2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
34 #include "guarded-list.h"
38 #include "mac-learning.h"
39 #include "meta-flow.h"
40 #include "multipath.h"
41 #include "netdev-vport.h"
46 #include "odp-execute.h"
49 #include "ofp-actions.h"
50 #include "ofp-parse.h"
51 #include "ofp-print.h"
52 #include "ofproto-dpif-governor.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-sflow.h"
56 #include "ofproto-dpif-upcall.h"
57 #include "ofproto-dpif-xlate.h"
58 #include "poll-loop.h"
63 #include "unaligned.h"
65 #include "vlan-bitmap.h"
68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
70 COVERAGE_DEFINE(ofproto_dpif_expired);
71 COVERAGE_DEFINE(facet_changed_rule);
72 COVERAGE_DEFINE(facet_revalidate);
73 COVERAGE_DEFINE(facet_unexpected);
74 COVERAGE_DEFINE(facet_suppress);
75 COVERAGE_DEFINE(subfacet_install_fail);
76 COVERAGE_DEFINE(packet_in_overflow);
77 COVERAGE_DEFINE(flow_mod_overflow);
79 /* Number of implemented OpenFlow tables. */
80 enum { N_TABLES = 255 };
81 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
82 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
92 * - Do include packets and bytes from facets that have been deleted or
93 * whose own statistics have been folded into the rule.
95 * - Do include packets and bytes sent "by hand" that were accounted to
96 * the rule without any facet being involved (this is a rare corner
97 * case in rule_execute()).
99 * - Do not include packet or bytes that can be obtained from any facet's
100 * packet_count or byte_count member or that can be obtained from the
101 * datapath by, e.g., dpif_flow_get() for any subfacet.
103 struct ovs_mutex stats_mutex;
104 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
105 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
108 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
110 static struct rule_dpif *rule_dpif_cast(const struct rule *);
113 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 void *aux; /* Key supplied by ofproto's client. */
116 char *name; /* Identifier for log messages. */
119 struct list ports; /* Contains "struct ofport"s. */
120 enum port_vlan_mode vlan_mode; /* VLAN mode */
121 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
122 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
123 * NULL if all VLANs are trunked. */
124 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
125 struct bond *bond; /* Nonnull iff more than one port. */
126 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
129 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
132 static void bundle_remove(struct ofport *);
133 static void bundle_update(struct ofbundle *);
134 static void bundle_destroy(struct ofbundle *);
135 static void bundle_del_port(struct ofport_dpif *);
136 static void bundle_run(struct ofbundle *);
137 static void bundle_wait(struct ofbundle *);
139 static void stp_run(struct ofproto_dpif *ofproto);
140 static void stp_wait(struct ofproto_dpif *ofproto);
141 static int set_stp_port(struct ofport *,
142 const struct ofproto_port_stp_settings *);
144 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
145 enum slow_path_reason,
146 uint64_t *stub, size_t stub_size,
147 const struct nlattr **actionsp,
148 size_t *actions_lenp);
150 /* A subfacet (see "struct subfacet" below) has three possible installation
153 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
154 * case just after the subfacet is created, just before the subfacet is
155 * destroyed, or if the datapath returns an error when we try to install a
158 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
160 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
161 * ofproto_dpif is installed in the datapath.
164 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
165 SF_FAST_PATH, /* Full actions are installed. */
166 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
169 /* A dpif flow and actions associated with a facet.
171 * See also the large comment on struct facet. */
174 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
175 struct list list_node; /* In struct facet's 'facets' list. */
176 struct facet *facet; /* Owning facet. */
177 struct dpif_backer *backer; /* Owning backer. */
179 enum odp_key_fitness key_fitness;
183 long long int used; /* Time last used; time created if not used. */
184 long long int created; /* Time created. */
186 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
187 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
189 enum subfacet_path path; /* Installed in datapath? */
192 #define SUBFACET_DESTROY_MAX_BATCH 50
194 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
195 static struct subfacet *subfacet_find(struct dpif_backer *,
196 const struct nlattr *key, size_t key_len,
198 static void subfacet_destroy(struct subfacet *);
199 static void subfacet_destroy__(struct subfacet *);
200 static void subfacet_destroy_batch(struct dpif_backer *,
201 struct subfacet **, int n);
202 static void subfacet_reset_dp_stats(struct subfacet *,
203 struct dpif_flow_stats *);
204 static void subfacet_update_stats(struct subfacet *,
205 const struct dpif_flow_stats *);
206 static int subfacet_install(struct subfacet *,
207 const struct ofpbuf *odp_actions,
208 struct dpif_flow_stats *);
209 static void subfacet_uninstall(struct subfacet *);
211 /* A unique, non-overlapping instantiation of an OpenFlow flow.
213 * A facet associates a "struct flow", which represents the Open vSwitch
214 * userspace idea of an exact-match flow, with one or more subfacets.
215 * While the facet is created based on an exact-match flow, it is stored
216 * within the ofproto based on the wildcards that could be expressed
217 * based on the flow table and other configuration. (See the 'wc'
218 * description in "struct xlate_out" for more details.)
220 * Each subfacet tracks the datapath's idea of the flow equivalent to
221 * the facet. When the kernel module (or other dpif implementation) and
222 * Open vSwitch userspace agree on the definition of a flow key, there
223 * is exactly one subfacet per facet. If the dpif implementation
224 * supports more-specific flow matching than userspace, however, a facet
225 * can have more than one subfacet. Examples include the dpif
226 * implementation not supporting the same wildcards as userspace or some
227 * distinction in flow that userspace simply doesn't understand.
229 * Flow expiration works in terms of subfacets, so a facet must have at
230 * least one subfacet or it will never expire, leaking memory. */
233 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
234 struct ofproto_dpif *ofproto;
237 struct list subfacets;
238 long long int used; /* Time last used; time created if not used. */
241 struct flow flow; /* Flow of the creating subfacet. */
242 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
246 * - Do include packets and bytes sent "by hand", e.g. with
249 * - Do include packets and bytes that were obtained from the datapath
250 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
251 * DPIF_FP_ZERO_STATS).
253 * - Do not include packets or bytes that can be obtained from the
254 * datapath for any existing subfacet.
256 uint64_t packet_count; /* Number of packets received. */
257 uint64_t byte_count; /* Number of bytes received. */
259 /* Resubmit statistics. */
260 uint64_t prev_packet_count; /* Number of packets from last stats push. */
261 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
262 long long int prev_used; /* Used time from last stats push. */
265 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
266 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
267 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
269 struct xlate_out xout;
271 /* Storage for a single subfacet, to reduce malloc() time and space
272 * overhead. (A facet always has at least one subfacet and in the common
273 * case has exactly one subfacet. However, 'one_subfacet' may not
274 * always be valid, since it could have been removed after newer
275 * subfacets were pushed onto the 'subfacets' list.) */
276 struct subfacet one_subfacet;
278 long long int learn_rl; /* Rate limiter for facet_learn(). */
281 static struct facet *facet_create(const struct flow_miss *);
282 static void facet_remove(struct facet *);
283 static void facet_free(struct facet *);
285 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
286 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
287 const struct flow *);
288 static bool facet_revalidate(struct facet *);
289 static bool facet_check_consistency(struct facet *);
291 static void facet_flush_stats(struct facet *);
293 static void facet_reset_counters(struct facet *);
294 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
295 struct dpif_flow_stats *, bool may_learn);
296 static void facet_push_stats(struct facet *, bool may_learn);
297 static void facet_learn(struct facet *);
298 static void facet_account(struct facet *);
299 static void push_all_stats(void);
301 static bool facet_is_controller_flow(struct facet *);
304 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
308 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
309 struct list bundle_node; /* In struct ofbundle's "ports" list. */
310 struct cfm *cfm; /* Connectivity Fault Management, if any. */
311 struct bfd *bfd; /* BFD, if any. */
312 bool may_enable; /* May be enabled in bonds. */
313 bool is_tunnel; /* This port is a tunnel. */
314 long long int carrier_seq; /* Carrier status changes. */
315 struct ofport_dpif *peer; /* Peer if patch port. */
318 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
319 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
320 long long int stp_state_entered;
322 /* Queue to DSCP mapping. */
323 struct ofproto_port_queue *qdscp;
326 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
328 * This is deprecated. It is only for compatibility with broken device
329 * drivers in old versions of Linux that do not properly support VLANs when
330 * VLAN devices are not used. When broken device drivers are no longer in
331 * widespread use, we will delete these interfaces. */
332 ofp_port_t realdev_ofp_port;
336 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
338 * This is deprecated. It is only for compatibility with broken device drivers
339 * in old versions of Linux that do not properly support VLANs when VLAN
340 * devices are not used. When broken device drivers are no longer in
341 * widespread use, we will delete these interfaces. */
342 struct vlan_splinter {
343 struct hmap_node realdev_vid_node;
344 struct hmap_node vlandev_node;
345 ofp_port_t realdev_ofp_port;
346 ofp_port_t vlandev_ofp_port;
350 static void vsp_remove(struct ofport_dpif *);
351 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
353 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
356 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
359 static struct ofport_dpif *
360 ofport_dpif_cast(const struct ofport *ofport)
362 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
365 static void port_run(struct ofport_dpif *);
366 static void port_run_fast(struct ofport_dpif *);
367 static void port_wait(struct ofport_dpif *);
368 static int set_bfd(struct ofport *, const struct smap *);
369 static int set_cfm(struct ofport *, const struct cfm_settings *);
370 static void ofport_update_peer(struct ofport_dpif *);
371 static void run_fast_rl(void);
372 static int run_fast(struct ofproto *);
374 struct dpif_completion {
375 struct list list_node;
376 struct ofoperation *op;
379 /* Reasons that we might need to revalidate every facet, and corresponding
382 * A value of 0 means that there is no need to revalidate.
384 * It would be nice to have some cleaner way to integrate with coverage
385 * counters, but with only a few reasons I guess this is good enough for
387 enum revalidate_reason {
388 REV_RECONFIGURE = 1, /* Switch configuration changed. */
389 REV_STP, /* Spanning tree protocol port status change. */
390 REV_BOND, /* Bonding changed. */
391 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
392 REV_FLOW_TABLE, /* Flow table changed. */
393 REV_MAC_LEARNING, /* Mac learning changed. */
394 REV_INCONSISTENCY /* Facet self-check failed. */
396 COVERAGE_DEFINE(rev_reconfigure);
397 COVERAGE_DEFINE(rev_stp);
398 COVERAGE_DEFINE(rev_bond);
399 COVERAGE_DEFINE(rev_port_toggled);
400 COVERAGE_DEFINE(rev_flow_table);
401 COVERAGE_DEFINE(rev_mac_learning);
402 COVERAGE_DEFINE(rev_inconsistency);
404 struct avg_subfacet_rates {
405 double add_rate; /* Moving average of new flows created per minute. */
406 double del_rate; /* Moving average of flows deleted per minute. */
409 /* All datapaths of a given type share a single dpif backer instance. */
415 struct timer next_expiration;
417 struct ovs_rwlock odp_to_ofport_lock;
418 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
420 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
422 /* Facet revalidation flags applying to facets which use this backer. */
423 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
425 struct hmap drop_keys; /* Set of dropped odp keys. */
426 bool recv_set_enable; /* Enables or disables receiving packets. */
428 struct hmap subfacets;
429 struct governor *governor;
431 /* Subfacet statistics.
433 * These keep track of the total number of subfacets added and deleted and
434 * flow life span. They are useful for computing the flow rates stats
435 * exposed via "ovs-appctl dpif/show". The goal is to learn about
436 * traffic patterns in ways that we can use later to improve Open vSwitch
437 * performance in new situations. */
438 long long int created; /* Time when it is created. */
439 unsigned max_n_subfacet; /* Maximum number of flows */
440 unsigned avg_n_subfacet; /* Average number of flows. */
441 long long int avg_subfacet_life; /* Average life span of subfacets. */
443 /* The average number of subfacets... */
444 struct avg_subfacet_rates hourly; /* ...over the last hour. */
445 struct avg_subfacet_rates daily; /* ...over the last day. */
446 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
447 long long int last_minute; /* Last time 'hourly' was updated. */
449 /* Number of subfacets added or deleted since 'last_minute'. */
450 unsigned subfacet_add_count;
451 unsigned subfacet_del_count;
453 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
454 unsigned long long int total_subfacet_add_count;
455 unsigned long long int total_subfacet_del_count;
457 /* Number of upcall handling threads. */
458 unsigned int n_handler_threads;
461 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
462 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
464 static void drop_key_clear(struct dpif_backer *);
465 static void update_moving_averages(struct dpif_backer *backer);
467 struct ofproto_dpif {
468 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
470 struct dpif_backer *backer;
472 /* Special OpenFlow rules. */
473 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
474 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
475 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
478 struct netflow *netflow;
479 struct dpif_sflow *sflow;
480 struct dpif_ipfix *ipfix;
481 struct hmap bundles; /* Contains "struct ofbundle"s. */
482 struct mac_learning *ml;
483 bool has_bonded_bundles;
484 struct mbridge *mbridge;
487 struct classifier facets; /* Contains 'struct facet's. */
488 long long int consistency_rl;
490 struct netdev_stats stats; /* To account packets generated and consumed in
495 long long int stp_last_tick;
497 /* VLAN splinters. */
498 struct ovs_mutex vsp_mutex;
499 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
500 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
503 struct sset ports; /* Set of standard port names. */
504 struct sset ghost_ports; /* Ports with no datapath port. */
505 struct sset port_poll_set; /* Queued names for port_poll() reply. */
506 int port_poll_errno; /* Last errno for port_poll() reply. */
508 /* Per ofproto's dpif stats. */
513 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
516 /* By default, flows in the datapath are wildcarded (megaflows). They
517 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
518 static bool enable_megaflows = true;
520 /* All existing ofproto_dpif instances, indexed by ->up.name. */
521 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
523 static void ofproto_dpif_unixctl_init(void);
525 static inline struct ofproto_dpif *
526 ofproto_dpif_cast(const struct ofproto *ofproto)
528 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
529 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
532 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
533 ofp_port_t ofp_port);
534 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
535 const struct ofpbuf *packet, struct ds *);
538 static void handle_upcalls(struct dpif_backer *);
540 /* Flow expiration. */
541 static int expire(struct dpif_backer *);
544 static void send_netflow_active_timeouts(struct ofproto_dpif *);
547 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
549 /* Global variables. */
550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
552 /* Initial mappings of port to bridge mappings. */
553 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
555 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
558 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
559 struct ofputil_flow_mod *fm)
561 ofproto_flow_mod(&ofproto->up, fm);
564 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
565 * Takes ownership of 'pin' and pin->packet. */
567 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
568 struct ofputil_packet_in *pin)
570 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
571 COVERAGE_INC(packet_in_overflow);
572 free(CONST_CAST(void *, pin->packet));
577 /* Factory functions. */
580 init(const struct shash *iface_hints)
582 struct shash_node *node;
584 /* Make a local copy, since we don't own 'iface_hints' elements. */
585 SHASH_FOR_EACH(node, iface_hints) {
586 const struct iface_hint *orig_hint = node->data;
587 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
589 new_hint->br_name = xstrdup(orig_hint->br_name);
590 new_hint->br_type = xstrdup(orig_hint->br_type);
591 new_hint->ofp_port = orig_hint->ofp_port;
593 shash_add(&init_ofp_ports, node->name, new_hint);
598 enumerate_types(struct sset *types)
600 dp_enumerate_types(types);
604 enumerate_names(const char *type, struct sset *names)
606 struct ofproto_dpif *ofproto;
609 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
610 if (strcmp(type, ofproto->up.type)) {
613 sset_add(names, ofproto->up.name);
620 del(const char *type, const char *name)
625 error = dpif_open(name, type, &dpif);
627 error = dpif_delete(dpif);
634 port_open_type(const char *datapath_type, const char *port_type)
636 return dpif_port_open_type(datapath_type, port_type);
639 /* Type functions. */
641 static void process_dpif_port_changes(struct dpif_backer *);
642 static void process_dpif_all_ports_changed(struct dpif_backer *);
643 static void process_dpif_port_change(struct dpif_backer *,
644 const char *devname);
645 static void process_dpif_port_error(struct dpif_backer *, int error);
647 static struct ofproto_dpif *
648 lookup_ofproto_dpif_by_port_name(const char *name)
650 struct ofproto_dpif *ofproto;
652 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
653 if (sset_contains(&ofproto->ports, name)) {
662 type_run(const char *type)
664 static long long int push_timer = LLONG_MIN;
665 struct dpif_backer *backer;
667 backer = shash_find_data(&all_dpif_backers, type);
669 /* This is not necessarily a problem, since backers are only
670 * created on demand. */
674 dpif_run(backer->dpif);
676 /* The most natural place to push facet statistics is when they're pulled
677 * from the datapath. However, when there are many flows in the datapath,
678 * this expensive operation can occur so frequently, that it reduces our
679 * ability to quickly set up flows. To reduce the cost, we push statistics
681 if (time_msec() > push_timer) {
682 push_timer = time_msec() + 2000;
686 /* If vswitchd started with other_config:flow_restore_wait set as "true",
687 * and the configuration has now changed to "false", enable receiving
688 * packets from the datapath. */
689 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
692 backer->recv_set_enable = true;
694 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
696 udpif_recv_set(backer->udpif, 0, false);
697 VLOG_ERR("Failed to enable receiving packets in dpif.");
700 udpif_recv_set(backer->udpif, n_handler_threads,
701 backer->recv_set_enable);
702 dpif_flow_flush(backer->dpif);
703 backer->need_revalidate = REV_RECONFIGURE;
706 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
707 * to reset the handler threads. */
708 if (backer->n_handler_threads != n_handler_threads) {
709 udpif_recv_set(backer->udpif, n_handler_threads,
710 backer->recv_set_enable);
711 backer->n_handler_threads = n_handler_threads;
714 if (backer->need_revalidate) {
715 struct ofproto_dpif *ofproto;
716 struct simap_node *node;
717 struct simap tmp_backers;
719 /* Handle tunnel garbage collection. */
720 simap_init(&tmp_backers);
721 simap_swap(&backer->tnl_backers, &tmp_backers);
723 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
724 struct ofport_dpif *iter;
726 if (backer != ofproto->backer) {
730 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
731 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
734 if (!iter->is_tunnel) {
738 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
739 namebuf, sizeof namebuf);
740 node = simap_find(&tmp_backers, dp_port);
742 simap_put(&backer->tnl_backers, dp_port, node->data);
743 simap_delete(&tmp_backers, node);
744 node = simap_find(&backer->tnl_backers, dp_port);
746 node = simap_find(&backer->tnl_backers, dp_port);
748 odp_port_t odp_port = ODPP_NONE;
750 if (!dpif_port_add(backer->dpif, iter->up.netdev,
752 simap_put(&backer->tnl_backers, dp_port,
753 odp_to_u32(odp_port));
754 node = simap_find(&backer->tnl_backers, dp_port);
759 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
760 if (tnl_port_reconfigure(iter, iter->up.netdev,
762 backer->need_revalidate = REV_RECONFIGURE;
767 SIMAP_FOR_EACH (node, &tmp_backers) {
768 dpif_port_del(backer->dpif, u32_to_odp(node->data));
770 simap_destroy(&tmp_backers);
772 switch (backer->need_revalidate) {
773 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
774 case REV_STP: COVERAGE_INC(rev_stp); break;
775 case REV_BOND: COVERAGE_INC(rev_bond); break;
776 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
777 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
778 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
779 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
781 backer->need_revalidate = 0;
783 /* Clear the drop_keys in case we should now be accepting some
784 * formerly dropped flows. */
785 drop_key_clear(backer);
787 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
788 struct facet *facet, *next;
789 struct ofport_dpif *ofport;
790 struct cls_cursor cursor;
791 struct ofbundle *bundle;
793 if (ofproto->backer != backer) {
797 ovs_rwlock_wrlock(&xlate_rwlock);
798 xlate_ofproto_set(ofproto, ofproto->up.name,
799 ofproto->backer->dpif, ofproto->miss_rule,
800 ofproto->no_packet_in_rule, ofproto->ml,
801 ofproto->stp, ofproto->mbridge,
802 ofproto->sflow, ofproto->ipfix,
803 ofproto->up.frag_handling,
804 ofproto->up.forward_bpdu,
805 connmgr_has_in_band(ofproto->up.connmgr),
806 ofproto->netflow != NULL);
808 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
809 xlate_bundle_set(ofproto, bundle, bundle->name,
810 bundle->vlan_mode, bundle->vlan,
811 bundle->trunks, bundle->use_priority_tags,
812 bundle->bond, bundle->lacp,
816 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
817 int stp_port = ofport->stp_port
818 ? stp_port_no(ofport->stp_port)
820 xlate_ofport_set(ofproto, ofport->bundle, ofport,
821 ofport->up.ofp_port, ofport->odp_port,
822 ofport->up.netdev, ofport->cfm,
823 ofport->bfd, ofport->peer, stp_port,
824 ofport->qdscp, ofport->n_qdscp,
825 ofport->up.pp.config, ofport->is_tunnel,
828 ovs_rwlock_unlock(&xlate_rwlock);
830 /* Only ofproto-dpif cares about the facet classifier so we just
831 * lock cls_cursor_init() to appease the thread safety analysis. */
832 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
833 cls_cursor_init(&cursor, &ofproto->facets, NULL);
834 ovs_rwlock_unlock(&ofproto->facets.rwlock);
835 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
836 facet_revalidate(facet);
841 udpif_revalidate(backer->udpif);
844 if (!backer->recv_set_enable) {
845 /* Wake up before a max of 1000ms. */
846 timer_set_duration(&backer->next_expiration, 1000);
847 } else if (timer_expired(&backer->next_expiration)) {
848 int delay = expire(backer);
849 timer_set_duration(&backer->next_expiration, delay);
852 process_dpif_port_changes(backer);
854 if (backer->governor) {
857 governor_run(backer->governor);
859 /* If the governor has shrunk to its minimum size and the number of
860 * subfacets has dwindled, then drop the governor entirely.
862 * For hysteresis, the number of subfacets to drop the governor is
863 * smaller than the number needed to trigger its creation. */
864 n_subfacets = hmap_count(&backer->subfacets);
865 if (n_subfacets * 4 < flow_eviction_threshold
866 && governor_is_idle(backer->governor)) {
867 governor_destroy(backer->governor);
868 backer->governor = NULL;
875 /* Check for and handle port changes in 'backer''s dpif. */
877 process_dpif_port_changes(struct dpif_backer *backer)
883 error = dpif_port_poll(backer->dpif, &devname);
889 process_dpif_all_ports_changed(backer);
893 process_dpif_port_change(backer, devname);
898 process_dpif_port_error(backer, error);
905 process_dpif_all_ports_changed(struct dpif_backer *backer)
907 struct ofproto_dpif *ofproto;
908 struct dpif_port dpif_port;
909 struct dpif_port_dump dump;
910 struct sset devnames;
913 sset_init(&devnames);
914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
915 if (ofproto->backer == backer) {
916 struct ofport *ofport;
918 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
919 sset_add(&devnames, netdev_get_name(ofport->netdev));
923 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
924 sset_add(&devnames, dpif_port.name);
927 SSET_FOR_EACH (devname, &devnames) {
928 process_dpif_port_change(backer, devname);
930 sset_destroy(&devnames);
934 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
936 struct ofproto_dpif *ofproto;
937 struct dpif_port port;
939 /* Don't report on the datapath's device. */
940 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
944 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
945 &all_ofproto_dpifs) {
946 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
951 ofproto = lookup_ofproto_dpif_by_port_name(devname);
952 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
953 /* The port was removed. If we know the datapath,
954 * report it through poll_set(). If we don't, it may be
955 * notifying us of a removal we initiated, so ignore it.
956 * If there's a pending ENOBUFS, let it stand, since
957 * everything will be reevaluated. */
958 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
959 sset_add(&ofproto->port_poll_set, devname);
960 ofproto->port_poll_errno = 0;
962 } else if (!ofproto) {
963 /* The port was added, but we don't know with which
964 * ofproto we should associate it. Delete it. */
965 dpif_port_del(backer->dpif, port.port_no);
967 struct ofport_dpif *ofport;
969 ofport = ofport_dpif_cast(shash_find_data(
970 &ofproto->up.port_by_name, devname));
972 && ofport->odp_port != port.port_no
973 && !odp_port_to_ofport(backer, port.port_no))
975 /* 'ofport''s datapath port number has changed from
976 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
977 * structures to match. */
978 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
979 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
980 ofport->odp_port = port.port_no;
981 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
982 hash_odp_port(port.port_no));
983 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
984 backer->need_revalidate = REV_RECONFIGURE;
987 dpif_port_destroy(&port);
990 /* Propagate 'error' to all ofprotos based on 'backer'. */
992 process_dpif_port_error(struct dpif_backer *backer, int error)
994 struct ofproto_dpif *ofproto;
996 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
997 if (ofproto->backer == backer) {
998 sset_clear(&ofproto->port_poll_set);
999 ofproto->port_poll_errno = error;
1005 dpif_backer_run_fast(struct dpif_backer *backer)
1007 handle_upcalls(backer);
1013 type_run_fast(const char *type)
1015 struct dpif_backer *backer;
1017 backer = shash_find_data(&all_dpif_backers, type);
1019 /* This is not necessarily a problem, since backers are only
1020 * created on demand. */
1024 return dpif_backer_run_fast(backer);
1030 static long long int port_rl = LLONG_MIN;
1032 if (time_msec() >= port_rl) {
1033 struct ofproto_dpif *ofproto;
1035 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1036 run_fast(&ofproto->up);
1038 port_rl = time_msec() + 200;
1043 type_wait(const char *type)
1045 struct dpif_backer *backer;
1047 backer = shash_find_data(&all_dpif_backers, type);
1049 /* This is not necessarily a problem, since backers are only
1050 * created on demand. */
1054 if (backer->governor) {
1055 governor_wait(backer->governor);
1058 timer_wait(&backer->next_expiration);
1059 dpif_wait(backer->dpif);
1060 udpif_wait(backer->udpif);
1063 /* Basic life-cycle. */
1065 static int add_internal_flows(struct ofproto_dpif *);
1067 static struct ofproto *
1070 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1071 return &ofproto->up;
1075 dealloc(struct ofproto *ofproto_)
1077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1082 close_dpif_backer(struct dpif_backer *backer)
1084 struct shash_node *node;
1086 ovs_assert(backer->refcount > 0);
1088 if (--backer->refcount) {
1092 drop_key_clear(backer);
1093 hmap_destroy(&backer->drop_keys);
1095 udpif_destroy(backer->udpif);
1097 simap_destroy(&backer->tnl_backers);
1098 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1099 hmap_destroy(&backer->odp_to_ofport_map);
1100 node = shash_find(&all_dpif_backers, backer->type);
1102 shash_delete(&all_dpif_backers, node);
1103 dpif_close(backer->dpif);
1105 ovs_assert(hmap_is_empty(&backer->subfacets));
1106 hmap_destroy(&backer->subfacets);
1107 governor_destroy(backer->governor);
1112 /* Datapath port slated for removal from datapath. */
1113 struct odp_garbage {
1114 struct list list_node;
1115 odp_port_t odp_port;
1119 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1121 struct dpif_backer *backer;
1122 struct dpif_port_dump port_dump;
1123 struct dpif_port port;
1124 struct shash_node *node;
1125 struct list garbage_list;
1126 struct odp_garbage *garbage, *next;
1132 backer = shash_find_data(&all_dpif_backers, type);
1139 backer_name = xasprintf("ovs-%s", type);
1141 /* Remove any existing datapaths, since we assume we're the only
1142 * userspace controlling the datapath. */
1144 dp_enumerate_names(type, &names);
1145 SSET_FOR_EACH(name, &names) {
1146 struct dpif *old_dpif;
1148 /* Don't remove our backer if it exists. */
1149 if (!strcmp(name, backer_name)) {
1153 if (dpif_open(name, type, &old_dpif)) {
1154 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1156 dpif_delete(old_dpif);
1157 dpif_close(old_dpif);
1160 sset_destroy(&names);
1162 backer = xmalloc(sizeof *backer);
1164 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1167 VLOG_ERR("failed to open datapath of type %s: %s", type,
1168 ovs_strerror(error));
1172 backer->udpif = udpif_create(backer, backer->dpif);
1174 backer->type = xstrdup(type);
1175 backer->governor = NULL;
1176 backer->refcount = 1;
1177 hmap_init(&backer->odp_to_ofport_map);
1178 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1179 hmap_init(&backer->drop_keys);
1180 hmap_init(&backer->subfacets);
1181 timer_set_duration(&backer->next_expiration, 1000);
1182 backer->need_revalidate = 0;
1183 simap_init(&backer->tnl_backers);
1184 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1187 if (backer->recv_set_enable) {
1188 dpif_flow_flush(backer->dpif);
1191 /* Loop through the ports already on the datapath and remove any
1192 * that we don't need anymore. */
1193 list_init(&garbage_list);
1194 dpif_port_dump_start(&port_dump, backer->dpif);
1195 while (dpif_port_dump_next(&port_dump, &port)) {
1196 node = shash_find(&init_ofp_ports, port.name);
1197 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1198 garbage = xmalloc(sizeof *garbage);
1199 garbage->odp_port = port.port_no;
1200 list_push_front(&garbage_list, &garbage->list_node);
1203 dpif_port_dump_done(&port_dump);
1205 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1206 dpif_port_del(backer->dpif, garbage->odp_port);
1207 list_remove(&garbage->list_node);
1211 shash_add(&all_dpif_backers, type, backer);
1213 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1215 VLOG_ERR("failed to listen on datapath of type %s: %s",
1216 type, ovs_strerror(error));
1217 close_dpif_backer(backer);
1220 udpif_recv_set(backer->udpif, n_handler_threads,
1221 backer->recv_set_enable);
1222 backer->n_handler_threads = n_handler_threads;
1224 backer->max_n_subfacet = 0;
1225 backer->created = time_msec();
1226 backer->last_minute = backer->created;
1227 memset(&backer->hourly, 0, sizeof backer->hourly);
1228 memset(&backer->daily, 0, sizeof backer->daily);
1229 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1230 backer->subfacet_add_count = 0;
1231 backer->subfacet_del_count = 0;
1232 backer->total_subfacet_add_count = 0;
1233 backer->total_subfacet_del_count = 0;
1234 backer->avg_n_subfacet = 0;
1235 backer->avg_subfacet_life = 0;
1241 construct(struct ofproto *ofproto_)
1243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1244 struct shash_node *node, *next;
1248 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1253 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1254 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1256 ofproto->netflow = NULL;
1257 ofproto->sflow = NULL;
1258 ofproto->ipfix = NULL;
1259 ofproto->stp = NULL;
1260 hmap_init(&ofproto->bundles);
1261 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1262 ofproto->mbridge = mbridge_create();
1263 ofproto->has_bonded_bundles = false;
1264 ovs_mutex_init(&ofproto->vsp_mutex);
1266 classifier_init(&ofproto->facets);
1267 ofproto->consistency_rl = LLONG_MIN;
1269 guarded_list_init(&ofproto->pins);
1271 ofproto_dpif_unixctl_init();
1273 hmap_init(&ofproto->vlandev_map);
1274 hmap_init(&ofproto->realdev_vid_map);
1276 sset_init(&ofproto->ports);
1277 sset_init(&ofproto->ghost_ports);
1278 sset_init(&ofproto->port_poll_set);
1279 ofproto->port_poll_errno = 0;
1281 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1282 struct iface_hint *iface_hint = node->data;
1284 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1285 /* Check if the datapath already has this port. */
1286 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1287 sset_add(&ofproto->ports, node->name);
1290 free(iface_hint->br_name);
1291 free(iface_hint->br_type);
1293 shash_delete(&init_ofp_ports, node);
1297 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1298 hash_string(ofproto->up.name, 0));
1299 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1301 ofproto_init_tables(ofproto_, N_TABLES);
1302 error = add_internal_flows(ofproto);
1303 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1306 ofproto->n_missed = 0;
1312 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1313 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1315 struct ofputil_flow_mod fm;
1318 match_init_catchall(&fm.match);
1320 match_set_reg(&fm.match, 0, id);
1321 fm.new_cookie = htonll(0);
1322 fm.cookie = htonll(0);
1323 fm.cookie_mask = htonll(0);
1324 fm.modify_cookie = false;
1325 fm.table_id = TBL_INTERNAL;
1326 fm.command = OFPFC_ADD;
1327 fm.idle_timeout = 0;
1328 fm.hard_timeout = 0;
1332 fm.ofpacts = ofpacts->data;
1333 fm.ofpacts_len = ofpacts->size;
1335 error = ofproto_flow_mod(&ofproto->up, &fm);
1337 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1338 id, ofperr_to_string(error));
1342 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1344 rule_dpif_unref(*rulep);
1353 add_internal_flows(struct ofproto_dpif *ofproto)
1355 struct ofpact_controller *controller;
1356 uint64_t ofpacts_stub[128 / 8];
1357 struct ofpbuf ofpacts;
1361 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1364 controller = ofpact_put_CONTROLLER(&ofpacts);
1365 controller->max_len = UINT16_MAX;
1366 controller->controller_id = 0;
1367 controller->reason = OFPR_NO_MATCH;
1368 ofpact_pad(&ofpacts);
1370 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1375 ofpbuf_clear(&ofpacts);
1376 error = add_internal_flow(ofproto, id++, &ofpacts,
1377 &ofproto->no_packet_in_rule);
1382 error = add_internal_flow(ofproto, id++, &ofpacts,
1383 &ofproto->drop_frags_rule);
1388 destruct(struct ofproto *ofproto_)
1390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1391 struct rule_dpif *rule, *next_rule;
1392 struct ofputil_packet_in *pin, *next_pin;
1393 struct facet *facet, *next_facet;
1394 struct cls_cursor cursor;
1395 struct oftable *table;
1398 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1399 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1400 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1401 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1402 facet_remove(facet);
1405 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1406 ovs_rwlock_wrlock(&xlate_rwlock);
1407 xlate_remove_ofproto(ofproto);
1408 ovs_rwlock_unlock(&xlate_rwlock);
1410 /* Ensure that the upcall processing threads have no remaining references
1411 * to the ofproto or anything in it. */
1412 udpif_synchronize(ofproto->backer->udpif);
1414 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1416 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1417 struct cls_cursor cursor;
1419 ovs_rwlock_rdlock(&table->cls.rwlock);
1420 cls_cursor_init(&cursor, &table->cls, NULL);
1421 ovs_rwlock_unlock(&table->cls.rwlock);
1422 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1423 ofproto_rule_delete(&ofproto->up, &rule->up);
1427 guarded_list_pop_all(&ofproto->pins, &pins);
1428 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1429 list_remove(&pin->list_node);
1430 free(CONST_CAST(void *, pin->packet));
1433 guarded_list_destroy(&ofproto->pins);
1435 mbridge_unref(ofproto->mbridge);
1437 netflow_destroy(ofproto->netflow);
1438 dpif_sflow_unref(ofproto->sflow);
1439 hmap_destroy(&ofproto->bundles);
1440 mac_learning_unref(ofproto->ml);
1442 classifier_destroy(&ofproto->facets);
1444 hmap_destroy(&ofproto->vlandev_map);
1445 hmap_destroy(&ofproto->realdev_vid_map);
1447 sset_destroy(&ofproto->ports);
1448 sset_destroy(&ofproto->ghost_ports);
1449 sset_destroy(&ofproto->port_poll_set);
1451 ovs_mutex_destroy(&ofproto->vsp_mutex);
1453 close_dpif_backer(ofproto->backer);
1457 run_fast(struct ofproto *ofproto_)
1459 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1460 struct ofputil_packet_in *pin, *next_pin;
1461 struct ofport_dpif *ofport;
1464 /* Do not perform any periodic activity required by 'ofproto' while
1465 * waiting for flow restore to complete. */
1466 if (ofproto_get_flow_restore_wait()) {
1470 guarded_list_pop_all(&ofproto->pins, &pins);
1471 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1472 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1473 list_remove(&pin->list_node);
1474 free(CONST_CAST(void *, pin->packet));
1478 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1479 port_run_fast(ofport);
1486 run(struct ofproto *ofproto_)
1488 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1489 struct ofport_dpif *ofport;
1490 struct ofbundle *bundle;
1493 if (mbridge_need_revalidate(ofproto->mbridge)) {
1494 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1495 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1496 mac_learning_flush(ofproto->ml);
1497 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1500 /* Do not perform any periodic activity below required by 'ofproto' while
1501 * waiting for flow restore to complete. */
1502 if (ofproto_get_flow_restore_wait()) {
1506 error = run_fast(ofproto_);
1511 if (ofproto->netflow) {
1512 if (netflow_run(ofproto->netflow)) {
1513 send_netflow_active_timeouts(ofproto);
1516 if (ofproto->sflow) {
1517 dpif_sflow_run(ofproto->sflow);
1519 if (ofproto->ipfix) {
1520 dpif_ipfix_run(ofproto->ipfix);
1523 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1526 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1531 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1532 if (mac_learning_run(ofproto->ml)) {
1533 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1535 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1537 /* Check the consistency of a random facet, to aid debugging. */
1538 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1539 if (time_msec() >= ofproto->consistency_rl
1540 && !classifier_is_empty(&ofproto->facets)
1541 && !ofproto->backer->need_revalidate) {
1542 struct cls_table *table;
1543 struct cls_rule *cr;
1544 struct facet *facet;
1546 ofproto->consistency_rl = time_msec() + 250;
1548 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1549 struct cls_table, hmap_node);
1550 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1552 facet = CONTAINER_OF(cr, struct facet, cr);
1554 if (!facet_check_consistency(facet)) {
1555 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1558 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1564 wait(struct ofproto *ofproto_)
1566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1567 struct ofport_dpif *ofport;
1568 struct ofbundle *bundle;
1570 if (ofproto_get_flow_restore_wait()) {
1574 if (ofproto->sflow) {
1575 dpif_sflow_wait(ofproto->sflow);
1577 if (ofproto->ipfix) {
1578 dpif_ipfix_wait(ofproto->ipfix);
1580 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1583 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1584 bundle_wait(bundle);
1586 if (ofproto->netflow) {
1587 netflow_wait(ofproto->netflow);
1589 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1590 mac_learning_wait(ofproto->ml);
1591 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1593 if (ofproto->backer->need_revalidate) {
1594 /* Shouldn't happen, but if it does just go around again. */
1595 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1596 poll_immediate_wake();
1601 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1603 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1604 struct cls_cursor cursor;
1605 size_t n_subfacets = 0;
1606 struct facet *facet;
1608 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1609 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1610 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1612 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1613 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1614 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1615 n_subfacets += list_size(&facet->subfacets);
1617 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1618 simap_increase(usage, "subfacets", n_subfacets);
1622 flush(struct ofproto *ofproto_)
1624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1625 struct subfacet *subfacet, *next_subfacet;
1626 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1630 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1631 &ofproto->backer->subfacets) {
1632 if (subfacet->facet->ofproto != ofproto) {
1636 if (subfacet->path != SF_NOT_INSTALLED) {
1637 batch[n_batch++] = subfacet;
1638 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1639 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1643 subfacet_destroy(subfacet);
1648 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1653 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1654 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1656 *arp_match_ip = true;
1657 *actions = (OFPUTIL_A_OUTPUT |
1658 OFPUTIL_A_SET_VLAN_VID |
1659 OFPUTIL_A_SET_VLAN_PCP |
1660 OFPUTIL_A_STRIP_VLAN |
1661 OFPUTIL_A_SET_DL_SRC |
1662 OFPUTIL_A_SET_DL_DST |
1663 OFPUTIL_A_SET_NW_SRC |
1664 OFPUTIL_A_SET_NW_DST |
1665 OFPUTIL_A_SET_NW_TOS |
1666 OFPUTIL_A_SET_TP_SRC |
1667 OFPUTIL_A_SET_TP_DST |
1672 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1674 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1675 struct dpif_dp_stats s;
1676 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1679 strcpy(ots->name, "classifier");
1681 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1682 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, true);
1683 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1685 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1688 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1689 ots->lookup_count = htonll(n_lookup);
1690 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1693 static struct ofport *
1696 struct ofport_dpif *port = xmalloc(sizeof *port);
1701 port_dealloc(struct ofport *port_)
1703 struct ofport_dpif *port = ofport_dpif_cast(port_);
1708 port_construct(struct ofport *port_)
1710 struct ofport_dpif *port = ofport_dpif_cast(port_);
1711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1712 const struct netdev *netdev = port->up.netdev;
1713 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1714 struct dpif_port dpif_port;
1717 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1718 port->bundle = NULL;
1721 port->may_enable = true;
1722 port->stp_port = NULL;
1723 port->stp_state = STP_DISABLED;
1724 port->is_tunnel = false;
1728 port->realdev_ofp_port = 0;
1729 port->vlandev_vid = 0;
1730 port->carrier_seq = netdev_get_carrier_resets(netdev);
1732 if (netdev_vport_is_patch(netdev)) {
1733 /* By bailing out here, we don't submit the port to the sFlow module
1734 * to be considered for counter polling export. This is correct
1735 * because the patch port represents an interface that sFlow considers
1736 * to be "internal" to the switch as a whole, and therefore not an
1737 * candidate for counter polling. */
1738 port->odp_port = ODPP_NONE;
1739 ofport_update_peer(port);
1743 error = dpif_port_query_by_name(ofproto->backer->dpif,
1744 netdev_vport_get_dpif_port(netdev, namebuf,
1751 port->odp_port = dpif_port.port_no;
1753 if (netdev_get_tunnel_config(netdev)) {
1754 tnl_port_add(port, port->up.netdev, port->odp_port);
1755 port->is_tunnel = true;
1757 /* Sanity-check that a mapping doesn't already exist. This
1758 * shouldn't happen for non-tunnel ports. */
1759 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1760 VLOG_ERR("port %s already has an OpenFlow port number",
1762 dpif_port_destroy(&dpif_port);
1766 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1767 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1768 hash_odp_port(port->odp_port));
1769 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1771 dpif_port_destroy(&dpif_port);
1773 if (ofproto->sflow) {
1774 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1781 port_destruct(struct ofport *port_)
1783 struct ofport_dpif *port = ofport_dpif_cast(port_);
1784 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1785 const char *devname = netdev_get_name(port->up.netdev);
1786 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1787 const char *dp_port_name;
1789 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1790 ovs_rwlock_wrlock(&xlate_rwlock);
1791 xlate_ofport_remove(port);
1792 ovs_rwlock_unlock(&xlate_rwlock);
1794 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1796 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1797 /* The underlying device is still there, so delete it. This
1798 * happens when the ofproto is being destroyed, since the caller
1799 * assumes that removal of attached ports will happen as part of
1801 if (!port->is_tunnel) {
1802 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1807 port->peer->peer = NULL;
1811 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1812 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1813 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1814 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1818 sset_find_and_delete(&ofproto->ports, devname);
1819 sset_find_and_delete(&ofproto->ghost_ports, devname);
1820 bundle_remove(port_);
1821 set_cfm(port_, NULL);
1822 set_bfd(port_, NULL);
1823 if (port->stp_port) {
1824 stp_port_disable(port->stp_port);
1826 if (ofproto->sflow) {
1827 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1834 port_modified(struct ofport *port_)
1836 struct ofport_dpif *port = ofport_dpif_cast(port_);
1838 if (port->bundle && port->bundle->bond) {
1839 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1843 cfm_set_netdev(port->cfm, port->up.netdev);
1847 bfd_set_netdev(port->bfd, port->up.netdev);
1850 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1852 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1856 ofport_update_peer(port);
1860 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1862 struct ofport_dpif *port = ofport_dpif_cast(port_);
1863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1864 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1866 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1867 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1868 OFPUTIL_PC_NO_PACKET_IN)) {
1869 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1871 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1872 bundle_update(port->bundle);
1878 set_sflow(struct ofproto *ofproto_,
1879 const struct ofproto_sflow_options *sflow_options)
1881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1882 struct dpif_sflow *ds = ofproto->sflow;
1884 if (sflow_options) {
1886 struct ofport_dpif *ofport;
1888 ds = ofproto->sflow = dpif_sflow_create();
1889 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1890 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1892 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1894 dpif_sflow_set_options(ds, sflow_options);
1897 dpif_sflow_unref(ds);
1898 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1899 ofproto->sflow = NULL;
1907 struct ofproto *ofproto_,
1908 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1909 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1910 size_t n_flow_exporters_options)
1912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1913 struct dpif_ipfix *di = ofproto->ipfix;
1914 bool has_options = bridge_exporter_options || flow_exporters_options;
1916 if (has_options && !di) {
1917 di = ofproto->ipfix = dpif_ipfix_create();
1921 /* Call set_options in any case to cleanly flush the flow
1922 * caches in the last exporters that are to be destroyed. */
1923 dpif_ipfix_set_options(
1924 di, bridge_exporter_options, flow_exporters_options,
1925 n_flow_exporters_options);
1928 dpif_ipfix_unref(di);
1929 ofproto->ipfix = NULL;
1937 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1939 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1946 struct ofproto_dpif *ofproto;
1948 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1949 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1950 ofport->cfm = cfm_create(ofport->up.netdev);
1953 if (cfm_configure(ofport->cfm, s)) {
1959 cfm_unref(ofport->cfm);
1965 get_cfm_status(const struct ofport *ofport_,
1966 struct ofproto_cfm_status *status)
1968 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1971 status->faults = cfm_get_fault(ofport->cfm);
1972 status->remote_opstate = cfm_get_opup(ofport->cfm);
1973 status->health = cfm_get_health(ofport->cfm);
1974 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1982 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1985 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1989 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1990 cfg, ofport->up.netdev);
1991 if (ofport->bfd != old) {
1992 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1999 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2001 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2004 bfd_get_status(ofport->bfd, smap);
2011 /* Spanning Tree. */
2014 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2016 struct ofproto_dpif *ofproto = ofproto_;
2017 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2018 struct ofport_dpif *ofport;
2020 ofport = stp_port_get_aux(sp);
2022 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2023 ofproto->up.name, port_num);
2025 struct eth_header *eth = pkt->l2;
2027 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2028 if (eth_addr_is_zero(eth->eth_src)) {
2029 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2030 "with unknown MAC", ofproto->up.name, port_num);
2032 send_packet(ofport, pkt);
2038 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2040 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2042 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2044 /* Only revalidate flows if the configuration changed. */
2045 if (!s != !ofproto->stp) {
2046 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2050 if (!ofproto->stp) {
2051 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2052 send_bpdu_cb, ofproto);
2053 ofproto->stp_last_tick = time_msec();
2056 stp_set_bridge_id(ofproto->stp, s->system_id);
2057 stp_set_bridge_priority(ofproto->stp, s->priority);
2058 stp_set_hello_time(ofproto->stp, s->hello_time);
2059 stp_set_max_age(ofproto->stp, s->max_age);
2060 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2062 struct ofport *ofport;
2064 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2065 set_stp_port(ofport, NULL);
2068 stp_unref(ofproto->stp);
2069 ofproto->stp = NULL;
2076 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2082 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2083 s->designated_root = stp_get_designated_root(ofproto->stp);
2084 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2093 update_stp_port_state(struct ofport_dpif *ofport)
2095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2096 enum stp_state state;
2098 /* Figure out new state. */
2099 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2103 if (ofport->stp_state != state) {
2104 enum ofputil_port_state of_state;
2107 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2108 netdev_get_name(ofport->up.netdev),
2109 stp_state_name(ofport->stp_state),
2110 stp_state_name(state));
2111 if (stp_learn_in_state(ofport->stp_state)
2112 != stp_learn_in_state(state)) {
2113 /* xxx Learning action flows should also be flushed. */
2114 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2115 mac_learning_flush(ofproto->ml);
2116 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2118 fwd_change = stp_forward_in_state(ofport->stp_state)
2119 != stp_forward_in_state(state);
2121 ofproto->backer->need_revalidate = REV_STP;
2122 ofport->stp_state = state;
2123 ofport->stp_state_entered = time_msec();
2125 if (fwd_change && ofport->bundle) {
2126 bundle_update(ofport->bundle);
2129 /* Update the STP state bits in the OpenFlow port description. */
2130 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2131 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2132 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2133 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2134 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2136 ofproto_port_set_state(&ofport->up, of_state);
2140 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2141 * caller is responsible for assigning STP port numbers and ensuring
2142 * there are no duplicates. */
2144 set_stp_port(struct ofport *ofport_,
2145 const struct ofproto_port_stp_settings *s)
2147 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2149 struct stp_port *sp = ofport->stp_port;
2151 if (!s || !s->enable) {
2153 ofport->stp_port = NULL;
2154 stp_port_disable(sp);
2155 update_stp_port_state(ofport);
2158 } else if (sp && stp_port_no(sp) != s->port_num
2159 && ofport == stp_port_get_aux(sp)) {
2160 /* The port-id changed, so disable the old one if it's not
2161 * already in use by another port. */
2162 stp_port_disable(sp);
2165 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2166 stp_port_enable(sp);
2168 stp_port_set_aux(sp, ofport);
2169 stp_port_set_priority(sp, s->priority);
2170 stp_port_set_path_cost(sp, s->path_cost);
2172 update_stp_port_state(ofport);
2178 get_stp_port_status(struct ofport *ofport_,
2179 struct ofproto_port_stp_status *s)
2181 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2183 struct stp_port *sp = ofport->stp_port;
2185 if (!ofproto->stp || !sp) {
2191 s->port_id = stp_port_get_id(sp);
2192 s->state = stp_port_get_state(sp);
2193 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2194 s->role = stp_port_get_role(sp);
2195 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2201 stp_run(struct ofproto_dpif *ofproto)
2204 long long int now = time_msec();
2205 long long int elapsed = now - ofproto->stp_last_tick;
2206 struct stp_port *sp;
2209 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2210 ofproto->stp_last_tick = now;
2212 while (stp_get_changed_port(ofproto->stp, &sp)) {
2213 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2216 update_stp_port_state(ofport);
2220 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2221 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2222 mac_learning_flush(ofproto->ml);
2223 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2229 stp_wait(struct ofproto_dpif *ofproto)
2232 poll_timer_wait(1000);
2237 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2240 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2243 if (ofport->n_qdscp != n_qdscp
2244 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2245 n_qdscp * sizeof *qdscp))) {
2246 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2247 free(ofport->qdscp);
2248 ofport->qdscp = n_qdscp
2249 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2251 ofport->n_qdscp = n_qdscp;
2259 /* Expires all MAC learning entries associated with 'bundle' and forces its
2260 * ofproto to revalidate every flow.
2262 * Normally MAC learning entries are removed only from the ofproto associated
2263 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2264 * are removed from every ofproto. When patch ports and SLB bonds are in use
2265 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2266 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2267 * with the host from which it migrated. */
2269 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2271 struct ofproto_dpif *ofproto = bundle->ofproto;
2272 struct mac_learning *ml = ofproto->ml;
2273 struct mac_entry *mac, *next_mac;
2275 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2276 ovs_rwlock_wrlock(&ml->rwlock);
2277 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2278 if (mac->port.p == bundle) {
2280 struct ofproto_dpif *o;
2282 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2284 struct mac_entry *e;
2286 ovs_rwlock_wrlock(&o->ml->rwlock);
2287 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2289 mac_learning_expire(o->ml, e);
2291 ovs_rwlock_unlock(&o->ml->rwlock);
2296 mac_learning_expire(ml, mac);
2299 ovs_rwlock_unlock(&ml->rwlock);
2302 static struct ofbundle *
2303 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2305 struct ofbundle *bundle;
2307 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2308 &ofproto->bundles) {
2309 if (bundle->aux == aux) {
2317 bundle_update(struct ofbundle *bundle)
2319 struct ofport_dpif *port;
2321 bundle->floodable = true;
2322 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2323 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2324 || !stp_forward_in_state(port->stp_state)) {
2325 bundle->floodable = false;
2332 bundle_del_port(struct ofport_dpif *port)
2334 struct ofbundle *bundle = port->bundle;
2336 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2338 list_remove(&port->bundle_node);
2339 port->bundle = NULL;
2342 lacp_slave_unregister(bundle->lacp, port);
2345 bond_slave_unregister(bundle->bond, port);
2348 bundle_update(bundle);
2352 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2353 struct lacp_slave_settings *lacp)
2355 struct ofport_dpif *port;
2357 port = get_ofp_port(bundle->ofproto, ofp_port);
2362 if (port->bundle != bundle) {
2363 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2365 bundle_remove(&port->up);
2368 port->bundle = bundle;
2369 list_push_back(&bundle->ports, &port->bundle_node);
2370 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2371 || !stp_forward_in_state(port->stp_state)) {
2372 bundle->floodable = false;
2376 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2377 lacp_slave_register(bundle->lacp, port, lacp);
2384 bundle_destroy(struct ofbundle *bundle)
2386 struct ofproto_dpif *ofproto;
2387 struct ofport_dpif *port, *next_port;
2393 ofproto = bundle->ofproto;
2394 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2396 ovs_rwlock_wrlock(&xlate_rwlock);
2397 xlate_bundle_remove(bundle);
2398 ovs_rwlock_unlock(&xlate_rwlock);
2400 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2401 bundle_del_port(port);
2404 bundle_flush_macs(bundle, true);
2405 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2407 free(bundle->trunks);
2408 lacp_unref(bundle->lacp);
2409 bond_unref(bundle->bond);
2414 bundle_set(struct ofproto *ofproto_, void *aux,
2415 const struct ofproto_bundle_settings *s)
2417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2418 bool need_flush = false;
2419 struct ofport_dpif *port;
2420 struct ofbundle *bundle;
2421 unsigned long *trunks;
2427 bundle_destroy(bundle_lookup(ofproto, aux));
2431 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2432 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2434 bundle = bundle_lookup(ofproto, aux);
2436 bundle = xmalloc(sizeof *bundle);
2438 bundle->ofproto = ofproto;
2439 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2440 hash_pointer(aux, 0));
2442 bundle->name = NULL;
2444 list_init(&bundle->ports);
2445 bundle->vlan_mode = PORT_VLAN_TRUNK;
2447 bundle->trunks = NULL;
2448 bundle->use_priority_tags = s->use_priority_tags;
2449 bundle->lacp = NULL;
2450 bundle->bond = NULL;
2452 bundle->floodable = true;
2453 mbridge_register_bundle(ofproto->mbridge, bundle);
2456 if (!bundle->name || strcmp(s->name, bundle->name)) {
2458 bundle->name = xstrdup(s->name);
2463 if (!bundle->lacp) {
2464 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2465 bundle->lacp = lacp_create();
2467 lacp_configure(bundle->lacp, s->lacp);
2469 lacp_unref(bundle->lacp);
2470 bundle->lacp = NULL;
2473 /* Update set of ports. */
2475 for (i = 0; i < s->n_slaves; i++) {
2476 if (!bundle_add_port(bundle, s->slaves[i],
2477 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2481 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2482 struct ofport_dpif *next_port;
2484 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2485 for (i = 0; i < s->n_slaves; i++) {
2486 if (s->slaves[i] == port->up.ofp_port) {
2491 bundle_del_port(port);
2495 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2497 if (list_is_empty(&bundle->ports)) {
2498 bundle_destroy(bundle);
2502 /* Set VLAN tagging mode */
2503 if (s->vlan_mode != bundle->vlan_mode
2504 || s->use_priority_tags != bundle->use_priority_tags) {
2505 bundle->vlan_mode = s->vlan_mode;
2506 bundle->use_priority_tags = s->use_priority_tags;
2511 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2512 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2514 if (vlan != bundle->vlan) {
2515 bundle->vlan = vlan;
2519 /* Get trunked VLANs. */
2520 switch (s->vlan_mode) {
2521 case PORT_VLAN_ACCESS:
2525 case PORT_VLAN_TRUNK:
2526 trunks = CONST_CAST(unsigned long *, s->trunks);
2529 case PORT_VLAN_NATIVE_UNTAGGED:
2530 case PORT_VLAN_NATIVE_TAGGED:
2531 if (vlan != 0 && (!s->trunks
2532 || !bitmap_is_set(s->trunks, vlan)
2533 || bitmap_is_set(s->trunks, 0))) {
2534 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2536 trunks = bitmap_clone(s->trunks, 4096);
2538 trunks = bitmap_allocate1(4096);
2540 bitmap_set1(trunks, vlan);
2541 bitmap_set0(trunks, 0);
2543 trunks = CONST_CAST(unsigned long *, s->trunks);
2550 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2551 free(bundle->trunks);
2552 if (trunks == s->trunks) {
2553 bundle->trunks = vlan_bitmap_clone(trunks);
2555 bundle->trunks = trunks;
2560 if (trunks != s->trunks) {
2565 if (!list_is_short(&bundle->ports)) {
2566 bundle->ofproto->has_bonded_bundles = true;
2568 if (bond_reconfigure(bundle->bond, s->bond)) {
2569 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2572 bundle->bond = bond_create(s->bond);
2573 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2576 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2577 bond_slave_register(bundle->bond, port, port->up.netdev);
2580 bond_unref(bundle->bond);
2581 bundle->bond = NULL;
2584 /* If we changed something that would affect MAC learning, un-learn
2585 * everything on this port and force flow revalidation. */
2587 bundle_flush_macs(bundle, false);
2594 bundle_remove(struct ofport *port_)
2596 struct ofport_dpif *port = ofport_dpif_cast(port_);
2597 struct ofbundle *bundle = port->bundle;
2600 bundle_del_port(port);
2601 if (list_is_empty(&bundle->ports)) {
2602 bundle_destroy(bundle);
2603 } else if (list_is_short(&bundle->ports)) {
2604 bond_unref(bundle->bond);
2605 bundle->bond = NULL;
2611 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2613 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2614 struct ofport_dpif *port = port_;
2615 uint8_t ea[ETH_ADDR_LEN];
2618 error = netdev_get_etheraddr(port->up.netdev, ea);
2620 struct ofpbuf packet;
2623 ofpbuf_init(&packet, 0);
2624 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2626 memcpy(packet_pdu, pdu, pdu_size);
2628 send_packet(port, &packet);
2629 ofpbuf_uninit(&packet);
2631 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2632 "%s (%s)", port->bundle->name,
2633 netdev_get_name(port->up.netdev), ovs_strerror(error));
2638 bundle_send_learning_packets(struct ofbundle *bundle)
2640 struct ofproto_dpif *ofproto = bundle->ofproto;
2641 struct ofpbuf *learning_packet;
2642 int error, n_packets, n_errors;
2643 struct mac_entry *e;
2644 struct list packets;
2646 list_init(&packets);
2647 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2648 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2649 if (e->port.p != bundle) {
2652 learning_packet = bond_compose_learning_packet(bundle->bond,
2655 learning_packet->private_p = port_void;
2656 list_push_back(&packets, &learning_packet->list_node);
2659 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2661 error = n_packets = n_errors = 0;
2662 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2665 ret = send_packet(learning_packet->private_p, learning_packet);
2672 ofpbuf_list_delete(&packets);
2675 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2676 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2677 "packets, last error was: %s",
2678 bundle->name, n_errors, n_packets, ovs_strerror(error));
2680 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2681 bundle->name, n_packets);
2686 bundle_run(struct ofbundle *bundle)
2689 lacp_run(bundle->lacp, send_pdu_cb);
2692 struct ofport_dpif *port;
2694 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2695 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2698 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2699 bundle->ofproto->backer->need_revalidate = REV_BOND;
2702 if (bond_should_send_learning_packets(bundle->bond)) {
2703 bundle_send_learning_packets(bundle);
2709 bundle_wait(struct ofbundle *bundle)
2712 lacp_wait(bundle->lacp);
2715 bond_wait(bundle->bond);
2722 mirror_set__(struct ofproto *ofproto_, void *aux,
2723 const struct ofproto_mirror_settings *s)
2725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2726 struct ofbundle **srcs, **dsts;
2731 mirror_destroy(ofproto->mbridge, aux);
2735 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2736 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2738 for (i = 0; i < s->n_srcs; i++) {
2739 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2742 for (i = 0; i < s->n_dsts; i++) {
2743 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2746 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2747 s->n_dsts, s->src_vlans,
2748 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2755 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2756 uint64_t *packets, uint64_t *bytes)
2759 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2764 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2767 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2768 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2769 mac_learning_flush(ofproto->ml);
2771 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2776 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2778 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2780 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2784 forward_bpdu_changed(struct ofproto *ofproto_)
2786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2787 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2791 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2795 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2796 mac_learning_set_idle_time(ofproto->ml, idle_time);
2797 mac_learning_set_max_entries(ofproto->ml, max_entries);
2798 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2803 static struct ofport_dpif *
2804 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2806 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2807 return ofport ? ofport_dpif_cast(ofport) : NULL;
2810 static struct ofport_dpif *
2811 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2813 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2814 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2818 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2819 struct ofproto_port *ofproto_port,
2820 struct dpif_port *dpif_port)
2822 ofproto_port->name = dpif_port->name;
2823 ofproto_port->type = dpif_port->type;
2824 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2828 ofport_update_peer(struct ofport_dpif *ofport)
2830 const struct ofproto_dpif *ofproto;
2831 struct dpif_backer *backer;
2834 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2838 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2839 backer->need_revalidate = REV_RECONFIGURE;
2842 ofport->peer->peer = NULL;
2843 ofport->peer = NULL;
2846 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2851 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2852 struct ofport *peer_ofport;
2853 struct ofport_dpif *peer;
2856 if (ofproto->backer != backer) {
2860 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2865 peer = ofport_dpif_cast(peer_ofport);
2866 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2867 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2869 ofport->peer = peer;
2870 ofport->peer->peer = ofport;
2880 port_run_fast(struct ofport_dpif *ofport)
2882 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2883 struct ofpbuf packet;
2885 ofpbuf_init(&packet, 0);
2886 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2887 send_packet(ofport, &packet);
2888 ofpbuf_uninit(&packet);
2891 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2892 struct ofpbuf packet;
2894 ofpbuf_init(&packet, 0);
2895 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2896 send_packet(ofport, &packet);
2897 ofpbuf_uninit(&packet);
2902 port_run(struct ofport_dpif *ofport)
2904 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2905 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2906 bool enable = netdev_get_carrier(ofport->up.netdev);
2907 bool cfm_enable = false;
2908 bool bfd_enable = false;
2910 ofport->carrier_seq = carrier_seq;
2912 port_run_fast(ofport);
2915 int cfm_opup = cfm_get_opup(ofport->cfm);
2917 cfm_run(ofport->cfm);
2918 cfm_enable = !cfm_get_fault(ofport->cfm);
2920 if (cfm_opup >= 0) {
2921 cfm_enable = cfm_enable && cfm_opup;
2926 bfd_run(ofport->bfd);
2927 bfd_enable = bfd_forwarding(ofport->bfd);
2930 if (ofport->bfd || ofport->cfm) {
2931 enable = enable && (cfm_enable || bfd_enable);
2934 if (ofport->bundle) {
2935 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2936 if (carrier_changed) {
2937 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2941 if (ofport->may_enable != enable) {
2942 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2943 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2946 ofport->may_enable = enable;
2950 port_wait(struct ofport_dpif *ofport)
2953 cfm_wait(ofport->cfm);
2957 bfd_wait(ofport->bfd);
2962 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2963 struct ofproto_port *ofproto_port)
2965 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2966 struct dpif_port dpif_port;
2969 if (sset_contains(&ofproto->ghost_ports, devname)) {
2970 const char *type = netdev_get_type_from_name(devname);
2972 /* We may be called before ofproto->up.port_by_name is populated with
2973 * the appropriate ofport. For this reason, we must get the name and
2974 * type from the netdev layer directly. */
2976 const struct ofport *ofport;
2978 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2979 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2980 ofproto_port->name = xstrdup(devname);
2981 ofproto_port->type = xstrdup(type);
2987 if (!sset_contains(&ofproto->ports, devname)) {
2990 error = dpif_port_query_by_name(ofproto->backer->dpif,
2991 devname, &dpif_port);
2993 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2999 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3002 const char *devname = netdev_get_name(netdev);
3003 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3004 const char *dp_port_name;
3006 if (netdev_vport_is_patch(netdev)) {
3007 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3011 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3012 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3013 odp_port_t port_no = ODPP_NONE;
3016 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3020 if (netdev_get_tunnel_config(netdev)) {
3021 simap_put(&ofproto->backer->tnl_backers,
3022 dp_port_name, odp_to_u32(port_no));
3026 if (netdev_get_tunnel_config(netdev)) {
3027 sset_add(&ofproto->ghost_ports, devname);
3029 sset_add(&ofproto->ports, devname);
3035 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3037 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3038 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3045 sset_find_and_delete(&ofproto->ghost_ports,
3046 netdev_get_name(ofport->up.netdev));
3047 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3048 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3049 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3051 /* The caller is going to close ofport->up.netdev. If this is a
3052 * bonded port, then the bond is using that netdev, so remove it
3053 * from the bond. The client will need to reconfigure everything
3054 * after deleting ports, so then the slave will get re-added. */
3055 bundle_remove(&ofport->up);
3062 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3064 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3069 error = netdev_get_stats(ofport->up.netdev, stats);
3071 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3074 /* ofproto->stats.tx_packets represents packets that we created
3075 * internally and sent to some port (e.g. packets sent with
3076 * send_packet()). Account for them as if they had come from
3077 * OFPP_LOCAL and got forwarded. */
3079 if (stats->rx_packets != UINT64_MAX) {
3080 stats->rx_packets += ofproto->stats.tx_packets;
3083 if (stats->rx_bytes != UINT64_MAX) {
3084 stats->rx_bytes += ofproto->stats.tx_bytes;
3087 /* ofproto->stats.rx_packets represents packets that were received on
3088 * some port and we processed internally and dropped (e.g. STP).
3089 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3091 if (stats->tx_packets != UINT64_MAX) {
3092 stats->tx_packets += ofproto->stats.rx_packets;
3095 if (stats->tx_bytes != UINT64_MAX) {
3096 stats->tx_bytes += ofproto->stats.rx_bytes;
3103 struct port_dump_state {
3108 struct ofproto_port port;
3113 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3115 *statep = xzalloc(sizeof(struct port_dump_state));
3120 port_dump_next(const struct ofproto *ofproto_, void *state_,
3121 struct ofproto_port *port)
3123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3124 struct port_dump_state *state = state_;
3125 const struct sset *sset;
3126 struct sset_node *node;
3128 if (state->has_port) {
3129 ofproto_port_destroy(&state->port);
3130 state->has_port = false;
3132 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3133 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3136 error = port_query_by_name(ofproto_, node->name, &state->port);
3138 *port = state->port;
3139 state->has_port = true;
3141 } else if (error != ENODEV) {
3146 if (!state->ghost) {
3147 state->ghost = true;
3150 return port_dump_next(ofproto_, state_, port);
3157 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3159 struct port_dump_state *state = state_;
3161 if (state->has_port) {
3162 ofproto_port_destroy(&state->port);
3169 port_poll(const struct ofproto *ofproto_, char **devnamep)
3171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3173 if (ofproto->port_poll_errno) {
3174 int error = ofproto->port_poll_errno;
3175 ofproto->port_poll_errno = 0;
3179 if (sset_is_empty(&ofproto->port_poll_set)) {
3183 *devnamep = sset_pop(&ofproto->port_poll_set);
3188 port_poll_wait(const struct ofproto *ofproto_)
3190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3191 dpif_port_poll_wait(ofproto->backer->dpif);
3195 port_is_lacp_current(const struct ofport *ofport_)
3197 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3198 return (ofport->bundle && ofport->bundle->lacp
3199 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3203 /* Upcall handling. */
3205 struct flow_miss_op {
3206 struct dpif_op dpif_op;
3208 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3209 struct xlate_out xout;
3210 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3212 struct ofpbuf mask; /* Flow mask for "put" ops. */
3213 struct odputil_keybuf maskbuf;
3215 /* If this is a "put" op, then a pointer to the subfacet that should
3216 * be marked as uninstalled if the operation fails. */
3217 struct subfacet *subfacet;
3220 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3221 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3222 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3223 * return value of true). However, for short flows the cost of bookkeeping is
3224 * much higher than the benefits, so when the datapath holds a large number of
3225 * flows we impose some heuristics to decide which flows are likely to be worth
3228 flow_miss_should_make_facet(struct flow_miss *miss)
3230 struct dpif_backer *backer = miss->ofproto->backer;
3233 switch (flow_miss_model) {
3234 case OFPROTO_HANDLE_MISS_AUTO:
3236 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3238 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3242 if (!backer->governor) {
3245 n_subfacets = hmap_count(&backer->subfacets);
3246 if (n_subfacets * 2 <= flow_eviction_threshold) {
3250 backer->governor = governor_create();
3253 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3254 return governor_should_install_flow(backer->governor, hash,
3255 list_size(&miss->packets));
3258 /* Handles 'miss', which matches 'facet'. May add any required datapath
3259 * operations to 'ops', incrementing '*n_ops' for each new op.
3261 * All of the packets in 'miss' are considered to have arrived at time
3262 * 'miss->stats.used'. This is really important only for new facets: if we
3263 * just called time_msec() here, then the new subfacet or its packets could
3264 * look (occasionally) as though it was used some time after the facet was
3265 * used. That can make a one-packet flow look like it has a nonzero duration,
3266 * which looks odd in e.g. NetFlow statistics. */
3268 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3269 struct flow_miss_op *ops, size_t *n_ops)
3271 enum subfacet_path want_path;
3272 struct subfacet *subfacet;
3274 facet->packet_count += miss->stats.n_packets;
3275 facet->prev_packet_count += miss->stats.n_packets;
3276 facet->byte_count += miss->stats.n_bytes;
3277 facet->prev_byte_count += miss->stats.n_bytes;
3279 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3281 /* Don't install the flow if it's the result of the "userspace"
3282 * action for an already installed facet. This can occur when a
3283 * datapath flow with wildcards has a "userspace" action and flows
3284 * sent to userspace result in a different subfacet, which will then
3285 * be rejected as overlapping by the datapath. */
3286 if (miss->upcall_type == DPIF_UC_ACTION
3287 && !list_is_empty(&facet->subfacets)) {
3291 subfacet = subfacet_create(facet, miss);
3292 if (subfacet->path != want_path) {
3293 struct flow_miss_op *op = &ops[(*n_ops)++];
3294 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3296 subfacet->path = want_path;
3298 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3299 if (enable_megaflows) {
3300 mask_to_netlink_attr(&op->mask, facet->ofproto,
3301 &facet->xout.wc.masks, &miss->flow);
3304 op->xout_garbage = false;
3305 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3306 op->subfacet = subfacet;
3307 put->flags = DPIF_FP_CREATE;
3308 put->key = miss->key;
3309 put->key_len = miss->key_len;
3310 put->mask = op->mask.data;
3311 put->mask_len = op->mask.size;
3313 if (want_path == SF_FAST_PATH) {
3314 put->actions = facet->xout.odp_actions.data;
3315 put->actions_len = facet->xout.odp_actions.size;
3317 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3318 op->slow_stub, sizeof op->slow_stub,
3319 &put->actions, &put->actions_len);
3325 /* Handles flow miss 'miss'. May add any required datapath operations
3326 * to 'ops', incrementing '*n_ops' for each new op. */
3328 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3331 struct facet *facet;
3333 miss->ofproto->n_missed += list_size(&miss->packets);
3335 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3337 /* There does not exist a bijection between 'struct flow' and datapath
3338 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3339 * assumption used throughout the facet and subfacet handling code.
3340 * Since we have to handle these misses in userspace anyway, we simply
3341 * skip facet creation, avoiding the problem altogether. */
3342 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3343 || !flow_miss_should_make_facet(miss)) {
3347 facet = facet_create(miss);
3349 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3352 static struct drop_key *
3353 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3356 struct drop_key *drop_key;
3358 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3359 &backer->drop_keys) {
3360 if (drop_key->key_len == key_len
3361 && !memcmp(drop_key->key, key, key_len)) {
3369 drop_key_clear(struct dpif_backer *backer)
3371 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3372 struct drop_key *drop_key, *next;
3374 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3377 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3379 if (error && !VLOG_DROP_WARN(&rl)) {
3380 struct ds ds = DS_EMPTY_INITIALIZER;
3381 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3382 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3383 ovs_strerror(error), ds_cstr(&ds));
3387 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3388 drop_key_destroy(drop_key);
3391 udpif_drop_key_clear(backer->udpif);
3395 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3397 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3398 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3399 struct flow_miss *miss;
3402 /* Process each element in the to-do list, constructing the set of
3403 * operations to batch. */
3405 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3406 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3408 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3410 /* Execute batch. */
3411 for (i = 0; i < n_ops; i++) {
3412 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3414 dpif_operate(backer->dpif, dpif_ops, n_ops);
3416 for (i = 0; i < n_ops; i++) {
3417 if (dpif_ops[i]->error != 0
3418 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3419 && flow_miss_ops[i].subfacet) {
3420 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3422 COVERAGE_INC(subfacet_install_fail);
3424 /* Zero-out subfacet counters when installation failed, but
3425 * datapath reported hits. This should not happen and
3426 * indicates a bug, since if the datapath flow exists, we
3427 * should not be attempting to create a new subfacet. A
3428 * buggy datapath could trigger this, so just zero out the
3429 * counters and log an error. */
3430 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3431 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3432 "datapath reported hits");
3433 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3436 subfacet->path = SF_NOT_INSTALLED;
3442 handle_sflow_upcall(struct dpif_backer *backer,
3443 const struct dpif_upcall *upcall)
3445 struct ofproto_dpif *ofproto;
3446 union user_action_cookie cookie;
3448 odp_port_t odp_in_port;
3450 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3451 &flow, NULL, &ofproto, &odp_in_port)
3452 || !ofproto->sflow) {
3456 memset(&cookie, 0, sizeof cookie);
3457 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3458 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3459 odp_in_port, &cookie);
3463 handle_flow_sample_upcall(struct dpif_backer *backer,
3464 const struct dpif_upcall *upcall)
3466 struct ofproto_dpif *ofproto;
3467 union user_action_cookie cookie;
3470 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3471 &flow, NULL, &ofproto, NULL)
3472 || !ofproto->ipfix) {
3476 memset(&cookie, 0, sizeof cookie);
3477 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3479 /* The flow reflects exactly the contents of the packet. Sample
3480 * the packet using it. */
3481 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3482 cookie.flow_sample.collector_set_id,
3483 cookie.flow_sample.probability,
3484 cookie.flow_sample.obs_domain_id,
3485 cookie.flow_sample.obs_point_id);
3489 handle_ipfix_upcall(struct dpif_backer *backer,
3490 const struct dpif_upcall *upcall)
3492 struct ofproto_dpif *ofproto;
3495 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3496 &flow, NULL, &ofproto, NULL)
3497 || !ofproto->ipfix) {
3501 /* The flow reflects exactly the contents of the packet. Sample
3502 * the packet using it. */
3503 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3507 handle_upcalls(struct dpif_backer *backer)
3509 struct flow_miss_batch *fmb;
3512 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3513 struct upcall *upcall = upcall_next(backer->udpif);
3519 switch (upcall->type) {
3521 handle_sflow_upcall(backer, &upcall->dpif_upcall);
3524 case FLOW_SAMPLE_UPCALL:
3525 handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3529 handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3539 upcall_destroy(upcall);
3542 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3543 struct drop_key *drop_key = drop_key_next(backer->udpif);
3548 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3549 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3550 hash_bytes(drop_key->key, drop_key->key_len, 0));
3551 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3552 drop_key->key, drop_key->key_len,
3553 NULL, 0, NULL, 0, NULL);
3555 drop_key_destroy(drop_key);
3559 fmb = flow_miss_batch_next(backer->udpif);
3561 handle_flow_misses(backer, fmb);
3562 flow_miss_batch_destroy(fmb);
3566 /* Flow expiration. */
3568 static int subfacet_max_idle(const struct dpif_backer *);
3569 static void update_stats(struct dpif_backer *);
3570 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3571 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3573 /* This function is called periodically by run(). Its job is to collect
3574 * updates for the flows that have been installed into the datapath, most
3575 * importantly when they last were used, and then use that information to
3576 * expire flows that have not been used recently.
3578 * Returns the number of milliseconds after which it should be called again. */
3580 expire(struct dpif_backer *backer)
3582 struct ofproto_dpif *ofproto;
3586 /* Periodically clear out the drop keys in an effort to keep them
3587 * relatively few. */
3588 drop_key_clear(backer);
3590 /* Update stats for each flow in the backer. */
3591 update_stats(backer);
3593 n_subfacets = hmap_count(&backer->subfacets);
3595 struct subfacet *subfacet;
3596 long long int total, now;
3600 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3601 total += now - subfacet->created;
3603 backer->avg_subfacet_life += total / n_subfacets;
3605 backer->avg_subfacet_life /= 2;
3607 backer->avg_n_subfacet += n_subfacets;
3608 backer->avg_n_subfacet /= 2;
3610 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3612 max_idle = subfacet_max_idle(backer);
3613 expire_subfacets(backer, max_idle);
3615 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3616 struct rule *rule, *next_rule;
3618 if (ofproto->backer != backer) {
3622 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3624 ovs_mutex_lock(&ofproto_mutex);
3625 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3626 &ofproto->up.expirable) {
3627 rule_expire(rule_dpif_cast(rule));
3629 ovs_mutex_unlock(&ofproto_mutex);
3631 /* All outstanding data in existing flows has been accounted, so it's a
3632 * good time to do bond rebalancing. */
3633 if (ofproto->has_bonded_bundles) {
3634 struct ofbundle *bundle;
3636 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3638 bond_rebalance(bundle->bond);
3644 return MIN(max_idle, 1000);
3647 /* Updates flow table statistics given that the datapath just reported 'stats'
3648 * as 'subfacet''s statistics. */
3650 update_subfacet_stats(struct subfacet *subfacet,
3651 const struct dpif_flow_stats *stats)
3653 struct facet *facet = subfacet->facet;
3654 struct dpif_flow_stats diff;
3656 diff.tcp_flags = stats->tcp_flags;
3657 diff.used = stats->used;
3659 if (stats->n_packets >= subfacet->dp_packet_count) {
3660 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3662 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3666 if (stats->n_bytes >= subfacet->dp_byte_count) {
3667 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3669 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3673 facet->ofproto->n_hit += diff.n_packets;
3674 subfacet->dp_packet_count = stats->n_packets;
3675 subfacet->dp_byte_count = stats->n_bytes;
3676 subfacet_update_stats(subfacet, &diff);
3678 if (facet->accounted_bytes < facet->byte_count) {
3680 facet_account(facet);
3681 facet->accounted_bytes = facet->byte_count;
3685 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3686 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3688 delete_unexpected_flow(struct dpif_backer *backer,
3689 const struct nlattr *key, size_t key_len)
3691 if (!VLOG_DROP_WARN(&rl)) {
3695 odp_flow_key_format(key, key_len, &s);
3696 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3700 COVERAGE_INC(facet_unexpected);
3701 dpif_flow_del(backer->dpif, key, key_len, NULL);
3704 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3706 * This function also pushes statistics updates to rules which each facet
3707 * resubmits into. Generally these statistics will be accurate. However, if a
3708 * facet changes the rule it resubmits into at some time in between
3709 * update_stats() runs, it is possible that statistics accrued to the
3710 * old rule will be incorrectly attributed to the new rule. This could be
3711 * avoided by calling update_stats() whenever rules are created or
3712 * deleted. However, the performance impact of making so many calls to the
3713 * datapath do not justify the benefit of having perfectly accurate statistics.
3715 * In addition, this function maintains per ofproto flow hit counts. The patch
3716 * port is not treated specially. e.g. A packet ingress from br0 patched into
3717 * br1 will increase the hit count of br0 by 1, however, does not affect
3718 * the hit or miss counts of br1.
3721 update_stats(struct dpif_backer *backer)
3723 const struct dpif_flow_stats *stats;
3724 struct dpif_flow_dump dump;
3725 const struct nlattr *key, *mask;
3726 size_t key_len, mask_len;
3728 dpif_flow_dump_start(&dump, backer->dpif);
3729 while (dpif_flow_dump_next(&dump, &key, &key_len,
3730 &mask, &mask_len, NULL, NULL, &stats)) {
3731 struct subfacet *subfacet;
3734 key_hash = odp_flow_key_hash(key, key_len);
3735 subfacet = subfacet_find(backer, key, key_len, key_hash);
3736 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3738 update_subfacet_stats(subfacet, stats);
3742 /* Stats are updated per-packet. */
3745 case SF_NOT_INSTALLED:
3747 delete_unexpected_flow(backer, key, key_len);
3752 dpif_flow_dump_done(&dump);
3754 update_moving_averages(backer);
3757 /* Calculates and returns the number of milliseconds of idle time after which
3758 * subfacets should expire from the datapath. When a subfacet expires, we fold
3759 * its statistics into its facet, and when a facet's last subfacet expires, we
3760 * fold its statistic into its rule. */
3762 subfacet_max_idle(const struct dpif_backer *backer)
3765 * Idle time histogram.
3767 * Most of the time a switch has a relatively small number of subfacets.
3768 * When this is the case we might as well keep statistics for all of them
3769 * in userspace and to cache them in the kernel datapath for performance as
3772 * As the number of subfacets increases, the memory required to maintain
3773 * statistics about them in userspace and in the kernel becomes
3774 * significant. However, with a large number of subfacets it is likely
3775 * that only a few of them are "heavy hitters" that consume a large amount
3776 * of bandwidth. At this point, only heavy hitters are worth caching in
3777 * the kernel and maintaining in userspaces; other subfacets we can
3780 * The technique used to compute the idle time is to build a histogram with
3781 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3782 * that is installed in the kernel gets dropped in the appropriate bucket.
3783 * After the histogram has been built, we compute the cutoff so that only
3784 * the most-recently-used 1% of subfacets (but at least
3785 * flow_eviction_threshold flows) are kept cached. At least
3786 * the most-recently-used bucket of subfacets is kept, so actually an
3787 * arbitrary number of subfacets can be kept in any given expiration run
3788 * (though the next run will delete most of those unless they receive
3791 * This requires a second pass through the subfacets, in addition to the
3792 * pass made by update_stats(), because the former function never looks at
3793 * uninstallable subfacets.
3795 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3796 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3797 int buckets[N_BUCKETS] = { 0 };
3798 int total, subtotal, bucket;
3799 struct subfacet *subfacet;
3803 total = hmap_count(&backer->subfacets);
3804 if (total <= flow_eviction_threshold) {
3805 return N_BUCKETS * BUCKET_WIDTH;
3808 /* Build histogram. */
3810 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3811 long long int idle = now - subfacet->used;
3812 int bucket = (idle <= 0 ? 0
3813 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3814 : (unsigned int) idle / BUCKET_WIDTH);
3818 /* Find the first bucket whose flows should be expired. */
3819 subtotal = bucket = 0;
3821 subtotal += buckets[bucket++];
3822 } while (bucket < N_BUCKETS &&
3823 subtotal < MAX(flow_eviction_threshold, total / 100));
3825 if (VLOG_IS_DBG_ENABLED()) {
3829 ds_put_cstr(&s, "keep");
3830 for (i = 0; i < N_BUCKETS; i++) {
3832 ds_put_cstr(&s, ", drop");
3835 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3838 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3842 return bucket * BUCKET_WIDTH;
3846 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3848 /* Cutoff time for most flows. */
3849 long long int normal_cutoff = time_msec() - dp_max_idle;
3851 /* We really want to keep flows for special protocols around, so use a more
3852 * conservative cutoff. */
3853 long long int special_cutoff = time_msec() - 10000;
3855 struct subfacet *subfacet, *next_subfacet;
3856 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3860 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3861 &backer->subfacets) {
3862 long long int cutoff;
3864 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3868 if (subfacet->used < cutoff) {
3869 if (subfacet->path != SF_NOT_INSTALLED) {
3870 batch[n_batch++] = subfacet;
3871 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3872 subfacet_destroy_batch(backer, batch, n_batch);
3876 subfacet_destroy(subfacet);
3882 subfacet_destroy_batch(backer, batch, n_batch);
3886 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3887 * then delete it entirely. */
3889 rule_expire(struct rule_dpif *rule)
3890 OVS_REQUIRES(ofproto_mutex)
3892 uint16_t idle_timeout, hard_timeout;
3893 long long int now = time_msec();
3896 ovs_assert(!rule->up.pending);
3898 /* Has 'rule' expired? */
3899 ovs_mutex_lock(&rule->up.mutex);
3900 hard_timeout = rule->up.hard_timeout;
3901 idle_timeout = rule->up.idle_timeout;
3902 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3903 reason = OFPRR_HARD_TIMEOUT;
3904 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3905 reason = OFPRR_IDLE_TIMEOUT;
3909 ovs_mutex_unlock(&rule->up.mutex);
3912 COVERAGE_INC(ofproto_dpif_expired);
3913 ofproto_rule_expire(&rule->up, reason);
3919 /* Creates and returns a new facet based on 'miss'.
3921 * The caller must already have determined that no facet with an identical
3922 * 'miss->flow' exists in 'miss->ofproto'.
3924 * 'rule' and 'xout' must have been created based on 'miss'.
3926 * 'facet'' statistics are initialized based on 'stats'.
3928 * The facet will initially have no subfacets. The caller should create (at
3929 * least) one subfacet with subfacet_create(). */
3930 static struct facet *
3931 facet_create(const struct flow_miss *miss)
3933 struct ofproto_dpif *ofproto = miss->ofproto;
3934 struct facet *facet;
3937 facet = xzalloc(sizeof *facet);
3938 facet->ofproto = miss->ofproto;
3939 facet->used = miss->stats.used;
3940 facet->flow = miss->flow;
3941 facet->learn_rl = time_msec() + 500;
3943 list_init(&facet->subfacets);
3944 netflow_flow_init(&facet->nf_flow);
3945 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3947 xlate_out_copy(&facet->xout, &miss->xout);
3949 match_init(&match, &facet->flow, &facet->xout.wc);
3950 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3951 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3952 classifier_insert(&ofproto->facets, &facet->cr);
3953 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3955 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3960 facet_free(struct facet *facet)
3963 xlate_out_uninit(&facet->xout);
3968 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3969 * 'packet', which arrived on 'in_port'. */
3971 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3972 const struct nlattr *odp_actions, size_t actions_len,
3973 struct ofpbuf *packet)
3975 struct odputil_keybuf keybuf;
3979 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3980 odp_flow_key_from_flow(&key, flow,
3981 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3983 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3984 odp_actions, actions_len, packet);
3988 /* Remove 'facet' from its ofproto and free up the associated memory:
3990 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3991 * rule's statistics, via subfacet_uninstall().
3993 * - Removes 'facet' from its rule and from ofproto->facets.
3996 facet_remove(struct facet *facet)
3998 struct subfacet *subfacet, *next_subfacet;
4000 ovs_assert(!list_is_empty(&facet->subfacets));
4002 /* First uninstall all of the subfacets to get final statistics. */
4003 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4004 subfacet_uninstall(subfacet);
4007 /* Flush the final stats to the rule.
4009 * This might require us to have at least one subfacet around so that we
4010 * can use its actions for accounting in facet_account(), which is why we
4011 * have uninstalled but not yet destroyed the subfacets. */
4012 facet_flush_stats(facet);
4014 /* Now we're really all done so destroy everything. */
4015 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4016 &facet->subfacets) {
4017 subfacet_destroy__(subfacet);
4019 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4020 classifier_remove(&facet->ofproto->facets, &facet->cr);
4021 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4022 cls_rule_destroy(&facet->cr);
4026 /* Feed information from 'facet' back into the learning table to keep it in
4027 * sync with what is actually flowing through the datapath. */
4029 facet_learn(struct facet *facet)
4031 long long int now = time_msec();
4033 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4037 facet->learn_rl = now + 500;
4039 if (!facet->xout.has_learn
4040 && !facet->xout.has_normal
4041 && (!facet->xout.has_fin_timeout
4042 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4046 facet_push_stats(facet, true);
4050 facet_account(struct facet *facet)
4052 const struct nlattr *a;
4057 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4060 n_bytes = facet->byte_count - facet->accounted_bytes;
4062 /* This loop feeds byte counters to bond_account() for rebalancing to use
4063 * as a basis. We also need to track the actual VLAN on which the packet
4064 * is going to be sent to ensure that it matches the one passed to
4065 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4068 * We use the actions from an arbitrary subfacet because they should all
4069 * be equally valid for our purpose. */
4070 vlan_tci = facet->flow.vlan_tci;
4071 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4072 facet->xout.odp_actions.size) {
4073 const struct ovs_action_push_vlan *vlan;
4074 struct ofport_dpif *port;
4076 switch (nl_attr_type(a)) {
4077 case OVS_ACTION_ATTR_OUTPUT:
4078 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4079 if (port && port->bundle && port->bundle->bond) {
4080 bond_account(port->bundle->bond, &facet->flow,
4081 vlan_tci_to_vid(vlan_tci), n_bytes);
4085 case OVS_ACTION_ATTR_POP_VLAN:
4086 vlan_tci = htons(0);
4089 case OVS_ACTION_ATTR_PUSH_VLAN:
4090 vlan = nl_attr_get(a);
4091 vlan_tci = vlan->vlan_tci;
4097 /* Returns true if the only action for 'facet' is to send to the controller.
4098 * (We don't report NetFlow expiration messages for such facets because they
4099 * are just part of the control logic for the network, not real traffic). */
4101 facet_is_controller_flow(struct facet *facet)
4104 struct ofproto_dpif *ofproto = facet->ofproto;
4105 const struct ofpact *ofpacts;
4106 struct rule_actions *actions;
4107 struct rule_dpif *rule;
4111 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4112 actions = rule_dpif_get_actions(rule);
4113 rule_dpif_unref(rule);
4115 ofpacts_len = actions->ofpacts_len;
4116 ofpacts = actions->ofpacts;
4117 is_controller = ofpacts_len > 0
4118 && ofpacts->type == OFPACT_CONTROLLER
4119 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4120 rule_actions_unref(actions);
4122 return is_controller;
4127 /* Folds all of 'facet''s statistics into its rule. Also updates the
4128 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4129 * 'facet''s statistics in the datapath should have been zeroed and folded into
4130 * its packet and byte counts before this function is called. */
4132 facet_flush_stats(struct facet *facet)
4134 struct ofproto_dpif *ofproto = facet->ofproto;
4135 struct subfacet *subfacet;
4137 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4138 ovs_assert(!subfacet->dp_byte_count);
4139 ovs_assert(!subfacet->dp_packet_count);
4142 facet_push_stats(facet, false);
4143 if (facet->accounted_bytes < facet->byte_count) {
4144 facet_account(facet);
4145 facet->accounted_bytes = facet->byte_count;
4148 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4149 struct ofexpired expired;
4150 expired.flow = facet->flow;
4151 expired.packet_count = facet->packet_count;
4152 expired.byte_count = facet->byte_count;
4153 expired.used = facet->used;
4154 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4157 /* Reset counters to prevent double counting if 'facet' ever gets
4159 facet_reset_counters(facet);
4161 netflow_flow_clear(&facet->nf_flow);
4162 facet->tcp_flags = 0;
4165 /* Searches 'ofproto''s table of facets for one which would be responsible for
4166 * 'flow'. Returns it if found, otherwise a null pointer.
4168 * The returned facet might need revalidation; use facet_lookup_valid()
4169 * instead if that is important. */
4170 static struct facet *
4171 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4173 struct cls_rule *cr;
4175 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4176 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4177 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4178 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4181 /* Searches 'ofproto''s table of facets for one capable that covers
4182 * 'flow'. Returns it if found, otherwise a null pointer.
4184 * The returned facet is guaranteed to be valid. */
4185 static struct facet *
4186 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4188 struct facet *facet;
4190 facet = facet_find(ofproto, flow);
4192 && ofproto->backer->need_revalidate
4193 && !facet_revalidate(facet)) {
4201 facet_check_consistency(struct facet *facet)
4203 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4205 struct xlate_out xout;
4206 struct xlate_in xin;
4209 /* Check the datapath actions for consistency. */
4210 xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4211 xlate_actions(&xin, &xout);
4213 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4214 && facet->xout.slow == xout.slow;
4215 if (!ok && !VLOG_DROP_WARN(&rl)) {
4216 struct ds s = DS_EMPTY_INITIALIZER;
4218 flow_format(&s, &facet->flow);
4219 ds_put_cstr(&s, ": inconsistency in facet");
4221 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4222 ds_put_cstr(&s, " (actions were: ");
4223 format_odp_actions(&s, facet->xout.odp_actions.data,
4224 facet->xout.odp_actions.size);
4225 ds_put_cstr(&s, ") (correct actions: ");
4226 format_odp_actions(&s, xout.odp_actions.data,
4227 xout.odp_actions.size);
4228 ds_put_char(&s, ')');
4231 if (facet->xout.slow != xout.slow) {
4232 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4237 xlate_out_uninit(&xout);
4242 /* Re-searches the classifier for 'facet':
4244 * - If the rule found is different from 'facet''s current rule, moves
4245 * 'facet' to the new rule and recompiles its actions.
4247 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4248 * where it is and recompiles its actions anyway.
4250 * - If any of 'facet''s subfacets correspond to a new flow according to
4251 * xlate_receive(), 'facet' is removed.
4253 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4255 facet_revalidate(struct facet *facet)
4257 struct ofproto_dpif *ofproto = facet->ofproto;
4258 struct rule_dpif *new_rule;
4259 struct subfacet *subfacet;
4260 struct flow_wildcards wc;
4261 struct xlate_out xout;
4262 struct xlate_in xin;
4264 COVERAGE_INC(facet_revalidate);
4266 /* Check that child subfacets still correspond to this facet. Tunnel
4267 * configuration changes could cause a subfacet's OpenFlow in_port to
4269 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4270 struct ofproto_dpif *recv_ofproto;
4271 struct flow recv_flow;
4274 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4275 subfacet->key_len, &recv_flow, NULL,
4276 &recv_ofproto, NULL);
4278 || recv_ofproto != ofproto
4279 || facet != facet_find(ofproto, &recv_flow)) {
4280 facet_remove(facet);
4285 flow_wildcards_init_catchall(&wc);
4286 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4288 /* Calculate new datapath actions.
4290 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4291 * emit a NetFlow expiration and, if so, we need to have the old state
4292 * around to properly compose it. */
4293 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4294 xlate_actions(&xin, &xout);
4295 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4296 /* Make sure non -packet fields are not masked. If not cleared,
4297 * the memcmp() below may fail, causing an otherwise valid facet
4299 flow_wildcards_clear_non_packet_fields(&xout.wc);
4301 /* A facet's slow path reason should only change under dramatic
4302 * circumstances. Rather than try to update everything, it's simpler to
4303 * remove the facet and start over.
4305 * More importantly, if a facet's wildcards change, it will be relatively
4306 * difficult to figure out if its subfacets still belong to it, and if not
4307 * which facet they may belong to. Again, to avoid the complexity, we
4308 * simply give up instead. */
4309 if (facet->xout.slow != xout.slow
4310 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4311 facet_remove(facet);
4312 xlate_out_uninit(&xout);
4313 rule_dpif_unref(new_rule);
4317 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4318 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4319 if (subfacet->path == SF_FAST_PATH) {
4320 struct dpif_flow_stats stats;
4322 subfacet_install(subfacet, &xout.odp_actions, &stats);
4323 subfacet_update_stats(subfacet, &stats);
4327 facet_flush_stats(facet);
4329 ofpbuf_clear(&facet->xout.odp_actions);
4330 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4331 xout.odp_actions.size);
4334 /* Update 'facet' now that we've taken care of all the old state. */
4335 facet->xout.slow = xout.slow;
4336 facet->xout.has_learn = xout.has_learn;
4337 facet->xout.has_normal = xout.has_normal;
4338 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4339 facet->xout.nf_output_iface = xout.nf_output_iface;
4340 facet->xout.mirrors = xout.mirrors;
4341 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4343 ovs_mutex_lock(&new_rule->up.mutex);
4344 facet->used = MAX(facet->used, new_rule->up.created);
4345 ovs_mutex_unlock(&new_rule->up.mutex);
4347 xlate_out_uninit(&xout);
4348 rule_dpif_unref(new_rule);
4353 facet_reset_counters(struct facet *facet)
4355 facet->packet_count = 0;
4356 facet->byte_count = 0;
4357 facet->prev_packet_count = 0;
4358 facet->prev_byte_count = 0;
4359 facet->accounted_bytes = 0;
4363 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4364 struct dpif_flow_stats *stats, bool may_learn)
4366 struct ofport_dpif *in_port;
4367 struct xlate_in xin;
4369 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4370 if (in_port && in_port->is_tunnel) {
4371 netdev_vport_inc_rx(in_port->up.netdev, stats);
4374 xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4375 xin.resubmit_stats = stats;
4376 xin.may_learn = may_learn;
4377 xlate_actions_for_side_effects(&xin);
4381 facet_push_stats(struct facet *facet, bool may_learn)
4383 struct dpif_flow_stats stats;
4385 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4386 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4387 ovs_assert(facet->used >= facet->prev_used);
4389 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4390 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4391 stats.used = facet->used;
4392 stats.tcp_flags = facet->tcp_flags;
4394 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4395 facet->prev_packet_count = facet->packet_count;
4396 facet->prev_byte_count = facet->byte_count;
4397 facet->prev_used = facet->used;
4399 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4401 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4402 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4403 stats.n_packets, stats.n_bytes);
4404 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4409 push_all_stats__(bool run_fast)
4411 static long long int rl = LLONG_MIN;
4412 struct ofproto_dpif *ofproto;
4414 if (time_msec() < rl) {
4418 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4419 struct cls_cursor cursor;
4420 struct facet *facet;
4422 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4423 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4424 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4425 facet_push_stats(facet, false);
4430 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4433 rl = time_msec() + 100;
4437 push_all_stats(void)
4439 push_all_stats__(true);
4443 rule_dpif_credit_stats(struct rule_dpif *rule,
4444 const struct dpif_flow_stats *stats)
4446 ovs_mutex_lock(&rule->stats_mutex);
4447 rule->packet_count += stats->n_packets;
4448 rule->byte_count += stats->n_bytes;
4449 rule->up.used = MAX(rule->up.used, stats->used);
4450 ovs_mutex_unlock(&rule->stats_mutex);
4454 rule_dpif_fail_open(const struct rule_dpif *rule)
4456 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4460 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4461 OVS_REQUIRES(rule->up.mutex)
4463 return rule->up.flow_cookie;
4467 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4468 uint16_t hard_timeout)
4470 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4473 /* Returns 'rule''s actions. The caller owns a reference on the returned
4474 * actions and must eventually release it (with rule_actions_unref()) to avoid
4476 struct rule_actions *
4477 rule_dpif_get_actions(const struct rule_dpif *rule)
4479 return rule_get_actions(&rule->up);
4484 static struct subfacet *
4485 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4486 size_t key_len, uint32_t key_hash)
4488 struct subfacet *subfacet;
4490 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4491 &backer->subfacets) {
4492 if (subfacet->key_len == key_len
4493 && !memcmp(key, subfacet->key, key_len)) {
4501 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4502 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4503 * existing subfacet if there is one, otherwise creates and returns a
4505 static struct subfacet *
4506 subfacet_create(struct facet *facet, struct flow_miss *miss)
4508 struct dpif_backer *backer = miss->ofproto->backer;
4509 enum odp_key_fitness key_fitness = miss->key_fitness;
4510 const struct nlattr *key = miss->key;
4511 size_t key_len = miss->key_len;
4513 struct subfacet *subfacet;
4515 key_hash = odp_flow_key_hash(key, key_len);
4517 if (list_is_empty(&facet->subfacets)) {
4518 subfacet = &facet->one_subfacet;
4520 subfacet = subfacet_find(backer, key, key_len, key_hash);
4522 if (subfacet->facet == facet) {
4526 /* This shouldn't happen. */
4527 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4528 subfacet_destroy(subfacet);
4531 subfacet = xmalloc(sizeof *subfacet);
4534 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4535 list_push_back(&facet->subfacets, &subfacet->list_node);
4536 subfacet->facet = facet;
4537 subfacet->key_fitness = key_fitness;
4538 subfacet->key = xmemdup(key, key_len);
4539 subfacet->key_len = key_len;
4540 subfacet->used = miss->stats.used;
4541 subfacet->created = subfacet->used;
4542 subfacet->dp_packet_count = 0;
4543 subfacet->dp_byte_count = 0;
4544 subfacet->path = SF_NOT_INSTALLED;
4545 subfacet->backer = backer;
4547 backer->subfacet_add_count++;
4551 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4552 * its facet within 'ofproto', and frees it. */
4554 subfacet_destroy__(struct subfacet *subfacet)
4556 struct facet *facet = subfacet->facet;
4557 struct ofproto_dpif *ofproto = facet->ofproto;
4559 /* Update ofproto stats before uninstall the subfacet. */
4560 ofproto->backer->subfacet_del_count++;
4562 subfacet_uninstall(subfacet);
4563 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4564 list_remove(&subfacet->list_node);
4565 free(subfacet->key);
4566 if (subfacet != &facet->one_subfacet) {
4571 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4572 * last remaining subfacet in its facet destroys the facet too. */
4574 subfacet_destroy(struct subfacet *subfacet)
4576 struct facet *facet = subfacet->facet;
4578 if (list_is_singleton(&facet->subfacets)) {
4579 /* facet_remove() needs at least one subfacet (it will remove it). */
4580 facet_remove(facet);
4582 subfacet_destroy__(subfacet);
4587 subfacet_destroy_batch(struct dpif_backer *backer,
4588 struct subfacet **subfacets, int n)
4590 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4591 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4592 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4595 for (i = 0; i < n; i++) {
4596 ops[i].type = DPIF_OP_FLOW_DEL;
4597 ops[i].u.flow_del.key = subfacets[i]->key;
4598 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4599 ops[i].u.flow_del.stats = &stats[i];
4603 dpif_operate(backer->dpif, opsp, n);
4604 for (i = 0; i < n; i++) {
4605 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4606 subfacets[i]->path = SF_NOT_INSTALLED;
4607 subfacet_destroy(subfacets[i]);
4612 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4613 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4614 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4615 * since 'subfacet' was last updated.
4617 * Returns 0 if successful, otherwise a positive errno value. */
4619 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4620 struct dpif_flow_stats *stats)
4622 struct facet *facet = subfacet->facet;
4623 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4624 const struct nlattr *actions = odp_actions->data;
4625 size_t actions_len = odp_actions->size;
4626 struct odputil_keybuf maskbuf;
4629 uint64_t slow_path_stub[128 / 8];
4630 enum dpif_flow_put_flags flags;
4633 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4636 flags |= DPIF_FP_ZERO_STATS;
4639 if (path == SF_SLOW_PATH) {
4640 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4641 slow_path_stub, sizeof slow_path_stub,
4642 &actions, &actions_len);
4645 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4646 if (enable_megaflows) {
4647 mask_to_netlink_attr(&mask, facet->ofproto,
4648 &facet->xout.wc.masks, &facet->flow);
4651 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4652 subfacet->key_len, mask.data, mask.size,
4653 actions, actions_len, stats);
4656 subfacet_reset_dp_stats(subfacet, stats);
4660 COVERAGE_INC(subfacet_install_fail);
4662 subfacet->path = path;
4667 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4669 subfacet_uninstall(struct subfacet *subfacet)
4671 if (subfacet->path != SF_NOT_INSTALLED) {
4672 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4673 struct dpif_flow_stats stats;
4676 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4677 subfacet->key_len, &stats);
4678 subfacet_reset_dp_stats(subfacet, &stats);
4680 subfacet_update_stats(subfacet, &stats);
4682 subfacet->path = SF_NOT_INSTALLED;
4684 ovs_assert(subfacet->dp_packet_count == 0);
4685 ovs_assert(subfacet->dp_byte_count == 0);
4689 /* Resets 'subfacet''s datapath statistics counters. This should be called
4690 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4691 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4692 * was reset in the datapath. 'stats' will be modified to include only
4693 * statistics new since 'subfacet' was last updated. */
4695 subfacet_reset_dp_stats(struct subfacet *subfacet,
4696 struct dpif_flow_stats *stats)
4699 && subfacet->dp_packet_count <= stats->n_packets
4700 && subfacet->dp_byte_count <= stats->n_bytes) {
4701 stats->n_packets -= subfacet->dp_packet_count;
4702 stats->n_bytes -= subfacet->dp_byte_count;
4705 subfacet->dp_packet_count = 0;
4706 subfacet->dp_byte_count = 0;
4709 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4711 * Because of the meaning of a subfacet's counters, it only makes sense to do
4712 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4713 * represents a packet that was sent by hand or if it represents statistics
4714 * that have been cleared out of the datapath. */
4716 subfacet_update_stats(struct subfacet *subfacet,
4717 const struct dpif_flow_stats *stats)
4719 if (stats->n_packets || stats->used > subfacet->used) {
4720 struct facet *facet = subfacet->facet;
4722 subfacet->used = MAX(subfacet->used, stats->used);
4723 facet->used = MAX(facet->used, stats->used);
4724 facet->packet_count += stats->n_packets;
4725 facet->byte_count += stats->n_bytes;
4726 facet->tcp_flags |= stats->tcp_flags;
4732 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4733 * the fields that were relevant as part of the lookup. */
4735 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4736 struct flow_wildcards *wc, struct rule_dpif **rule)
4738 struct ofport_dpif *port;
4740 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4743 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4745 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4746 flow->in_port.ofp_port);
4749 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4750 ofproto->no_packet_in_rule, rule);
4754 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4755 const struct flow *flow, struct flow_wildcards *wc,
4756 uint8_t table_id, struct rule_dpif **rule)
4758 const struct cls_rule *cls_rule;
4759 struct classifier *cls;
4763 if (table_id >= N_TABLES) {
4768 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4769 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4772 cls = &ofproto->up.tables[table_id].cls;
4773 ovs_rwlock_rdlock(&cls->rwlock);
4774 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4775 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4776 /* We must pretend that transport ports are unavailable. */
4777 struct flow ofpc_normal_flow = *flow;
4778 ofpc_normal_flow.tp_src = htons(0);
4779 ofpc_normal_flow.tp_dst = htons(0);
4780 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4781 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4782 cls_rule = &ofproto->drop_frags_rule->up.cr;
4784 flow_wildcards_init_exact(wc);
4787 cls_rule = classifier_lookup(cls, flow, wc);
4790 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4791 rule_dpif_ref(*rule);
4792 ovs_rwlock_unlock(&cls->rwlock);
4794 return *rule != NULL;
4797 /* Given a port configuration (specified as zero if there's no port), chooses
4798 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4799 * flow table miss. */
4801 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4802 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4804 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4805 rule_dpif_ref(*rule);
4809 rule_dpif_ref(struct rule_dpif *rule)
4812 ofproto_rule_ref(&rule->up);
4817 rule_dpif_unref(struct rule_dpif *rule)
4820 ofproto_rule_unref(&rule->up);
4825 complete_operation(struct rule_dpif *rule)
4826 OVS_REQUIRES(ofproto_mutex)
4828 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4830 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4831 ofoperation_complete(rule->up.pending, 0);
4834 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4836 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4839 static struct rule *
4842 struct rule_dpif *rule = xmalloc(sizeof *rule);
4847 rule_dealloc(struct rule *rule_)
4849 struct rule_dpif *rule = rule_dpif_cast(rule_);
4854 rule_construct(struct rule *rule_)
4856 struct rule_dpif *rule = rule_dpif_cast(rule_);
4857 ovs_mutex_init(&rule->stats_mutex);
4858 ovs_mutex_lock(&rule->stats_mutex);
4859 rule->packet_count = 0;
4860 rule->byte_count = 0;
4861 ovs_mutex_unlock(&rule->stats_mutex);
4866 rule_insert(struct rule *rule_)
4867 OVS_REQUIRES(ofproto_mutex)
4869 struct rule_dpif *rule = rule_dpif_cast(rule_);
4870 complete_operation(rule);
4874 rule_delete(struct rule *rule_)
4875 OVS_REQUIRES(ofproto_mutex)
4877 struct rule_dpif *rule = rule_dpif_cast(rule_);
4878 complete_operation(rule);
4882 rule_destruct(struct rule *rule_)
4884 struct rule_dpif *rule = rule_dpif_cast(rule_);
4885 ovs_mutex_destroy(&rule->stats_mutex);
4889 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4892 struct rule_dpif *rule = rule_dpif_cast(rule_);
4894 /* push_all_stats() can handle flow misses which, when using the learn
4895 * action, can cause rules to be added and deleted. This can corrupt our
4896 * caller's datastructures which assume that rule_get_stats() doesn't have
4897 * an impact on the flow table. To be safe, we disable miss handling. */
4899 push_all_stats__(false);
4902 /* Start from historical data for 'rule' itself that are no longer tracked
4903 * in facets. This counts, for example, facets that have expired. */
4904 ovs_mutex_lock(&rule->stats_mutex);
4905 *packets = rule->packet_count;
4906 *bytes = rule->byte_count;
4907 ovs_mutex_unlock(&rule->stats_mutex);
4911 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4912 struct ofpbuf *packet)
4914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4915 struct dpif_flow_stats stats;
4916 struct xlate_out xout;
4917 struct xlate_in xin;
4919 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4920 rule_dpif_credit_stats(rule, &stats);
4922 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4923 xin.resubmit_stats = &stats;
4924 xlate_actions(&xin, &xout);
4926 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4927 xout.odp_actions.size, packet);
4929 xlate_out_uninit(&xout);
4933 rule_execute(struct rule *rule, const struct flow *flow,
4934 struct ofpbuf *packet)
4936 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4937 ofpbuf_delete(packet);
4942 rule_modify_actions(struct rule *rule_, bool reset_counters)
4943 OVS_REQUIRES(ofproto_mutex)
4945 struct rule_dpif *rule = rule_dpif_cast(rule_);
4947 if (reset_counters) {
4948 ovs_mutex_lock(&rule->stats_mutex);
4949 rule->packet_count = 0;
4950 rule->byte_count = 0;
4951 ovs_mutex_unlock(&rule->stats_mutex);
4954 complete_operation(rule);
4957 /* Sends 'packet' out 'ofport'.
4958 * May modify 'packet'.
4959 * Returns 0 if successful, otherwise a positive errno value. */
4961 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4963 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4964 uint64_t odp_actions_stub[1024 / 8];
4965 struct ofpbuf key, odp_actions;
4966 struct dpif_flow_stats stats;
4967 struct odputil_keybuf keybuf;
4968 struct ofpact_output output;
4969 struct xlate_out xout;
4970 struct xlate_in xin;
4972 union flow_in_port in_port_;
4975 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4976 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4978 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4979 in_port_.ofp_port = OFPP_NONE;
4980 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4981 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4983 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4985 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4986 output.port = ofport->up.ofp_port;
4989 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4990 xin.ofpacts_len = sizeof output;
4991 xin.ofpacts = &output.ofpact;
4992 xin.resubmit_stats = &stats;
4993 xlate_actions(&xin, &xout);
4995 error = dpif_execute(ofproto->backer->dpif,
4997 xout.odp_actions.data, xout.odp_actions.size,
4999 xlate_out_uninit(&xout);
5002 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5003 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5004 ovs_strerror(error));
5007 ofproto->stats.tx_packets++;
5008 ofproto->stats.tx_bytes += packet->size;
5012 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5013 * The action will state 'slow' as the reason that the action is in the slow
5014 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5015 * dump-flows" output to see why a flow is in the slow path.)
5017 * The 'stub_size' bytes in 'stub' will be used to store the action.
5018 * 'stub_size' must be large enough for the action.
5020 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5023 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5024 enum slow_path_reason slow,
5025 uint64_t *stub, size_t stub_size,
5026 const struct nlattr **actionsp, size_t *actions_lenp)
5028 union user_action_cookie cookie;
5031 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5032 cookie.slow_path.unused = 0;
5033 cookie.slow_path.reason = slow;
5035 ofpbuf_use_stack(&buf, stub, stub_size);
5036 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5037 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5039 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5041 odp_port_t odp_port;
5044 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5045 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5046 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5048 *actionsp = buf.data;
5049 *actions_lenp = buf.size;
5053 set_frag_handling(struct ofproto *ofproto_,
5054 enum ofp_config_flags frag_handling)
5056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5057 if (frag_handling != OFPC_FRAG_REASM) {
5058 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5066 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5067 const struct flow *flow,
5068 const struct ofpact *ofpacts, size_t ofpacts_len)
5070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5071 struct odputil_keybuf keybuf;
5072 struct dpif_flow_stats stats;
5073 struct xlate_out xout;
5074 struct xlate_in xin;
5078 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5079 odp_flow_key_from_flow(&key, flow,
5080 ofp_port_to_odp_port(ofproto,
5081 flow->in_port.ofp_port));
5083 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5085 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5086 xin.resubmit_stats = &stats;
5087 xin.ofpacts_len = ofpacts_len;
5088 xin.ofpacts = ofpacts;
5090 xlate_actions(&xin, &xout);
5091 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5092 xout.odp_actions.data, xout.odp_actions.size, packet);
5093 xlate_out_uninit(&xout);
5101 set_netflow(struct ofproto *ofproto_,
5102 const struct netflow_options *netflow_options)
5104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5106 if (netflow_options) {
5107 if (!ofproto->netflow) {
5108 ofproto->netflow = netflow_create();
5109 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5111 return netflow_set_options(ofproto->netflow, netflow_options);
5112 } else if (ofproto->netflow) {
5113 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5114 netflow_destroy(ofproto->netflow);
5115 ofproto->netflow = NULL;
5122 get_netflow_ids(const struct ofproto *ofproto_,
5123 uint8_t *engine_type, uint8_t *engine_id)
5125 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5127 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5131 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5133 if (!facet_is_controller_flow(facet) &&
5134 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5135 struct subfacet *subfacet;
5136 struct ofexpired expired;
5138 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5139 if (subfacet->path == SF_FAST_PATH) {
5140 struct dpif_flow_stats stats;
5142 subfacet_install(subfacet, &facet->xout.odp_actions,
5144 subfacet_update_stats(subfacet, &stats);
5148 expired.flow = facet->flow;
5149 expired.packet_count = facet->packet_count;
5150 expired.byte_count = facet->byte_count;
5151 expired.used = facet->used;
5152 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5157 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5159 struct cls_cursor cursor;
5160 struct facet *facet;
5162 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5163 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5164 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5165 send_active_timeout(ofproto, facet);
5167 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5170 static struct ofproto_dpif *
5171 ofproto_dpif_lookup(const char *name)
5173 struct ofproto_dpif *ofproto;
5175 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5176 hash_string(name, 0), &all_ofproto_dpifs) {
5177 if (!strcmp(ofproto->up.name, name)) {
5185 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5186 const char *argv[], void *aux OVS_UNUSED)
5188 struct ofproto_dpif *ofproto;
5191 ofproto = ofproto_dpif_lookup(argv[1]);
5193 unixctl_command_reply_error(conn, "no such bridge");
5196 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5197 mac_learning_flush(ofproto->ml);
5198 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5200 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5201 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5202 mac_learning_flush(ofproto->ml);
5203 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5207 unixctl_command_reply(conn, "table successfully flushed");
5210 static struct ofport_dpif *
5211 ofbundle_get_a_port(const struct ofbundle *bundle)
5213 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5218 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5219 const char *argv[], void *aux OVS_UNUSED)
5221 struct ds ds = DS_EMPTY_INITIALIZER;
5222 const struct ofproto_dpif *ofproto;
5223 const struct mac_entry *e;
5225 ofproto = ofproto_dpif_lookup(argv[1]);
5227 unixctl_command_reply_error(conn, "no such bridge");
5231 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5232 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5233 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5234 struct ofbundle *bundle = e->port.p;
5235 char name[OFP_MAX_PORT_NAME_LEN];
5237 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5239 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5240 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5241 mac_entry_age(ofproto->ml, e));
5243 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5244 unixctl_command_reply(conn, ds_cstr(&ds));
5249 struct xlate_out xout;
5250 struct xlate_in xin;
5256 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5258 struct rule_actions *actions;
5261 ds_put_char_multiple(result, '\t', level);
5263 ds_put_cstr(result, "No match\n");
5267 ovs_mutex_lock(&rule->up.mutex);
5268 cookie = rule->up.flow_cookie;
5269 ovs_mutex_unlock(&rule->up.mutex);
5271 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5272 rule ? rule->up.table_id : 0, ntohll(cookie));
5273 cls_rule_format(&rule->up.cr, result);
5274 ds_put_char(result, '\n');
5276 actions = rule_dpif_get_actions(rule);
5278 ds_put_char_multiple(result, '\t', level);
5279 ds_put_cstr(result, "OpenFlow ");
5280 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5281 ds_put_char(result, '\n');
5283 rule_actions_unref(actions);
5287 trace_format_flow(struct ds *result, int level, const char *title,
5288 struct trace_ctx *trace)
5290 ds_put_char_multiple(result, '\t', level);
5291 ds_put_format(result, "%s: ", title);
5292 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5293 ds_put_cstr(result, "unchanged");
5295 flow_format(result, &trace->xin.flow);
5296 trace->flow = trace->xin.flow;
5298 ds_put_char(result, '\n');
5302 trace_format_regs(struct ds *result, int level, const char *title,
5303 struct trace_ctx *trace)
5307 ds_put_char_multiple(result, '\t', level);
5308 ds_put_format(result, "%s:", title);
5309 for (i = 0; i < FLOW_N_REGS; i++) {
5310 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5312 ds_put_char(result, '\n');
5316 trace_format_odp(struct ds *result, int level, const char *title,
5317 struct trace_ctx *trace)
5319 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5321 ds_put_char_multiple(result, '\t', level);
5322 ds_put_format(result, "%s: ", title);
5323 format_odp_actions(result, odp_actions->data, odp_actions->size);
5324 ds_put_char(result, '\n');
5328 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5330 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5331 struct ds *result = trace->result;
5333 ds_put_char(result, '\n');
5334 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5335 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5336 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5337 trace_format_rule(result, recurse + 1, rule);
5341 trace_report(struct xlate_in *xin, const char *s, int recurse)
5343 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5344 struct ds *result = trace->result;
5346 ds_put_char_multiple(result, '\t', recurse);
5347 ds_put_cstr(result, s);
5348 ds_put_char(result, '\n');
5352 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5353 void *aux OVS_UNUSED)
5355 const struct dpif_backer *backer;
5356 struct ofproto_dpif *ofproto;
5357 struct ofpbuf odp_key, odp_mask;
5358 struct ofpbuf *packet;
5366 ofpbuf_init(&odp_key, 0);
5367 ofpbuf_init(&odp_mask, 0);
5369 /* Handle "-generate" or a hex string as the last argument. */
5370 if (!strcmp(argv[argc - 1], "-generate")) {
5371 packet = ofpbuf_new(0);
5374 const char *error = eth_from_hex(argv[argc - 1], &packet);
5377 } else if (argc == 4) {
5378 /* The 3-argument form must end in "-generate' or a hex string. */
5379 unixctl_command_reply_error(conn, error);
5384 /* Parse the flow and determine whether a datapath or
5385 * bridge is specified. If function odp_flow_key_from_string()
5386 * returns 0, the flow is a odp_flow. If function
5387 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5388 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5389 /* If the odp_flow is the second argument,
5390 * the datapath name is the first argument. */
5392 const char *dp_type;
5393 if (!strncmp(argv[1], "ovs-", 4)) {
5394 dp_type = argv[1] + 4;
5398 backer = shash_find_data(&all_dpif_backers, dp_type);
5400 unixctl_command_reply_error(conn, "Cannot find datapath "
5405 /* No datapath name specified, so there should be only one
5407 struct shash_node *node;
5408 if (shash_count(&all_dpif_backers) != 1) {
5409 unixctl_command_reply_error(conn, "Must specify datapath "
5410 "name, there is more than one type of datapath");
5413 node = shash_first(&all_dpif_backers);
5414 backer = node->data;
5417 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5418 NULL, &ofproto, NULL)) {
5419 unixctl_command_reply_error(conn, "Invalid datapath flow");
5422 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5423 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5425 unixctl_command_reply_error(conn, "Must specify bridge name");
5429 ofproto = ofproto_dpif_lookup(argv[1]);
5431 unixctl_command_reply_error(conn, "Unknown bridge name");
5435 unixctl_command_reply_error(conn, "Bad flow syntax");
5439 /* Generate a packet, if requested. */
5441 if (!packet->size) {
5442 flow_compose(packet, &flow);
5444 union flow_in_port in_port_;
5446 in_port_ = flow.in_port;
5447 ds_put_cstr(&result, "Packet: ");
5448 s = ofp_packet_to_string(packet->data, packet->size);
5449 ds_put_cstr(&result, s);
5452 /* Use the metadata from the flow and the packet argument
5453 * to reconstruct the flow. */
5454 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5459 ofproto_trace(ofproto, &flow, packet, &result);
5460 unixctl_command_reply(conn, ds_cstr(&result));
5463 ds_destroy(&result);
5464 ofpbuf_delete(packet);
5465 ofpbuf_uninit(&odp_key);
5466 ofpbuf_uninit(&odp_mask);
5470 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5471 const struct ofpbuf *packet, struct ds *ds)
5473 struct rule_dpif *rule;
5474 struct flow_wildcards wc;
5476 ds_put_cstr(ds, "Flow: ");
5477 flow_format(ds, flow);
5478 ds_put_char(ds, '\n');
5480 flow_wildcards_init_catchall(&wc);
5481 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5483 trace_format_rule(ds, 0, rule);
5484 if (rule == ofproto->miss_rule) {
5485 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5486 } else if (rule == ofproto->no_packet_in_rule) {
5487 ds_put_cstr(ds, "\nNo match, packets dropped because "
5488 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5489 } else if (rule == ofproto->drop_frags_rule) {
5490 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5491 "and the fragment handling mode is \"drop\".\n");
5495 uint64_t odp_actions_stub[1024 / 8];
5496 struct ofpbuf odp_actions;
5497 struct trace_ctx trace;
5501 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5504 ofpbuf_use_stub(&odp_actions,
5505 odp_actions_stub, sizeof odp_actions_stub);
5506 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5507 trace.xin.resubmit_hook = trace_resubmit;
5508 trace.xin.report_hook = trace_report;
5510 xlate_actions(&trace.xin, &trace.xout);
5511 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5513 ds_put_char(ds, '\n');
5514 trace_format_flow(ds, 0, "Final flow", &trace);
5516 match_init(&match, flow, &trace.xout.wc);
5517 ds_put_cstr(ds, "Relevant fields: ");
5518 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5519 ds_put_char(ds, '\n');
5521 ds_put_cstr(ds, "Datapath actions: ");
5522 format_odp_actions(ds, trace.xout.odp_actions.data,
5523 trace.xout.odp_actions.size);
5525 if (trace.xout.slow) {
5526 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5527 "slow path because it:");
5528 switch (trace.xout.slow) {
5530 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5533 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5536 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5539 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5541 case SLOW_CONTROLLER:
5542 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5543 "to the OpenFlow controller.");
5550 xlate_out_uninit(&trace.xout);
5553 rule_dpif_unref(rule);
5556 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5557 * 'reply' describing the results. */
5559 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5561 struct cls_cursor cursor;
5562 struct facet *facet;
5566 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5567 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5568 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5569 if (!facet_check_consistency(facet)) {
5573 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5575 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5579 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5580 ofproto->up.name, errors);
5582 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5587 ofproto_dpif_self_check(struct unixctl_conn *conn,
5588 int argc, const char *argv[], void *aux OVS_UNUSED)
5590 struct ds reply = DS_EMPTY_INITIALIZER;
5591 struct ofproto_dpif *ofproto;
5594 ofproto = ofproto_dpif_lookup(argv[1]);
5596 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5597 "ofproto/list for help)");
5600 ofproto_dpif_self_check__(ofproto, &reply);
5602 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5603 ofproto_dpif_self_check__(ofproto, &reply);
5607 unixctl_command_reply(conn, ds_cstr(&reply));
5611 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5612 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5613 * to destroy 'ofproto_shash' and free the returned value. */
5614 static const struct shash_node **
5615 get_ofprotos(struct shash *ofproto_shash)
5617 const struct ofproto_dpif *ofproto;
5619 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5620 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5621 shash_add_nocopy(ofproto_shash, name, ofproto);
5624 return shash_sort(ofproto_shash);
5628 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5629 const char *argv[] OVS_UNUSED,
5630 void *aux OVS_UNUSED)
5632 struct ds ds = DS_EMPTY_INITIALIZER;
5633 struct shash ofproto_shash;
5634 const struct shash_node **sorted_ofprotos;
5637 shash_init(&ofproto_shash);
5638 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5639 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5640 const struct shash_node *node = sorted_ofprotos[i];
5641 ds_put_format(&ds, "%s\n", node->name);
5644 shash_destroy(&ofproto_shash);
5645 free(sorted_ofprotos);
5647 unixctl_command_reply(conn, ds_cstr(&ds));
5652 show_dp_rates(struct ds *ds, const char *heading,
5653 const struct avg_subfacet_rates *rates)
5655 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5656 heading, rates->add_rate, rates->del_rate);
5660 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5662 const struct shash_node **ofprotos;
5663 struct ofproto_dpif *ofproto;
5664 struct shash ofproto_shash;
5665 uint64_t n_hit, n_missed;
5666 long long int minutes;
5669 n_hit = n_missed = 0;
5670 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5671 if (ofproto->backer == backer) {
5672 n_missed += ofproto->n_missed;
5673 n_hit += ofproto->n_hit;
5677 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5678 dpif_name(backer->dpif), n_hit, n_missed);
5679 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5680 " life span: %lldms\n", hmap_count(&backer->subfacets),
5681 backer->avg_n_subfacet, backer->max_n_subfacet,
5682 backer->avg_subfacet_life);
5684 minutes = (time_msec() - backer->created) / (1000 * 60);
5685 if (minutes >= 60) {
5686 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5688 if (minutes >= 60 * 24) {
5689 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5691 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5693 shash_init(&ofproto_shash);
5694 ofprotos = get_ofprotos(&ofproto_shash);
5695 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5696 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5697 const struct shash_node **ports;
5700 if (ofproto->backer != backer) {
5704 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5705 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5707 ports = shash_sort(&ofproto->up.port_by_name);
5708 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5709 const struct shash_node *node = ports[j];
5710 struct ofport *ofport = node->data;
5712 odp_port_t odp_port;
5714 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5717 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5718 if (odp_port != ODPP_NONE) {
5719 ds_put_format(ds, "%"PRIu32":", odp_port);
5721 ds_put_cstr(ds, "none:");
5724 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5727 if (!netdev_get_config(ofport->netdev, &config)) {
5728 const struct smap_node **nodes;
5731 nodes = smap_sort(&config);
5732 for (i = 0; i < smap_count(&config); i++) {
5733 const struct smap_node *node = nodes[i];
5734 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5735 node->key, node->value);
5739 smap_destroy(&config);
5741 ds_put_char(ds, ')');
5742 ds_put_char(ds, '\n');
5746 shash_destroy(&ofproto_shash);
5751 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5752 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5754 struct ds ds = DS_EMPTY_INITIALIZER;
5755 const struct shash_node **backers;
5758 backers = shash_sort(&all_dpif_backers);
5759 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5760 dpif_show_backer(backers[i]->data, &ds);
5764 unixctl_command_reply(conn, ds_cstr(&ds));
5768 /* Dump the megaflow (facet) cache. This is useful to check the
5769 * correctness of flow wildcarding, since the same mechanism is used for
5770 * both xlate caching and kernel wildcarding.
5772 * It's important to note that in the output the flow description uses
5773 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5775 * This command is only needed for advanced debugging, so it's not
5776 * documented in the man page. */
5778 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5779 int argc OVS_UNUSED, const char *argv[],
5780 void *aux OVS_UNUSED)
5782 struct ds ds = DS_EMPTY_INITIALIZER;
5783 const struct ofproto_dpif *ofproto;
5784 long long int now = time_msec();
5785 struct cls_cursor cursor;
5786 struct facet *facet;
5788 ofproto = ofproto_dpif_lookup(argv[1]);
5790 unixctl_command_reply_error(conn, "no such bridge");
5794 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5795 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5796 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5797 cls_rule_format(&facet->cr, &ds);
5798 ds_put_cstr(&ds, ", ");
5799 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5800 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5801 ds_put_cstr(&ds, "Datapath actions: ");
5802 if (facet->xout.slow) {
5803 uint64_t slow_path_stub[128 / 8];
5804 const struct nlattr *actions;
5807 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5808 slow_path_stub, sizeof slow_path_stub,
5809 &actions, &actions_len);
5810 format_odp_actions(&ds, actions, actions_len);
5812 format_odp_actions(&ds, facet->xout.odp_actions.data,
5813 facet->xout.odp_actions.size);
5815 ds_put_cstr(&ds, "\n");
5817 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5819 ds_chomp(&ds, '\n');
5820 unixctl_command_reply(conn, ds_cstr(&ds));
5824 /* Disable using the megaflows.
5826 * This command is only needed for advanced debugging, so it's not
5827 * documented in the man page. */
5829 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5830 int argc OVS_UNUSED,
5831 const char *argv[] OVS_UNUSED,
5832 void *aux OVS_UNUSED)
5834 struct ofproto_dpif *ofproto;
5836 enable_megaflows = false;
5838 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5839 flush(&ofproto->up);
5842 unixctl_command_reply(conn, "megaflows disabled");
5845 /* Re-enable using megaflows.
5847 * This command is only needed for advanced debugging, so it's not
5848 * documented in the man page. */
5850 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5851 int argc OVS_UNUSED,
5852 const char *argv[] OVS_UNUSED,
5853 void *aux OVS_UNUSED)
5855 struct ofproto_dpif *ofproto;
5857 enable_megaflows = true;
5859 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5860 flush(&ofproto->up);
5863 unixctl_command_reply(conn, "megaflows enabled");
5867 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5868 int argc OVS_UNUSED, const char *argv[],
5869 void *aux OVS_UNUSED)
5871 struct ds ds = DS_EMPTY_INITIALIZER;
5872 const struct ofproto_dpif *ofproto;
5873 struct subfacet *subfacet;
5875 ofproto = ofproto_dpif_lookup(argv[1]);
5877 unixctl_command_reply_error(conn, "no such bridge");
5881 update_stats(ofproto->backer);
5883 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5884 struct facet *facet = subfacet->facet;
5885 struct odputil_keybuf maskbuf;
5888 if (facet->ofproto != ofproto) {
5892 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5893 if (enable_megaflows) {
5894 mask_to_netlink_attr(&mask, facet->ofproto,
5895 &facet->xout.wc.masks, &facet->flow);
5898 odp_flow_format(subfacet->key, subfacet->key_len,
5899 mask.data, mask.size, &ds, false);
5901 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5902 subfacet->dp_packet_count, subfacet->dp_byte_count);
5903 if (subfacet->used) {
5904 ds_put_format(&ds, "%.3fs",
5905 (time_msec() - subfacet->used) / 1000.0);
5907 ds_put_format(&ds, "never");
5909 if (subfacet->facet->tcp_flags) {
5910 ds_put_cstr(&ds, ", flags:");
5911 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5914 ds_put_cstr(&ds, ", actions:");
5915 if (facet->xout.slow) {
5916 uint64_t slow_path_stub[128 / 8];
5917 const struct nlattr *actions;
5920 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5921 slow_path_stub, sizeof slow_path_stub,
5922 &actions, &actions_len);
5923 format_odp_actions(&ds, actions, actions_len);
5925 format_odp_actions(&ds, facet->xout.odp_actions.data,
5926 facet->xout.odp_actions.size);
5928 ds_put_char(&ds, '\n');
5931 unixctl_command_reply(conn, ds_cstr(&ds));
5936 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5937 int argc OVS_UNUSED, const char *argv[],
5938 void *aux OVS_UNUSED)
5940 struct ds ds = DS_EMPTY_INITIALIZER;
5941 struct ofproto_dpif *ofproto;
5943 ofproto = ofproto_dpif_lookup(argv[1]);
5945 unixctl_command_reply_error(conn, "no such bridge");
5949 flush(&ofproto->up);
5951 unixctl_command_reply(conn, ds_cstr(&ds));
5956 ofproto_dpif_unixctl_init(void)
5958 static bool registered;
5964 unixctl_command_register(
5966 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5967 1, 3, ofproto_unixctl_trace, NULL);
5968 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5969 ofproto_unixctl_fdb_flush, NULL);
5970 unixctl_command_register("fdb/show", "bridge", 1, 1,
5971 ofproto_unixctl_fdb_show, NULL);
5972 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5973 ofproto_dpif_self_check, NULL);
5974 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5975 ofproto_unixctl_dpif_dump_dps, NULL);
5976 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5978 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5979 ofproto_unixctl_dpif_dump_flows, NULL);
5980 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5981 ofproto_unixctl_dpif_del_flows, NULL);
5982 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5983 ofproto_unixctl_dpif_dump_megaflows, NULL);
5984 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5985 ofproto_unixctl_dpif_disable_megaflows, NULL);
5986 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5987 ofproto_unixctl_dpif_enable_megaflows, NULL);
5990 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5992 * This is deprecated. It is only for compatibility with broken device drivers
5993 * in old versions of Linux that do not properly support VLANs when VLAN
5994 * devices are not used. When broken device drivers are no longer in
5995 * widespread use, we will delete these interfaces. */
5998 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6001 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6003 if (realdev_ofp_port == ofport->realdev_ofp_port
6004 && vid == ofport->vlandev_vid) {
6008 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6010 if (ofport->realdev_ofp_port) {
6013 if (realdev_ofp_port && ofport->bundle) {
6014 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6015 * themselves be part of a bundle. */
6016 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6019 ofport->realdev_ofp_port = realdev_ofp_port;
6020 ofport->vlandev_vid = vid;
6022 if (realdev_ofp_port) {
6023 vsp_add(ofport, realdev_ofp_port, vid);
6030 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6032 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6036 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6037 OVS_EXCLUDED(ofproto->vsp_mutex)
6041 ovs_mutex_lock(&ofproto->vsp_mutex);
6042 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6043 ovs_mutex_unlock(&ofproto->vsp_mutex);
6048 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6049 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6050 OVS_REQUIRES(ofproto->vsp_mutex)
6052 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6053 int vid = vlan_tci_to_vid(vlan_tci);
6054 const struct vlan_splinter *vsp;
6056 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6057 hash_realdev_vid(realdev_ofp_port, vid),
6058 &ofproto->realdev_vid_map) {
6059 if (vsp->realdev_ofp_port == realdev_ofp_port
6060 && vsp->vid == vid) {
6061 return vsp->vlandev_ofp_port;
6065 return realdev_ofp_port;
6068 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6069 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6070 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6071 * 'vlan_tci' 9, it would return the port number of eth0.9.
6073 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6074 * function just returns its 'realdev_ofp_port' argument. */
6076 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6077 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6078 OVS_EXCLUDED(ofproto->vsp_mutex)
6082 ovs_mutex_lock(&ofproto->vsp_mutex);
6083 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6084 ovs_mutex_unlock(&ofproto->vsp_mutex);
6088 static struct vlan_splinter *
6089 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6091 struct vlan_splinter *vsp;
6093 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6094 hash_ofp_port(vlandev_ofp_port),
6095 &ofproto->vlandev_map) {
6096 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6104 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6105 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6106 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6107 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6108 * eth0 and store 9 in '*vid'.
6110 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6111 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6114 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6115 ofp_port_t vlandev_ofp_port, int *vid)
6116 OVS_REQUIRES(ofproto->vsp_mutex)
6118 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6119 const struct vlan_splinter *vsp;
6121 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6126 return vsp->realdev_ofp_port;
6132 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6133 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6134 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6135 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6136 * always the case unless VLAN splinters are enabled), returns false without
6137 * making any changes. */
6139 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6140 OVS_EXCLUDED(ofproto->vsp_mutex)
6145 ovs_mutex_lock(&ofproto->vsp_mutex);
6146 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6147 ovs_mutex_unlock(&ofproto->vsp_mutex);
6152 /* Cause the flow to be processed as if it came in on the real device with
6153 * the VLAN device's VLAN ID. */
6154 flow->in_port.ofp_port = realdev;
6155 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6160 vsp_remove(struct ofport_dpif *port)
6162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6163 struct vlan_splinter *vsp;
6165 ovs_mutex_lock(&ofproto->vsp_mutex);
6166 vsp = vlandev_find(ofproto, port->up.ofp_port);
6168 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6169 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6172 port->realdev_ofp_port = 0;
6174 VLOG_ERR("missing vlan device record");
6176 ovs_mutex_unlock(&ofproto->vsp_mutex);
6180 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6184 ovs_mutex_lock(&ofproto->vsp_mutex);
6185 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6186 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6187 == realdev_ofp_port)) {
6188 struct vlan_splinter *vsp;
6190 vsp = xmalloc(sizeof *vsp);
6191 vsp->realdev_ofp_port = realdev_ofp_port;
6192 vsp->vlandev_ofp_port = port->up.ofp_port;
6195 port->realdev_ofp_port = realdev_ofp_port;
6197 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6198 hash_ofp_port(port->up.ofp_port));
6199 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6200 hash_realdev_vid(realdev_ofp_port, vid));
6202 VLOG_ERR("duplicate vlan device record");
6204 ovs_mutex_unlock(&ofproto->vsp_mutex);
6208 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6210 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6211 return ofport ? ofport->odp_port : ODPP_NONE;
6214 struct ofport_dpif *
6215 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6217 struct ofport_dpif *port;
6219 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6220 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6221 &backer->odp_to_ofport_map) {
6222 if (port->odp_port == odp_port) {
6223 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6228 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6233 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6235 struct ofport_dpif *port;
6237 port = odp_port_to_ofport(ofproto->backer, odp_port);
6238 if (port && &ofproto->up == port->up.ofproto) {
6239 return port->up.ofp_port;
6245 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6246 * most heavily weighted element. 'base' designates the rate of decay: after
6247 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6250 exp_mavg(double *avg, int base, double new)
6252 *avg = (*avg * (base - 1) + new) / base;
6256 update_moving_averages(struct dpif_backer *backer)
6258 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6259 long long int minutes = (time_msec() - backer->created) / min_ms;
6262 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6264 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6267 backer->lifetime.add_rate = 0.0;
6268 backer->lifetime.del_rate = 0.0;
6271 /* Update hourly averages on the minute boundaries. */
6272 if (time_msec() - backer->last_minute >= min_ms) {
6273 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6274 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6276 /* Update daily averages on the hour boundaries. */
6277 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6278 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6279 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6282 backer->total_subfacet_add_count += backer->subfacet_add_count;
6283 backer->total_subfacet_del_count += backer->subfacet_del_count;
6284 backer->subfacet_add_count = 0;
6285 backer->subfacet_del_count = 0;
6286 backer->last_minute += min_ms;
6290 /* Appends a representation of 'mask' as OVS_KEY_ATTR_* attributes to 'buf'.
6291 * Ommit vlan mask if the flow's vlan was added by VLAN splinter. */
6293 mask_to_netlink_attr(struct ofpbuf *buf, const struct ofproto_dpif *ofproto,
6294 const struct flow *mask, const struct flow *flow)
6296 ofp_port_t vlandev_port;
6297 struct flow orig_flow;
6299 vlandev_port = vsp_realdev_to_vlandev(ofproto, flow->in_port.ofp_port,
6302 if (vlandev_port != flow->in_port.ofp_port) {
6303 /* This flow was received from the VLAN port, do not
6304 * generate vlan masks. */
6306 orig_flow.vlan_tci = 0;
6310 odp_flow_key_from_mask(buf, mask, flow, UINT32_MAX);
6313 const struct ofproto_class ofproto_dpif_class = {
6348 port_is_lacp_current,
6349 NULL, /* rule_choose_table */
6358 rule_modify_actions,
6372 get_stp_port_status,
6379 is_mirror_output_bundle,
6380 forward_bpdu_changed,
6381 set_mac_table_config,
6383 NULL, /* meter_get_features */
6384 NULL, /* meter_set */
6385 NULL, /* meter_get */
6386 NULL, /* meter_del */