = vlan_tci_to_vid(vlan_tci);
6096 const struct vlan_splinter *vsp;
6098 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6099 hash_realdev_vid(realdev_ofp_port, vid),
6100 &ofproto->realdev_vid_map) {
6101 if (vsp->realdev_ofp_port == realdev_ofp_port
6102 && vsp->vid == vid) {
6103 return vsp->vlandev_ofp_port;
6107 return realdev_ofp_port;
6110 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6111 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6112 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6113 * 'vlan_tci' 9, it would return the port number of eth0.9.
6115 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6116 * function just returns its 'realdev_ofp_port' argument. */
6118 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6119 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6120 OVS_EXCLUDED(ofproto->vsp_mutex)
6124 ovs_mutex_lock(&ofproto->vsp_mutex);
6125 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6126 ovs_mutex_unlock(&ofproto->vsp_mutex);
6130 static struct vlan_splinter *
6131 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6133 struct vlan_splinter *vsp;
6135 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6136 hash_ofp_port(vlandev_ofp_port),
6137 &ofproto->vlandev_map) {
6138 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6146 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6147 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6148 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6149 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6150 * eth0 and store 9 in '*vid'.
6152 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6153 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6156 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6157 ofp_port_t vlandev_ofp_port, int *vid)
6158 OVS_REQUIRES(ofproto->vsp_mutex)
6160 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6161 const struct vlan_splinter *vsp;
6163 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6168 return vsp->realdev_ofp_port;
6174 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6175 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6176 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6177 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6178 * always the case unless VLAN splinters are enabled), returns false without
6179 * making any changes. */
6181 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6182 OVS_EXCLUDED(ofproto->vsp_mutex)
6187 ovs_mutex_lock(&ofproto->vsp_mutex);
6188 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6189 ovs_mutex_unlock(&ofproto->vsp_mutex);
6194 /* Cause the flow to be processed as if it came in on the real device with
6195 * the VLAN device's VLAN ID. */
6196 flow->in_port.ofp_port = realdev;
6197 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6202 vsp_remove(struct ofport_dpif *port)
6204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6205 struct vlan_splinter *vsp;
6207 ovs_mutex_lock(&ofproto->vsp_mutex);
6208 vsp = vlandev_find(ofproto, port->up.ofp_port);
6210 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6211 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6214 port->realdev_ofp_port = 0;
6216 VLOG_ERR("missing vlan device record");
6218 ovs_mutex_unlock(&ofproto->vsp_mutex);
6222 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6226 ovs_mutex_lock(&ofproto->vsp_mutex);
6227 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6228 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6229 == realdev_ofp_port)) {
6230 struct vlan_splinter *vsp;
6232 vsp = xmalloc(sizeof *vsp);
6233 vsp->realdev_ofp_port = realdev_ofp_port;
6234 vsp->vlandev_ofp_port = port->up.ofp_port;
6237 port->realdev_ofp_port = realdev_ofp_port;
6239 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6240 hash_ofp_port(port->up.ofp_port));
6241 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6242 hash_realdev_vid(realdev_ofp_port, vid));
6244 VLOG_ERR("duplicate vlan device record");
6246 ovs_mutex_unlock(&ofproto->vsp_mutex);
6250 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6252 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6253 return ofport ? ofport->odp_port : ODPP_NONE;
6256 struct ofport_dpif *
6257 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6259 struct ofport_dpif *port;
6261 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6262 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6263 &backer->odp_to_ofport_map) {
6264 if (port->odp_port == odp_port) {
6265 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6270 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6275 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6277 struct ofport_dpif *port;
6279 port = odp_port_to_ofport(ofproto->backer, odp_port);
6280 if (port && &ofproto->up == port->up.ofproto) {
6281 return port->up.ofp_port;
6287 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6288 * most heavily weighted element. 'base' designates the rate of decay: after
6289 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6292 exp_mavg(double *avg, int base, double new)
6294 *avg = (*avg * (base - 1) + new) / base;
6298 update_moving_averages(struct dpif_backer *backer)
6300 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6301 long long int minutes = (time_msec() - backer->created) / min_ms;
6304 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6306 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6309 backer->lifetime.add_rate = 0.0;
6310 backer->lifetime.del_rate = 0.0;
6313 /* Update hourly averages on the minute boundaries. */
6314 if (time_msec() - backer->last_minute >= min_ms) {
6315 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6316 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6318 /* Update daily averages on the hour boundaries. */
6319 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6320 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6321 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6324 backer->total_subfacet_add_count += backer->subfacet_add_count;
6325 backer->total_subfacet_del_count += backer->subfacet_del_count;
6326 backer->subfacet_add_count = 0;
6327 backer->subfacet_del_count = 0;
6328 backer->last_minute += min_ms;
6332 const struct ofproto_class ofproto_dpif_class = {
6367 port_is_lacp_current,
6368 NULL, /* rule_choose_table */
6377 rule_modify_actions,
6391 get_stp_port_status,
6398 is_mirror_output_bundle,
6399 forward_bpdu_changed,
6400 set_mac_table_config,
6402 NULL, /* meter_get_features */
6403 NULL, /* meter_set */
6404 NULL, /* meter_get */
6405 NULL, /* meter_del */