2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-mirror.h"
54 #include "ofproto-dpif-sflow.h"
55 #include "ofproto-dpif-xlate.h"
56 #include "poll-loop.h"
61 #include "unaligned.h"
63 #include "vlan-bitmap.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
68 COVERAGE_DEFINE(ofproto_dpif_expired);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(facet_unexpected);
72 COVERAGE_DEFINE(facet_suppress);
73 COVERAGE_DEFINE(subfacet_install_fail);
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
83 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
85 struct flow_wildcards *wc);
87 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
88 static void rule_invalidate(const struct rule_dpif *);
89 static tag_type rule_calculate_tag(const struct flow *,
90 const struct minimask *, uint32_t secret);
93 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
94 struct ofproto_dpif *ofproto; /* Owning ofproto. */
95 void *aux; /* Key supplied by ofproto's client. */
96 char *name; /* Identifier for log messages. */
99 struct list ports; /* Contains "struct ofport"s. */
100 enum port_vlan_mode vlan_mode; /* VLAN mode */
101 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
102 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
103 * NULL if all VLANs are trunked. */
104 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
105 struct bond *bond; /* Nonnull iff more than one port. */
106 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
109 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
112 static void bundle_remove(struct ofport *);
113 static void bundle_update(struct ofbundle *);
114 static void bundle_destroy(struct ofbundle *);
115 static void bundle_del_port(struct ofport_dpif *);
116 static void bundle_run(struct ofbundle *);
117 static void bundle_wait(struct ofbundle *);
119 static void stp_run(struct ofproto_dpif *ofproto);
120 static void stp_wait(struct ofproto_dpif *ofproto);
121 static int set_stp_port(struct ofport *,
122 const struct ofproto_port_stp_settings *);
124 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
125 enum slow_path_reason,
126 uint64_t *stub, size_t stub_size,
127 const struct nlattr **actionsp,
128 size_t *actions_lenp);
130 /* A subfacet (see "struct subfacet" below) has three possible installation
133 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
134 * case just after the subfacet is created, just before the subfacet is
135 * destroyed, or if the datapath returns an error when we try to install a
138 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
140 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
141 * ofproto_dpif is installed in the datapath.
144 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
145 SF_FAST_PATH, /* Full actions are installed. */
146 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
149 /* A dpif flow and actions associated with a facet.
151 * See also the large comment on struct facet. */
154 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
155 struct list list_node; /* In struct facet's 'facets' list. */
156 struct facet *facet; /* Owning facet. */
157 struct dpif_backer *backer; /* Owning backer. */
159 enum odp_key_fitness key_fitness;
163 long long int used; /* Time last used; time created if not used. */
164 long long int created; /* Time created. */
166 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
167 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
169 enum subfacet_path path; /* Installed in datapath? */
172 #define SUBFACET_DESTROY_MAX_BATCH 50
174 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
176 static struct subfacet *subfacet_find(struct dpif_backer *,
177 const struct nlattr *key, size_t key_len,
179 static void subfacet_destroy(struct subfacet *);
180 static void subfacet_destroy__(struct subfacet *);
181 static void subfacet_destroy_batch(struct dpif_backer *,
182 struct subfacet **, int n);
183 static void subfacet_reset_dp_stats(struct subfacet *,
184 struct dpif_flow_stats *);
185 static void subfacet_update_stats(struct subfacet *,
186 const struct dpif_flow_stats *);
187 static int subfacet_install(struct subfacet *,
188 const struct ofpbuf *odp_actions,
189 struct dpif_flow_stats *);
190 static void subfacet_uninstall(struct subfacet *);
192 /* A unique, non-overlapping instantiation of an OpenFlow flow.
194 * A facet associates a "struct flow", which represents the Open vSwitch
195 * userspace idea of an exact-match flow, with one or more subfacets.
196 * While the facet is created based on an exact-match flow, it is stored
197 * within the ofproto based on the wildcards that could be expressed
198 * based on the flow table and other configuration. (See the 'wc'
199 * description in "struct xlate_out" for more details.)
201 * Each subfacet tracks the datapath's idea of the flow equivalent to
202 * the facet. When the kernel module (or other dpif implementation) and
203 * Open vSwitch userspace agree on the definition of a flow key, there
204 * is exactly one subfacet per facet. If the dpif implementation
205 * supports more-specific flow matching than userspace, however, a facet
206 * can have more than one subfacet. Examples include the dpif
207 * implementation not supporting the same wildcards as userspace or some
208 * distinction in flow that userspace simply doesn't understand.
210 * Flow expiration works in terms of subfacets, so a facet must have at
211 * least one subfacet or it will never expire, leaking memory. */
214 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
215 struct ofproto_dpif *ofproto;
218 struct list subfacets;
219 long long int used; /* Time last used; time created if not used. */
222 struct flow flow; /* Flow of the creating subfacet. */
223 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
227 * - Do include packets and bytes sent "by hand", e.g. with
230 * - Do include packets and bytes that were obtained from the datapath
231 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
232 * DPIF_FP_ZERO_STATS).
234 * - Do not include packets or bytes that can be obtained from the
235 * datapath for any existing subfacet.
237 uint64_t packet_count; /* Number of packets received. */
238 uint64_t byte_count; /* Number of bytes received. */
240 /* Resubmit statistics. */
241 uint64_t prev_packet_count; /* Number of packets from last stats push. */
242 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
243 long long int prev_used; /* Used time from last stats push. */
246 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
247 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
248 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
250 struct xlate_out xout;
251 bool fail_open; /* Facet matched the fail open rule. */
253 /* Storage for a single subfacet, to reduce malloc() time and space
254 * overhead. (A facet always has at least one subfacet and in the common
255 * case has exactly one subfacet. However, 'one_subfacet' may not
256 * always be valid, since it could have been removed after newer
257 * subfacets were pushed onto the 'subfacets' list.) */
258 struct subfacet one_subfacet;
260 long long int learn_rl; /* Rate limiter for facet_learn(). */
263 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
265 struct dpif_flow_stats *);
266 static void facet_remove(struct facet *);
267 static void facet_free(struct facet *);
269 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
270 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
271 const struct flow *);
272 static bool facet_revalidate(struct facet *);
273 static bool facet_check_consistency(struct facet *);
275 static void facet_flush_stats(struct facet *);
277 static void facet_reset_counters(struct facet *);
278 static void facet_push_stats(struct facet *, bool may_learn);
279 static void facet_learn(struct facet *);
280 static void facet_account(struct facet *);
281 static void push_all_stats(void);
283 static bool facet_is_controller_flow(struct facet *);
286 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
290 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
291 struct list bundle_node; /* In struct ofbundle's "ports" list. */
292 struct cfm *cfm; /* Connectivity Fault Management, if any. */
293 struct bfd *bfd; /* BFD, if any. */
294 tag_type tag; /* Tag associated with this port. */
295 bool may_enable; /* May be enabled in bonds. */
296 bool is_tunnel; /* This port is a tunnel. */
297 long long int carrier_seq; /* Carrier status changes. */
298 struct ofport_dpif *peer; /* Peer if patch port. */
301 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
302 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
303 long long int stp_state_entered;
305 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
307 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
309 * This is deprecated. It is only for compatibility with broken device
310 * drivers in old versions of Linux that do not properly support VLANs when
311 * VLAN devices are not used. When broken device drivers are no longer in
312 * widespread use, we will delete these interfaces. */
313 ofp_port_t realdev_ofp_port;
317 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
318 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
319 * traffic egressing the 'ofport' with that priority should be marked with. */
320 struct priority_to_dscp {
321 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
322 uint32_t priority; /* Priority of this queue (see struct flow). */
324 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
327 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
329 * This is deprecated. It is only for compatibility with broken device drivers
330 * in old versions of Linux that do not properly support VLANs when VLAN
331 * devices are not used. When broken device drivers are no longer in
332 * widespread use, we will delete these interfaces. */
333 struct vlan_splinter {
334 struct hmap_node realdev_vid_node;
335 struct hmap_node vlandev_node;
336 ofp_port_t realdev_ofp_port;
337 ofp_port_t vlandev_ofp_port;
341 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
342 static void vsp_remove(struct ofport_dpif *);
343 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
345 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
348 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
351 static struct ofport_dpif *
352 ofport_dpif_cast(const struct ofport *ofport)
354 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
357 static void port_run(struct ofport_dpif *);
358 static void port_run_fast(struct ofport_dpif *);
359 static void port_wait(struct ofport_dpif *);
360 static int set_bfd(struct ofport *, const struct smap *);
361 static int set_cfm(struct ofport *, const struct cfm_settings *);
362 static void ofport_clear_priorities(struct ofport_dpif *);
363 static void ofport_update_peer(struct ofport_dpif *);
364 static void run_fast_rl(void);
366 struct dpif_completion {
367 struct list list_node;
368 struct ofoperation *op;
371 /* Extra information about a classifier table.
372 * Currently used just for optimized flow revalidation. */
374 /* If either of these is nonnull, then this table has a form that allows
375 * flows to be tagged to avoid revalidating most flows for the most common
376 * kinds of flow table changes. */
377 struct cls_table *catchall_table; /* Table that wildcards all fields. */
378 struct cls_table *other_table; /* Table with any other wildcard set. */
379 uint32_t basis; /* Keeps each table's tags separate. */
382 /* Reasons that we might need to revalidate every facet, and corresponding
385 * A value of 0 means that there is no need to revalidate.
387 * It would be nice to have some cleaner way to integrate with coverage
388 * counters, but with only a few reasons I guess this is good enough for
390 enum revalidate_reason {
391 REV_RECONFIGURE = 1, /* Switch configuration changed. */
392 REV_STP, /* Spanning tree protocol port status change. */
393 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
394 REV_FLOW_TABLE, /* Flow table changed. */
395 REV_MAC_LEARNING, /* Mac learning changed. */
396 REV_INCONSISTENCY /* Facet self-check failed. */
398 COVERAGE_DEFINE(rev_reconfigure);
399 COVERAGE_DEFINE(rev_stp);
400 COVERAGE_DEFINE(rev_port_toggled);
401 COVERAGE_DEFINE(rev_flow_table);
402 COVERAGE_DEFINE(rev_mac_learning);
403 COVERAGE_DEFINE(rev_inconsistency);
405 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
406 * These are datapath flows which have no associated ofproto, if they did we
407 * would use facets. */
409 struct hmap_node hmap_node;
414 struct avg_subfacet_rates {
415 double add_rate; /* Moving average of new flows created per minute. */
416 double del_rate; /* Moving average of flows deleted per minute. */
419 /* All datapaths of a given type share a single dpif backer instance. */
424 struct timer next_expiration;
425 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
427 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
429 /* Facet revalidation flags applying to facets which use this backer. */
430 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
431 struct tag_set revalidate_set; /* Revalidate only matching facets. */
433 struct hmap drop_keys; /* Set of dropped odp keys. */
434 bool recv_set_enable; /* Enables or disables receiving packets. */
436 struct hmap subfacets;
437 struct governor *governor;
439 /* Subfacet statistics.
441 * These keep track of the total number of subfacets added and deleted and
442 * flow life span. They are useful for computing the flow rates stats
443 * exposed via "ovs-appctl dpif/show". The goal is to learn about
444 * traffic patterns in ways that we can use later to improve Open vSwitch
445 * performance in new situations. */
446 long long int created; /* Time when it is created. */
447 unsigned max_n_subfacet; /* Maximum number of flows */
448 unsigned avg_n_subfacet; /* Average number of flows. */
449 long long int avg_subfacet_life; /* Average life span of subfacets. */
451 /* The average number of subfacets... */
452 struct avg_subfacet_rates hourly; /* ...over the last hour. */
453 struct avg_subfacet_rates daily; /* ...over the last day. */
454 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
455 long long int last_minute; /* Last time 'hourly' was updated. */
457 /* Number of subfacets added or deleted since 'last_minute'. */
458 unsigned subfacet_add_count;
459 unsigned subfacet_del_count;
461 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
462 unsigned long long int total_subfacet_add_count;
463 unsigned long long int total_subfacet_del_count;
466 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
467 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
469 static void drop_key_clear(struct dpif_backer *);
470 static struct ofport_dpif *
471 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
472 static void update_moving_averages(struct dpif_backer *backer);
474 struct ofproto_dpif {
475 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
477 struct dpif_backer *backer;
479 /* Special OpenFlow rules. */
480 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
481 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
482 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
485 struct netflow *netflow;
486 struct dpif_sflow *sflow;
487 struct dpif_ipfix *ipfix;
488 struct hmap bundles; /* Contains "struct ofbundle"s. */
489 struct mac_learning *ml;
490 bool has_bonded_bundles;
491 struct mbridge *mbridge;
494 struct classifier facets; /* Contains 'struct facet's. */
495 long long int consistency_rl;
498 struct table_dpif tables[N_TABLES];
500 /* Support for debugging async flow mods. */
501 struct list completions;
503 struct netdev_stats stats; /* To account packets generated and consumed in
508 long long int stp_last_tick;
510 /* VLAN splinters. */
511 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
512 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
515 struct sset ports; /* Set of standard port names. */
516 struct sset ghost_ports; /* Ports with no datapath port. */
517 struct sset port_poll_set; /* Queued names for port_poll() reply. */
518 int port_poll_errno; /* Last errno for port_poll() reply. */
520 /* Per ofproto's dpif stats. */
525 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
526 * for debugging the asynchronous flow_mod implementation.) */
529 /* By default, flows in the datapath are wildcarded (megaflows). They
530 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
531 static bool enable_megaflows = true;
533 /* All existing ofproto_dpif instances, indexed by ->up.name. */
534 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
536 static void ofproto_dpif_unixctl_init(void);
538 static inline struct ofproto_dpif *
539 ofproto_dpif_cast(const struct ofproto *ofproto)
541 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
542 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
545 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
546 ofp_port_t ofp_port);
549 #define FLOW_MISS_MAX_BATCH 50
550 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
552 /* Flow expiration. */
553 static int expire(struct dpif_backer *);
556 static void send_netflow_active_timeouts(struct ofproto_dpif *);
559 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
561 /* Global variables. */
562 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
564 /* Initial mappings of port to bridge mappings. */
565 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
568 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
569 struct ofputil_flow_mod *fm)
571 return ofproto_flow_mod(&ofproto->up, fm);
575 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
576 struct ofputil_packet_in *pin)
578 connmgr_send_packet_in(ofproto->up.connmgr, pin);
581 /* Factory functions. */
584 init(const struct shash *iface_hints)
586 struct shash_node *node;
588 /* Make a local copy, since we don't own 'iface_hints' elements. */
589 SHASH_FOR_EACH(node, iface_hints) {
590 const struct iface_hint *orig_hint = node->data;
591 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
593 new_hint->br_name = xstrdup(orig_hint->br_name);
594 new_hint->br_type = xstrdup(orig_hint->br_type);
595 new_hint->ofp_port = orig_hint->ofp_port;
597 shash_add(&init_ofp_ports, node->name, new_hint);
602 enumerate_types(struct sset *types)
604 dp_enumerate_types(types);
608 enumerate_names(const char *type, struct sset *names)
610 struct ofproto_dpif *ofproto;
613 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
614 if (strcmp(type, ofproto->up.type)) {
617 sset_add(names, ofproto->up.name);
624 del(const char *type, const char *name)
629 error = dpif_open(name, type, &dpif);
631 error = dpif_delete(dpif);
638 port_open_type(const char *datapath_type, const char *port_type)
640 return dpif_port_open_type(datapath_type, port_type);
643 /* Type functions. */
645 static void process_dpif_port_changes(struct dpif_backer *);
646 static void process_dpif_all_ports_changed(struct dpif_backer *);
647 static void process_dpif_port_change(struct dpif_backer *,
648 const char *devname);
649 static void process_dpif_port_error(struct dpif_backer *, int error);
651 static struct ofproto_dpif *
652 lookup_ofproto_dpif_by_port_name(const char *name)
654 struct ofproto_dpif *ofproto;
656 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
657 if (sset_contains(&ofproto->ports, name)) {
666 type_run(const char *type)
668 static long long int push_timer = LLONG_MIN;
669 struct dpif_backer *backer;
671 backer = shash_find_data(&all_dpif_backers, type);
673 /* This is not necessarily a problem, since backers are only
674 * created on demand. */
678 dpif_run(backer->dpif);
680 /* The most natural place to push facet statistics is when they're pulled
681 * from the datapath. However, when there are many flows in the datapath,
682 * this expensive operation can occur so frequently, that it reduces our
683 * ability to quickly set up flows. To reduce the cost, we push statistics
685 if (time_msec() > push_timer) {
686 push_timer = time_msec() + 2000;
690 /* If vswitchd started with other_config:flow_restore_wait set as "true",
691 * and the configuration has now changed to "false", enable receiving
692 * packets from the datapath. */
693 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
696 backer->recv_set_enable = true;
698 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
700 VLOG_ERR("Failed to enable receiving packets in dpif.");
703 dpif_flow_flush(backer->dpif);
704 backer->need_revalidate = REV_RECONFIGURE;
707 if (backer->need_revalidate
708 || !tag_set_is_empty(&backer->revalidate_set)) {
709 struct tag_set revalidate_set = backer->revalidate_set;
710 bool need_revalidate = backer->need_revalidate;
711 struct ofproto_dpif *ofproto;
712 struct simap_node *node;
713 struct simap tmp_backers;
715 /* Handle tunnel garbage collection. */
716 simap_init(&tmp_backers);
717 simap_swap(&backer->tnl_backers, &tmp_backers);
719 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
720 struct ofport_dpif *iter;
722 if (backer != ofproto->backer) {
726 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
727 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
730 if (!iter->is_tunnel) {
734 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
735 namebuf, sizeof namebuf);
736 node = simap_find(&tmp_backers, dp_port);
738 simap_put(&backer->tnl_backers, dp_port, node->data);
739 simap_delete(&tmp_backers, node);
740 node = simap_find(&backer->tnl_backers, dp_port);
742 node = simap_find(&backer->tnl_backers, dp_port);
744 odp_port_t odp_port = ODPP_NONE;
746 if (!dpif_port_add(backer->dpif, iter->up.netdev,
748 simap_put(&backer->tnl_backers, dp_port,
749 odp_to_u32(odp_port));
750 node = simap_find(&backer->tnl_backers, dp_port);
755 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
756 if (tnl_port_reconfigure(iter, iter->up.netdev,
758 backer->need_revalidate = REV_RECONFIGURE;
763 SIMAP_FOR_EACH (node, &tmp_backers) {
764 dpif_port_del(backer->dpif, u32_to_odp(node->data));
766 simap_destroy(&tmp_backers);
768 switch (backer->need_revalidate) {
769 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
770 case REV_STP: COVERAGE_INC(rev_stp); break;
771 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
772 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
773 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
774 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
777 if (backer->need_revalidate) {
778 /* Clear the drop_keys in case we should now be accepting some
779 * formerly dropped flows. */
780 drop_key_clear(backer);
783 /* Clear the revalidation flags. */
784 tag_set_init(&backer->revalidate_set);
785 backer->need_revalidate = 0;
787 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
788 struct facet *facet, *next;
789 struct cls_cursor cursor;
791 if (ofproto->backer != backer) {
795 if (need_revalidate) {
796 struct ofport_dpif *ofport;
797 struct ofbundle *bundle;
799 xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml,
800 ofproto->mbridge, ofproto->sflow,
801 ofproto->ipfix, ofproto->up.frag_handling,
802 ofproto->up.forward_bpdu,
803 connmgr_has_in_band(ofproto->up.connmgr),
804 ofproto->netflow != NULL,
805 ofproto->stp != NULL);
807 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
808 xlate_bundle_set(ofproto, bundle, bundle->name,
809 bundle->vlan_mode, bundle->vlan,
810 bundle->trunks, bundle->use_priority_tags,
811 bundle->bond, bundle->lacp,
815 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
816 xlate_ofport_set(ofproto, ofport->bundle, ofport,
817 ofport->up.ofp_port, ofport->odp_port,
818 ofport->up.netdev, ofport->cfm,
819 ofport->bfd, ofport->peer,
820 ofport->up.pp.config, ofport->stp_state,
821 ofport->is_tunnel, ofport->may_enable);
825 cls_cursor_init(&cursor, &ofproto->facets, NULL);
826 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
828 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
829 facet_revalidate(facet);
836 if (!backer->recv_set_enable) {
837 /* Wake up before a max of 1000ms. */
838 timer_set_duration(&backer->next_expiration, 1000);
839 } else if (timer_expired(&backer->next_expiration)) {
840 int delay = expire(backer);
841 timer_set_duration(&backer->next_expiration, delay);
844 process_dpif_port_changes(backer);
846 if (backer->governor) {
849 governor_run(backer->governor);
851 /* If the governor has shrunk to its minimum size and the number of
852 * subfacets has dwindled, then drop the governor entirely.
854 * For hysteresis, the number of subfacets to drop the governor is
855 * smaller than the number needed to trigger its creation. */
856 n_subfacets = hmap_count(&backer->subfacets);
857 if (n_subfacets * 4 < flow_eviction_threshold
858 && governor_is_idle(backer->governor)) {
859 governor_destroy(backer->governor);
860 backer->governor = NULL;
867 /* Check for and handle port changes in 'backer''s dpif. */
869 process_dpif_port_changes(struct dpif_backer *backer)
875 error = dpif_port_poll(backer->dpif, &devname);
881 process_dpif_all_ports_changed(backer);
885 process_dpif_port_change(backer, devname);
890 process_dpif_port_error(backer, error);
897 process_dpif_all_ports_changed(struct dpif_backer *backer)
899 struct ofproto_dpif *ofproto;
900 struct dpif_port dpif_port;
901 struct dpif_port_dump dump;
902 struct sset devnames;
905 sset_init(&devnames);
906 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
907 if (ofproto->backer == backer) {
908 struct ofport *ofport;
910 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
911 sset_add(&devnames, netdev_get_name(ofport->netdev));
915 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
916 sset_add(&devnames, dpif_port.name);
919 SSET_FOR_EACH (devname, &devnames) {
920 process_dpif_port_change(backer, devname);
922 sset_destroy(&devnames);
926 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
928 struct ofproto_dpif *ofproto;
929 struct dpif_port port;
931 /* Don't report on the datapath's device. */
932 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
936 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
937 &all_ofproto_dpifs) {
938 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
943 ofproto = lookup_ofproto_dpif_by_port_name(devname);
944 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
945 /* The port was removed. If we know the datapath,
946 * report it through poll_set(). If we don't, it may be
947 * notifying us of a removal we initiated, so ignore it.
948 * If there's a pending ENOBUFS, let it stand, since
949 * everything will be reevaluated. */
950 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
951 sset_add(&ofproto->port_poll_set, devname);
952 ofproto->port_poll_errno = 0;
954 } else if (!ofproto) {
955 /* The port was added, but we don't know with which
956 * ofproto we should associate it. Delete it. */
957 dpif_port_del(backer->dpif, port.port_no);
959 struct ofport_dpif *ofport;
961 ofport = ofport_dpif_cast(shash_find_data(
962 &ofproto->up.port_by_name, devname));
964 && ofport->odp_port != port.port_no
965 && !odp_port_to_ofport(backer, port.port_no))
967 /* 'ofport''s datapath port number has changed from
968 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
969 * structures to match. */
970 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
971 ofport->odp_port = port.port_no;
972 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
973 hash_odp_port(port.port_no));
974 backer->need_revalidate = REV_RECONFIGURE;
977 dpif_port_destroy(&port);
980 /* Propagate 'error' to all ofprotos based on 'backer'. */
982 process_dpif_port_error(struct dpif_backer *backer, int error)
984 struct ofproto_dpif *ofproto;
986 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
987 if (ofproto->backer == backer) {
988 sset_clear(&ofproto->port_poll_set);
989 ofproto->port_poll_errno = error;
995 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
999 /* If recv_set_enable is false, we should not handle upcalls. */
1000 if (!backer->recv_set_enable) {
1004 /* Handle one or more batches of upcalls, until there's nothing left to do
1005 * or until we do a fixed total amount of work.
1007 * We do work in batches because it can be much cheaper to set up a number
1008 * of flows and fire off their patches all at once. We do multiple batches
1009 * because in some cases handling a packet can cause another packet to be
1010 * queued almost immediately as part of the return flow. Both
1011 * optimizations can make major improvements on some benchmarks and
1012 * presumably for real traffic as well. */
1014 while (work < max_batch) {
1015 int retval = handle_upcalls(backer, max_batch - work);
1026 type_run_fast(const char *type)
1028 struct dpif_backer *backer;
1030 backer = shash_find_data(&all_dpif_backers, type);
1032 /* This is not necessarily a problem, since backers are only
1033 * created on demand. */
1037 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1043 static long long int port_rl = LLONG_MIN;
1044 static unsigned int backer_rl = 0;
1046 if (time_msec() >= port_rl) {
1047 struct ofproto_dpif *ofproto;
1048 struct ofport_dpif *ofport;
1050 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1052 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1053 port_run_fast(ofport);
1056 port_rl = time_msec() + 200;
1059 /* XXX: We have to be careful not to do too much work in this function. If
1060 * we call dpif_backer_run_fast() too often, or with too large a batch,
1061 * performance improves signifcantly, but at a cost. It's possible for the
1062 * number of flows in the datapath to increase without bound, and for poll
1063 * loops to take 10s of seconds. The correct solution to this problem,
1064 * long term, is to separate flow miss handling into it's own thread so it
1065 * isn't affected by revalidations, and expirations. Until then, this is
1066 * the best we can do. */
1067 if (++backer_rl >= 10) {
1068 struct shash_node *node;
1071 SHASH_FOR_EACH (node, &all_dpif_backers) {
1072 dpif_backer_run_fast(node->data, 1);
1078 type_wait(const char *type)
1080 struct dpif_backer *backer;
1082 backer = shash_find_data(&all_dpif_backers, type);
1084 /* This is not necessarily a problem, since backers are only
1085 * created on demand. */
1089 if (backer->governor) {
1090 governor_wait(backer->governor);
1093 timer_wait(&backer->next_expiration);
1096 /* Basic life-cycle. */
1098 static int add_internal_flows(struct ofproto_dpif *);
1100 static struct ofproto *
1103 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1104 return &ofproto->up;
1108 dealloc(struct ofproto *ofproto_)
1110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1115 close_dpif_backer(struct dpif_backer *backer)
1117 struct shash_node *node;
1119 ovs_assert(backer->refcount > 0);
1121 if (--backer->refcount) {
1125 drop_key_clear(backer);
1126 hmap_destroy(&backer->drop_keys);
1128 simap_destroy(&backer->tnl_backers);
1129 hmap_destroy(&backer->odp_to_ofport_map);
1130 node = shash_find(&all_dpif_backers, backer->type);
1132 shash_delete(&all_dpif_backers, node);
1133 dpif_close(backer->dpif);
1135 ovs_assert(hmap_is_empty(&backer->subfacets));
1136 hmap_destroy(&backer->subfacets);
1137 governor_destroy(backer->governor);
1142 /* Datapath port slated for removal from datapath. */
1143 struct odp_garbage {
1144 struct list list_node;
1145 odp_port_t odp_port;
1149 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1151 struct dpif_backer *backer;
1152 struct dpif_port_dump port_dump;
1153 struct dpif_port port;
1154 struct shash_node *node;
1155 struct list garbage_list;
1156 struct odp_garbage *garbage, *next;
1162 backer = shash_find_data(&all_dpif_backers, type);
1169 backer_name = xasprintf("ovs-%s", type);
1171 /* Remove any existing datapaths, since we assume we're the only
1172 * userspace controlling the datapath. */
1174 dp_enumerate_names(type, &names);
1175 SSET_FOR_EACH(name, &names) {
1176 struct dpif *old_dpif;
1178 /* Don't remove our backer if it exists. */
1179 if (!strcmp(name, backer_name)) {
1183 if (dpif_open(name, type, &old_dpif)) {
1184 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1186 dpif_delete(old_dpif);
1187 dpif_close(old_dpif);
1190 sset_destroy(&names);
1192 backer = xmalloc(sizeof *backer);
1194 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1197 VLOG_ERR("failed to open datapath of type %s: %s", type,
1198 ovs_strerror(error));
1203 backer->type = xstrdup(type);
1204 backer->governor = NULL;
1205 backer->refcount = 1;
1206 hmap_init(&backer->odp_to_ofport_map);
1207 hmap_init(&backer->drop_keys);
1208 hmap_init(&backer->subfacets);
1209 timer_set_duration(&backer->next_expiration, 1000);
1210 backer->need_revalidate = 0;
1211 simap_init(&backer->tnl_backers);
1212 tag_set_init(&backer->revalidate_set);
1213 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1216 if (backer->recv_set_enable) {
1217 dpif_flow_flush(backer->dpif);
1220 /* Loop through the ports already on the datapath and remove any
1221 * that we don't need anymore. */
1222 list_init(&garbage_list);
1223 dpif_port_dump_start(&port_dump, backer->dpif);
1224 while (dpif_port_dump_next(&port_dump, &port)) {
1225 node = shash_find(&init_ofp_ports, port.name);
1226 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1227 garbage = xmalloc(sizeof *garbage);
1228 garbage->odp_port = port.port_no;
1229 list_push_front(&garbage_list, &garbage->list_node);
1232 dpif_port_dump_done(&port_dump);
1234 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1235 dpif_port_del(backer->dpif, garbage->odp_port);
1236 list_remove(&garbage->list_node);
1240 shash_add(&all_dpif_backers, type, backer);
1242 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1244 VLOG_ERR("failed to listen on datapath of type %s: %s",
1245 type, ovs_strerror(error));
1246 close_dpif_backer(backer);
1250 backer->max_n_subfacet = 0;
1251 backer->created = time_msec();
1252 backer->last_minute = backer->created;
1253 memset(&backer->hourly, 0, sizeof backer->hourly);
1254 memset(&backer->daily, 0, sizeof backer->daily);
1255 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1256 backer->subfacet_add_count = 0;
1257 backer->subfacet_del_count = 0;
1258 backer->total_subfacet_add_count = 0;
1259 backer->total_subfacet_del_count = 0;
1260 backer->avg_n_subfacet = 0;
1261 backer->avg_subfacet_life = 0;
1267 construct(struct ofproto *ofproto_)
1269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1270 struct shash_node *node, *next;
1271 odp_port_t max_ports;
1275 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1280 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1281 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1282 ofp_to_u16(OFPP_MAX))));
1284 ofproto->netflow = NULL;
1285 ofproto->sflow = NULL;
1286 ofproto->ipfix = NULL;
1287 ofproto->stp = NULL;
1288 hmap_init(&ofproto->bundles);
1289 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1290 ofproto->mbridge = mbridge_create();
1291 ofproto->has_bonded_bundles = false;
1293 classifier_init(&ofproto->facets);
1294 ofproto->consistency_rl = LLONG_MIN;
1296 for (i = 0; i < N_TABLES; i++) {
1297 struct table_dpif *table = &ofproto->tables[i];
1299 table->catchall_table = NULL;
1300 table->other_table = NULL;
1301 table->basis = random_uint32();
1304 list_init(&ofproto->completions);
1306 ofproto_dpif_unixctl_init();
1308 hmap_init(&ofproto->vlandev_map);
1309 hmap_init(&ofproto->realdev_vid_map);
1311 sset_init(&ofproto->ports);
1312 sset_init(&ofproto->ghost_ports);
1313 sset_init(&ofproto->port_poll_set);
1314 ofproto->port_poll_errno = 0;
1316 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1317 struct iface_hint *iface_hint = node->data;
1319 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1320 /* Check if the datapath already has this port. */
1321 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1322 sset_add(&ofproto->ports, node->name);
1325 free(iface_hint->br_name);
1326 free(iface_hint->br_type);
1328 shash_delete(&init_ofp_ports, node);
1332 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1333 hash_string(ofproto->up.name, 0));
1334 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1336 ofproto_init_tables(ofproto_, N_TABLES);
1337 error = add_internal_flows(ofproto);
1338 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1341 ofproto->n_missed = 0;
1347 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1348 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1350 struct ofputil_flow_mod fm;
1353 match_init_catchall(&fm.match);
1355 match_set_reg(&fm.match, 0, id);
1356 fm.new_cookie = htonll(0);
1357 fm.cookie = htonll(0);
1358 fm.cookie_mask = htonll(0);
1359 fm.modify_cookie = false;
1360 fm.table_id = TBL_INTERNAL;
1361 fm.command = OFPFC_ADD;
1362 fm.idle_timeout = 0;
1363 fm.hard_timeout = 0;
1367 fm.ofpacts = ofpacts->data;
1368 fm.ofpacts_len = ofpacts->size;
1370 error = ofproto_flow_mod(&ofproto->up, &fm);
1372 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1373 id, ofperr_to_string(error));
1377 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1379 ovs_assert(*rulep != NULL);
1385 add_internal_flows(struct ofproto_dpif *ofproto)
1387 struct ofpact_controller *controller;
1388 uint64_t ofpacts_stub[128 / 8];
1389 struct ofpbuf ofpacts;
1393 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1396 controller = ofpact_put_CONTROLLER(&ofpacts);
1397 controller->max_len = UINT16_MAX;
1398 controller->controller_id = 0;
1399 controller->reason = OFPR_NO_MATCH;
1400 ofpact_pad(&ofpacts);
1402 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1407 ofpbuf_clear(&ofpacts);
1408 error = add_internal_flow(ofproto, id++, &ofpacts,
1409 &ofproto->no_packet_in_rule);
1414 error = add_internal_flow(ofproto, id++, &ofpacts,
1415 &ofproto->drop_frags_rule);
1420 complete_operations(struct ofproto_dpif *ofproto)
1422 struct dpif_completion *c, *next;
1424 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1425 ofoperation_complete(c->op, 0);
1426 list_remove(&c->list_node);
1432 destruct(struct ofproto *ofproto_)
1434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1435 struct rule_dpif *rule, *next_rule;
1436 struct oftable *table;
1438 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1439 xlate_remove_ofproto(ofproto);
1441 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1442 complete_operations(ofproto);
1444 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1445 struct cls_cursor cursor;
1447 cls_cursor_init(&cursor, &table->cls, NULL);
1448 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1449 ofproto_rule_destroy(&rule->up);
1453 mbridge_unref(ofproto->mbridge);
1455 netflow_destroy(ofproto->netflow);
1456 dpif_sflow_unref(ofproto->sflow);
1457 hmap_destroy(&ofproto->bundles);
1458 mac_learning_unref(ofproto->ml);
1460 classifier_destroy(&ofproto->facets);
1462 hmap_destroy(&ofproto->vlandev_map);
1463 hmap_destroy(&ofproto->realdev_vid_map);
1465 sset_destroy(&ofproto->ports);
1466 sset_destroy(&ofproto->ghost_ports);
1467 sset_destroy(&ofproto->port_poll_set);
1469 close_dpif_backer(ofproto->backer);
1473 run_fast(struct ofproto *ofproto_)
1475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1476 struct ofport_dpif *ofport;
1478 /* Do not perform any periodic activity required by 'ofproto' while
1479 * waiting for flow restore to complete. */
1480 if (ofproto_get_flow_restore_wait()) {
1484 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1485 port_run_fast(ofport);
1492 run(struct ofproto *ofproto_)
1494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1495 struct ofport_dpif *ofport;
1496 struct ofbundle *bundle;
1500 complete_operations(ofproto);
1503 if (mbridge_need_revalidate(ofproto->mbridge)) {
1504 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1505 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1506 mac_learning_flush(ofproto->ml);
1507 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1510 /* Do not perform any periodic activity below required by 'ofproto' while
1511 * waiting for flow restore to complete. */
1512 if (ofproto_get_flow_restore_wait()) {
1516 error = run_fast(ofproto_);
1521 if (ofproto->netflow) {
1522 if (netflow_run(ofproto->netflow)) {
1523 send_netflow_active_timeouts(ofproto);
1526 if (ofproto->sflow) {
1527 dpif_sflow_run(ofproto->sflow);
1530 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1533 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1538 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1539 if (mac_learning_run(ofproto->ml)) {
1540 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1542 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1544 /* Check the consistency of a random facet, to aid debugging. */
1545 if (time_msec() >= ofproto->consistency_rl
1546 && !classifier_is_empty(&ofproto->facets)
1547 && !ofproto->backer->need_revalidate) {
1548 struct cls_table *table;
1549 struct cls_rule *cr;
1550 struct facet *facet;
1552 ofproto->consistency_rl = time_msec() + 250;
1554 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1555 struct cls_table, hmap_node);
1556 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1558 facet = CONTAINER_OF(cr, struct facet, cr);
1560 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1561 facet->xout.tags)) {
1562 if (!facet_check_consistency(facet)) {
1563 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1572 wait(struct ofproto *ofproto_)
1574 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1575 struct ofport_dpif *ofport;
1576 struct ofbundle *bundle;
1578 if (!clogged && !list_is_empty(&ofproto->completions)) {
1579 poll_immediate_wake();
1582 if (ofproto_get_flow_restore_wait()) {
1586 dpif_wait(ofproto->backer->dpif);
1587 dpif_recv_wait(ofproto->backer->dpif);
1588 if (ofproto->sflow) {
1589 dpif_sflow_wait(ofproto->sflow);
1591 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1592 poll_immediate_wake();
1594 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1597 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1598 bundle_wait(bundle);
1600 if (ofproto->netflow) {
1601 netflow_wait(ofproto->netflow);
1603 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1604 mac_learning_wait(ofproto->ml);
1605 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1607 if (ofproto->backer->need_revalidate) {
1608 /* Shouldn't happen, but if it does just go around again. */
1609 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1610 poll_immediate_wake();
1615 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1617 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1618 struct cls_cursor cursor;
1619 size_t n_subfacets = 0;
1620 struct facet *facet;
1622 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1624 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1625 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1626 n_subfacets += list_size(&facet->subfacets);
1628 simap_increase(usage, "subfacets", n_subfacets);
1632 flush(struct ofproto *ofproto_)
1634 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1635 struct subfacet *subfacet, *next_subfacet;
1636 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1640 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1641 &ofproto->backer->subfacets) {
1642 if (subfacet->facet->ofproto != ofproto) {
1646 if (subfacet->path != SF_NOT_INSTALLED) {
1647 batch[n_batch++] = subfacet;
1648 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1649 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1653 subfacet_destroy(subfacet);
1658 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1663 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1664 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1666 *arp_match_ip = true;
1667 *actions = (OFPUTIL_A_OUTPUT |
1668 OFPUTIL_A_SET_VLAN_VID |
1669 OFPUTIL_A_SET_VLAN_PCP |
1670 OFPUTIL_A_STRIP_VLAN |
1671 OFPUTIL_A_SET_DL_SRC |
1672 OFPUTIL_A_SET_DL_DST |
1673 OFPUTIL_A_SET_NW_SRC |
1674 OFPUTIL_A_SET_NW_DST |
1675 OFPUTIL_A_SET_NW_TOS |
1676 OFPUTIL_A_SET_TP_SRC |
1677 OFPUTIL_A_SET_TP_DST |
1682 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1684 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1685 struct dpif_dp_stats s;
1686 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1689 strcpy(ots->name, "classifier");
1691 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1692 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1693 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1694 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1696 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1697 ots->lookup_count = htonll(n_lookup);
1698 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1701 static struct ofport *
1704 struct ofport_dpif *port = xmalloc(sizeof *port);
1709 port_dealloc(struct ofport *port_)
1711 struct ofport_dpif *port = ofport_dpif_cast(port_);
1716 port_construct(struct ofport *port_)
1718 struct ofport_dpif *port = ofport_dpif_cast(port_);
1719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1720 const struct netdev *netdev = port->up.netdev;
1721 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1722 struct dpif_port dpif_port;
1725 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1726 port->bundle = NULL;
1729 port->tag = tag_create_random();
1730 port->may_enable = true;
1731 port->stp_port = NULL;
1732 port->stp_state = STP_DISABLED;
1733 port->is_tunnel = false;
1735 hmap_init(&port->priorities);
1736 port->realdev_ofp_port = 0;
1737 port->vlandev_vid = 0;
1738 port->carrier_seq = netdev_get_carrier_resets(netdev);
1740 if (netdev_vport_is_patch(netdev)) {
1741 /* By bailing out here, we don't submit the port to the sFlow module
1742 * to be considered for counter polling export. This is correct
1743 * because the patch port represents an interface that sFlow considers
1744 * to be "internal" to the switch as a whole, and therefore not an
1745 * candidate for counter polling. */
1746 port->odp_port = ODPP_NONE;
1747 ofport_update_peer(port);
1751 error = dpif_port_query_by_name(ofproto->backer->dpif,
1752 netdev_vport_get_dpif_port(netdev, namebuf,
1759 port->odp_port = dpif_port.port_no;
1761 if (netdev_get_tunnel_config(netdev)) {
1762 tnl_port_add(port, port->up.netdev, port->odp_port);
1763 port->is_tunnel = true;
1765 /* Sanity-check that a mapping doesn't already exist. This
1766 * shouldn't happen for non-tunnel ports. */
1767 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1768 VLOG_ERR("port %s already has an OpenFlow port number",
1770 dpif_port_destroy(&dpif_port);
1774 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1775 hash_odp_port(port->odp_port));
1777 dpif_port_destroy(&dpif_port);
1779 if (ofproto->sflow) {
1780 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1787 port_destruct(struct ofport *port_)
1789 struct ofport_dpif *port = ofport_dpif_cast(port_);
1790 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1791 const char *devname = netdev_get_name(port->up.netdev);
1792 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1793 const char *dp_port_name;
1795 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1796 xlate_ofport_remove(port);
1798 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1800 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1801 /* The underlying device is still there, so delete it. This
1802 * happens when the ofproto is being destroyed, since the caller
1803 * assumes that removal of attached ports will happen as part of
1805 if (!port->is_tunnel) {
1806 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1811 port->peer->peer = NULL;
1815 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1816 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1820 sset_find_and_delete(&ofproto->ports, devname);
1821 sset_find_and_delete(&ofproto->ghost_ports, devname);
1822 bundle_remove(port_);
1823 set_cfm(port_, NULL);
1824 set_bfd(port_, NULL);
1825 if (ofproto->sflow) {
1826 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1829 ofport_clear_priorities(port);
1830 hmap_destroy(&port->priorities);
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);
1846 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1848 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1852 ofport_update_peer(port);
1856 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1858 struct ofport_dpif *port = ofport_dpif_cast(port_);
1859 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1860 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1862 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1863 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1864 OFPUTIL_PC_NO_PACKET_IN)) {
1865 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1867 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1868 bundle_update(port->bundle);
1874 set_sflow(struct ofproto *ofproto_,
1875 const struct ofproto_sflow_options *sflow_options)
1877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1878 struct dpif_sflow *ds = ofproto->sflow;
1880 if (sflow_options) {
1882 struct ofport_dpif *ofport;
1884 ds = ofproto->sflow = dpif_sflow_create();
1885 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1886 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1888 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1890 dpif_sflow_set_options(ds, sflow_options);
1893 dpif_sflow_unref(ds);
1894 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1895 ofproto->sflow = NULL;
1903 struct ofproto *ofproto_,
1904 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1905 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1906 size_t n_flow_exporters_options)
1908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1909 struct dpif_ipfix *di = ofproto->ipfix;
1911 if (bridge_exporter_options || flow_exporters_options) {
1913 di = ofproto->ipfix = dpif_ipfix_create();
1915 dpif_ipfix_set_options(
1916 di, bridge_exporter_options, flow_exporters_options,
1917 n_flow_exporters_options);
1920 dpif_ipfix_unref(di);
1921 ofproto->ipfix = NULL;
1928 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1930 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1937 struct ofproto_dpif *ofproto;
1939 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1940 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1941 ofport->cfm = cfm_create(ofport->up.netdev);
1944 if (cfm_configure(ofport->cfm, s)) {
1950 cfm_unref(ofport->cfm);
1956 get_cfm_status(const struct ofport *ofport_,
1957 struct ofproto_cfm_status *status)
1959 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1962 status->faults = cfm_get_fault(ofport->cfm);
1963 status->remote_opstate = cfm_get_opup(ofport->cfm);
1964 status->health = cfm_get_health(ofport->cfm);
1965 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1973 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1976 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1980 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1981 if (ofport->bfd != old) {
1982 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1989 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1991 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1994 bfd_get_status(ofport->bfd, smap);
2001 /* Spanning Tree. */
2004 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2006 struct ofproto_dpif *ofproto = ofproto_;
2007 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2008 struct ofport_dpif *ofport;
2010 ofport = stp_port_get_aux(sp);
2012 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2013 ofproto->up.name, port_num);
2015 struct eth_header *eth = pkt->l2;
2017 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2018 if (eth_addr_is_zero(eth->eth_src)) {
2019 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2020 "with unknown MAC", ofproto->up.name, port_num);
2022 send_packet(ofport, pkt);
2028 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2030 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2034 /* Only revalidate flows if the configuration changed. */
2035 if (!s != !ofproto->stp) {
2036 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2040 if (!ofproto->stp) {
2041 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2042 send_bpdu_cb, ofproto);
2043 ofproto->stp_last_tick = time_msec();
2046 stp_set_bridge_id(ofproto->stp, s->system_id);
2047 stp_set_bridge_priority(ofproto->stp, s->priority);
2048 stp_set_hello_time(ofproto->stp, s->hello_time);
2049 stp_set_max_age(ofproto->stp, s->max_age);
2050 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2052 struct ofport *ofport;
2054 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2055 set_stp_port(ofport, NULL);
2058 stp_unref(ofproto->stp);
2059 ofproto->stp = NULL;
2066 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2072 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2073 s->designated_root = stp_get_designated_root(ofproto->stp);
2074 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2083 update_stp_port_state(struct ofport_dpif *ofport)
2085 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2086 enum stp_state state;
2088 /* Figure out new state. */
2089 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2093 if (ofport->stp_state != state) {
2094 enum ofputil_port_state of_state;
2097 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2098 netdev_get_name(ofport->up.netdev),
2099 stp_state_name(ofport->stp_state),
2100 stp_state_name(state));
2101 if (stp_learn_in_state(ofport->stp_state)
2102 != stp_learn_in_state(state)) {
2103 /* xxx Learning action flows should also be flushed. */
2104 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2105 mac_learning_flush(ofproto->ml);
2106 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2108 fwd_change = stp_forward_in_state(ofport->stp_state)
2109 != stp_forward_in_state(state);
2111 ofproto->backer->need_revalidate = REV_STP;
2112 ofport->stp_state = state;
2113 ofport->stp_state_entered = time_msec();
2115 if (fwd_change && ofport->bundle) {
2116 bundle_update(ofport->bundle);
2119 /* Update the STP state bits in the OpenFlow port description. */
2120 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2121 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2122 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2123 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2124 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2126 ofproto_port_set_state(&ofport->up, of_state);
2130 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2131 * caller is responsible for assigning STP port numbers and ensuring
2132 * there are no duplicates. */
2134 set_stp_port(struct ofport *ofport_,
2135 const struct ofproto_port_stp_settings *s)
2137 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2139 struct stp_port *sp = ofport->stp_port;
2141 if (!s || !s->enable) {
2143 ofport->stp_port = NULL;
2144 stp_port_disable(sp);
2145 update_stp_port_state(ofport);
2148 } else if (sp && stp_port_no(sp) != s->port_num
2149 && ofport == stp_port_get_aux(sp)) {
2150 /* The port-id changed, so disable the old one if it's not
2151 * already in use by another port. */
2152 stp_port_disable(sp);
2155 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2156 stp_port_enable(sp);
2158 stp_port_set_aux(sp, ofport);
2159 stp_port_set_priority(sp, s->priority);
2160 stp_port_set_path_cost(sp, s->path_cost);
2162 update_stp_port_state(ofport);
2168 get_stp_port_status(struct ofport *ofport_,
2169 struct ofproto_port_stp_status *s)
2171 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2173 struct stp_port *sp = ofport->stp_port;
2175 if (!ofproto->stp || !sp) {
2181 s->port_id = stp_port_get_id(sp);
2182 s->state = stp_port_get_state(sp);
2183 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2184 s->role = stp_port_get_role(sp);
2185 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2191 stp_run(struct ofproto_dpif *ofproto)
2194 long long int now = time_msec();
2195 long long int elapsed = now - ofproto->stp_last_tick;
2196 struct stp_port *sp;
2199 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2200 ofproto->stp_last_tick = now;
2202 while (stp_get_changed_port(ofproto->stp, &sp)) {
2203 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2206 update_stp_port_state(ofport);
2210 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2211 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2212 mac_learning_flush(ofproto->ml);
2213 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2219 stp_wait(struct ofproto_dpif *ofproto)
2222 poll_timer_wait(1000);
2226 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
2227 * were used to make the determination.*/
2229 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
2231 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2232 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2236 stp_process_packet(const struct ofport_dpif *ofport,
2237 const struct ofpbuf *packet)
2239 struct ofpbuf payload = *packet;
2240 struct eth_header *eth = payload.data;
2241 struct stp_port *sp = ofport->stp_port;
2243 /* Sink packets on ports that have STP disabled when the bridge has
2245 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2249 /* Trim off padding on payload. */
2250 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2251 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2254 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2255 stp_received_bpdu(sp, payload.data, payload.size);
2260 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
2261 uint32_t queue_id, uint32_t *priority)
2263 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
2266 static struct priority_to_dscp *
2267 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2269 struct priority_to_dscp *pdscp;
2272 hash = hash_int(priority, 0);
2273 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2274 if (pdscp->priority == priority) {
2282 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2283 uint32_t priority, uint8_t *dscp)
2285 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2286 *dscp = pdscp ? pdscp->dscp : 0;
2287 return pdscp != NULL;
2291 ofport_clear_priorities(struct ofport_dpif *ofport)
2293 struct priority_to_dscp *pdscp, *next;
2295 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2296 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2302 set_queues(struct ofport *ofport_,
2303 const struct ofproto_port_queue *qdscp_list,
2306 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2308 struct hmap new = HMAP_INITIALIZER(&new);
2311 for (i = 0; i < n_qdscp; i++) {
2312 struct priority_to_dscp *pdscp;
2316 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2317 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2322 pdscp = get_priority(ofport, priority);
2324 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2326 pdscp = xmalloc(sizeof *pdscp);
2327 pdscp->priority = priority;
2329 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2332 if (pdscp->dscp != dscp) {
2334 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2337 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2340 if (!hmap_is_empty(&ofport->priorities)) {
2341 ofport_clear_priorities(ofport);
2342 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2345 hmap_swap(&new, &ofport->priorities);
2353 /* Expires all MAC learning entries associated with 'bundle' and forces its
2354 * ofproto to revalidate every flow.
2356 * Normally MAC learning entries are removed only from the ofproto associated
2357 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2358 * are removed from every ofproto. When patch ports and SLB bonds are in use
2359 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2360 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2361 * with the host from which it migrated. */
2363 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2365 struct ofproto_dpif *ofproto = bundle->ofproto;
2366 struct mac_learning *ml = ofproto->ml;
2367 struct mac_entry *mac, *next_mac;
2369 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2370 ovs_rwlock_wrlock(&ml->rwlock);
2371 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2372 if (mac->port.p == bundle) {
2374 struct ofproto_dpif *o;
2376 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2378 struct mac_entry *e;
2380 ovs_rwlock_wrlock(&o->ml->rwlock);
2381 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2383 mac_learning_expire(o->ml, e);
2385 ovs_rwlock_unlock(&o->ml->rwlock);
2390 mac_learning_expire(ml, mac);
2393 ovs_rwlock_unlock(&ml->rwlock);
2396 static struct ofbundle *
2397 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2399 struct ofbundle *bundle;
2401 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2402 &ofproto->bundles) {
2403 if (bundle->aux == aux) {
2411 bundle_update(struct ofbundle *bundle)
2413 struct ofport_dpif *port;
2415 bundle->floodable = true;
2416 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2417 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2418 || !stp_forward_in_state(port->stp_state)) {
2419 bundle->floodable = false;
2426 bundle_del_port(struct ofport_dpif *port)
2428 struct ofbundle *bundle = port->bundle;
2430 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2432 list_remove(&port->bundle_node);
2433 port->bundle = NULL;
2436 lacp_slave_unregister(bundle->lacp, port);
2439 bond_slave_unregister(bundle->bond, port);
2442 bundle_update(bundle);
2446 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2447 struct lacp_slave_settings *lacp)
2449 struct ofport_dpif *port;
2451 port = get_ofp_port(bundle->ofproto, ofp_port);
2456 if (port->bundle != bundle) {
2457 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2459 bundle_del_port(port);
2462 port->bundle = bundle;
2463 list_push_back(&bundle->ports, &port->bundle_node);
2464 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2465 || !stp_forward_in_state(port->stp_state)) {
2466 bundle->floodable = false;
2470 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2471 lacp_slave_register(bundle->lacp, port, lacp);
2478 bundle_destroy(struct ofbundle *bundle)
2480 struct ofproto_dpif *ofproto;
2481 struct ofport_dpif *port, *next_port;
2487 ofproto = bundle->ofproto;
2488 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2490 xlate_bundle_remove(bundle);
2492 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2493 bundle_del_port(port);
2496 bundle_flush_macs(bundle, true);
2497 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2499 free(bundle->trunks);
2500 lacp_unref(bundle->lacp);
2501 bond_unref(bundle->bond);
2506 bundle_set(struct ofproto *ofproto_, void *aux,
2507 const struct ofproto_bundle_settings *s)
2509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2510 bool need_flush = false;
2511 struct ofport_dpif *port;
2512 struct ofbundle *bundle;
2513 unsigned long *trunks;
2519 bundle_destroy(bundle_lookup(ofproto, aux));
2523 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2524 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2526 bundle = bundle_lookup(ofproto, aux);
2528 bundle = xmalloc(sizeof *bundle);
2530 bundle->ofproto = ofproto;
2531 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2532 hash_pointer(aux, 0));
2534 bundle->name = NULL;
2536 list_init(&bundle->ports);
2537 bundle->vlan_mode = PORT_VLAN_TRUNK;
2539 bundle->trunks = NULL;
2540 bundle->use_priority_tags = s->use_priority_tags;
2541 bundle->lacp = NULL;
2542 bundle->bond = NULL;
2544 bundle->floodable = true;
2545 mbridge_register_bundle(ofproto->mbridge, bundle);
2548 if (!bundle->name || strcmp(s->name, bundle->name)) {
2550 bundle->name = xstrdup(s->name);
2555 if (!bundle->lacp) {
2556 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2557 bundle->lacp = lacp_create();
2559 lacp_configure(bundle->lacp, s->lacp);
2561 lacp_unref(bundle->lacp);
2562 bundle->lacp = NULL;
2565 /* Update set of ports. */
2567 for (i = 0; i < s->n_slaves; i++) {
2568 if (!bundle_add_port(bundle, s->slaves[i],
2569 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2573 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2574 struct ofport_dpif *next_port;
2576 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2577 for (i = 0; i < s->n_slaves; i++) {
2578 if (s->slaves[i] == port->up.ofp_port) {
2583 bundle_del_port(port);
2587 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2589 if (list_is_empty(&bundle->ports)) {
2590 bundle_destroy(bundle);
2594 /* Set VLAN tagging mode */
2595 if (s->vlan_mode != bundle->vlan_mode
2596 || s->use_priority_tags != bundle->use_priority_tags) {
2597 bundle->vlan_mode = s->vlan_mode;
2598 bundle->use_priority_tags = s->use_priority_tags;
2603 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2604 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2606 if (vlan != bundle->vlan) {
2607 bundle->vlan = vlan;
2611 /* Get trunked VLANs. */
2612 switch (s->vlan_mode) {
2613 case PORT_VLAN_ACCESS:
2617 case PORT_VLAN_TRUNK:
2618 trunks = CONST_CAST(unsigned long *, s->trunks);
2621 case PORT_VLAN_NATIVE_UNTAGGED:
2622 case PORT_VLAN_NATIVE_TAGGED:
2623 if (vlan != 0 && (!s->trunks
2624 || !bitmap_is_set(s->trunks, vlan)
2625 || bitmap_is_set(s->trunks, 0))) {
2626 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2628 trunks = bitmap_clone(s->trunks, 4096);
2630 trunks = bitmap_allocate1(4096);
2632 bitmap_set1(trunks, vlan);
2633 bitmap_set0(trunks, 0);
2635 trunks = CONST_CAST(unsigned long *, s->trunks);
2642 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2643 free(bundle->trunks);
2644 if (trunks == s->trunks) {
2645 bundle->trunks = vlan_bitmap_clone(trunks);
2647 bundle->trunks = trunks;
2652 if (trunks != s->trunks) {
2657 if (!list_is_short(&bundle->ports)) {
2658 bundle->ofproto->has_bonded_bundles = true;
2660 if (bond_reconfigure(bundle->bond, s->bond)) {
2661 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2664 bundle->bond = bond_create(s->bond);
2665 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2668 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2669 bond_slave_register(bundle->bond, port, port->up.netdev);
2672 bond_unref(bundle->bond);
2673 bundle->bond = NULL;
2676 /* If we changed something that would affect MAC learning, un-learn
2677 * everything on this port and force flow revalidation. */
2679 bundle_flush_macs(bundle, false);
2686 bundle_remove(struct ofport *port_)
2688 struct ofport_dpif *port = ofport_dpif_cast(port_);
2689 struct ofbundle *bundle = port->bundle;
2692 bundle_del_port(port);
2693 if (list_is_empty(&bundle->ports)) {
2694 bundle_destroy(bundle);
2695 } else if (list_is_short(&bundle->ports)) {
2696 bond_unref(bundle->bond);
2697 bundle->bond = NULL;
2703 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2705 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2706 struct ofport_dpif *port = port_;
2707 uint8_t ea[ETH_ADDR_LEN];
2710 error = netdev_get_etheraddr(port->up.netdev, ea);
2712 struct ofpbuf packet;
2715 ofpbuf_init(&packet, 0);
2716 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2718 memcpy(packet_pdu, pdu, pdu_size);
2720 send_packet(port, &packet);
2721 ofpbuf_uninit(&packet);
2723 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2724 "%s (%s)", port->bundle->name,
2725 netdev_get_name(port->up.netdev), ovs_strerror(error));
2730 bundle_send_learning_packets(struct ofbundle *bundle)
2732 struct ofproto_dpif *ofproto = bundle->ofproto;
2733 int error, n_packets, n_errors;
2734 struct mac_entry *e;
2736 error = n_packets = n_errors = 0;
2737 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2738 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2739 if (e->port.p != bundle) {
2740 struct ofpbuf *learning_packet;
2741 struct ofport_dpif *port;
2745 /* The assignment to "port" is unnecessary but makes "grep"ing for
2746 * struct ofport_dpif more effective. */
2747 learning_packet = bond_compose_learning_packet(bundle->bond,
2751 ret = send_packet(port, learning_packet);
2752 ofpbuf_delete(learning_packet);
2760 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2763 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2764 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2765 "packets, last error was: %s",
2766 bundle->name, n_errors, n_packets, ovs_strerror(error));
2768 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2769 bundle->name, n_packets);
2774 bundle_run(struct ofbundle *bundle)
2777 lacp_run(bundle->lacp, send_pdu_cb);
2780 struct ofport_dpif *port;
2782 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2783 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2786 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2787 lacp_status(bundle->lacp));
2788 if (bond_should_send_learning_packets(bundle->bond)) {
2789 bundle_send_learning_packets(bundle);
2795 bundle_wait(struct ofbundle *bundle)
2798 lacp_wait(bundle->lacp);
2801 bond_wait(bundle->bond);
2808 mirror_set__(struct ofproto *ofproto_, void *aux,
2809 const struct ofproto_mirror_settings *s)
2811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2812 struct ofbundle **srcs, **dsts;
2817 mirror_destroy(ofproto->mbridge, aux);
2821 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2822 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2824 for (i = 0; i < s->n_srcs; i++) {
2825 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2828 for (i = 0; i < s->n_dsts; i++) {
2829 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2832 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2833 s->n_dsts, s->src_vlans,
2834 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2841 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2842 uint64_t *packets, uint64_t *bytes)
2845 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2850 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2852 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2853 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2854 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2855 mac_learning_flush(ofproto->ml);
2857 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2862 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2865 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2866 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2870 forward_bpdu_changed(struct ofproto *ofproto_)
2872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2873 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2877 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2881 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2882 mac_learning_set_idle_time(ofproto->ml, idle_time);
2883 mac_learning_set_max_entries(ofproto->ml, max_entries);
2884 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2889 static struct ofport_dpif *
2890 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2892 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2893 return ofport ? ofport_dpif_cast(ofport) : NULL;
2896 static struct ofport_dpif *
2897 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2899 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2900 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2904 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2905 struct ofproto_port *ofproto_port,
2906 struct dpif_port *dpif_port)
2908 ofproto_port->name = dpif_port->name;
2909 ofproto_port->type = dpif_port->type;
2910 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2914 ofport_update_peer(struct ofport_dpif *ofport)
2916 const struct ofproto_dpif *ofproto;
2917 struct dpif_backer *backer;
2918 const char *peer_name;
2920 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2924 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2925 backer->need_revalidate = REV_RECONFIGURE;
2928 ofport->peer->peer = NULL;
2929 ofport->peer = NULL;
2932 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2937 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2938 struct ofport *peer_ofport;
2939 struct ofport_dpif *peer;
2940 const char *peer_peer;
2942 if (ofproto->backer != backer) {
2946 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2951 peer = ofport_dpif_cast(peer_ofport);
2952 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2953 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2955 ofport->peer = peer;
2956 ofport->peer->peer = ofport;
2964 port_run_fast(struct ofport_dpif *ofport)
2966 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2967 struct ofpbuf packet;
2969 ofpbuf_init(&packet, 0);
2970 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2971 send_packet(ofport, &packet);
2972 ofpbuf_uninit(&packet);
2975 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2976 struct ofpbuf packet;
2978 ofpbuf_init(&packet, 0);
2979 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2980 send_packet(ofport, &packet);
2981 ofpbuf_uninit(&packet);
2986 port_run(struct ofport_dpif *ofport)
2988 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2989 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2990 bool enable = netdev_get_carrier(ofport->up.netdev);
2992 ofport->carrier_seq = carrier_seq;
2994 port_run_fast(ofport);
2997 int cfm_opup = cfm_get_opup(ofport->cfm);
2999 cfm_run(ofport->cfm);
3000 enable = enable && !cfm_get_fault(ofport->cfm);
3002 if (cfm_opup >= 0) {
3003 enable = enable && cfm_opup;
3008 bfd_run(ofport->bfd);
3009 enable = enable && bfd_forwarding(ofport->bfd);
3012 if (ofport->bundle) {
3013 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3014 if (carrier_changed) {
3015 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3019 if (ofport->may_enable != enable) {
3020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3021 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3024 ofport->may_enable = enable;
3028 port_wait(struct ofport_dpif *ofport)
3031 cfm_wait(ofport->cfm);
3035 bfd_wait(ofport->bfd);
3040 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3041 struct ofproto_port *ofproto_port)
3043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3044 struct dpif_port dpif_port;
3047 if (sset_contains(&ofproto->ghost_ports, devname)) {
3048 const char *type = netdev_get_type_from_name(devname);
3050 /* We may be called before ofproto->up.port_by_name is populated with
3051 * the appropriate ofport. For this reason, we must get the name and
3052 * type from the netdev layer directly. */
3054 const struct ofport *ofport;
3056 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3057 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3058 ofproto_port->name = xstrdup(devname);
3059 ofproto_port->type = xstrdup(type);
3065 if (!sset_contains(&ofproto->ports, devname)) {
3068 error = dpif_port_query_by_name(ofproto->backer->dpif,
3069 devname, &dpif_port);
3071 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3077 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3079 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3080 const char *devname = netdev_get_name(netdev);
3081 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3082 const char *dp_port_name;
3084 if (netdev_vport_is_patch(netdev)) {
3085 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3089 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3090 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3091 odp_port_t port_no = ODPP_NONE;
3094 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3098 if (netdev_get_tunnel_config(netdev)) {
3099 simap_put(&ofproto->backer->tnl_backers,
3100 dp_port_name, odp_to_u32(port_no));
3104 if (netdev_get_tunnel_config(netdev)) {
3105 sset_add(&ofproto->ghost_ports, devname);
3107 sset_add(&ofproto->ports, devname);
3113 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3116 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3123 sset_find_and_delete(&ofproto->ghost_ports,
3124 netdev_get_name(ofport->up.netdev));
3125 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3126 if (!ofport->is_tunnel) {
3127 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3129 /* The caller is going to close ofport->up.netdev. If this is a
3130 * bonded port, then the bond is using that netdev, so remove it
3131 * from the bond. The client will need to reconfigure everything
3132 * after deleting ports, so then the slave will get re-added. */
3133 bundle_remove(&ofport->up);
3140 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3142 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3147 error = netdev_get_stats(ofport->up.netdev, stats);
3149 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3152 /* ofproto->stats.tx_packets represents packets that we created
3153 * internally and sent to some port (e.g. packets sent with
3154 * send_packet()). Account for them as if they had come from
3155 * OFPP_LOCAL and got forwarded. */
3157 if (stats->rx_packets != UINT64_MAX) {
3158 stats->rx_packets += ofproto->stats.tx_packets;
3161 if (stats->rx_bytes != UINT64_MAX) {
3162 stats->rx_bytes += ofproto->stats.tx_bytes;
3165 /* ofproto->stats.rx_packets represents packets that were received on
3166 * some port and we processed internally and dropped (e.g. STP).
3167 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3169 if (stats->tx_packets != UINT64_MAX) {
3170 stats->tx_packets += ofproto->stats.rx_packets;
3173 if (stats->tx_bytes != UINT64_MAX) {
3174 stats->tx_bytes += ofproto->stats.rx_bytes;
3181 struct port_dump_state {
3186 struct ofproto_port port;
3191 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3193 *statep = xzalloc(sizeof(struct port_dump_state));
3198 port_dump_next(const struct ofproto *ofproto_, void *state_,
3199 struct ofproto_port *port)
3201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3202 struct port_dump_state *state = state_;
3203 const struct sset *sset;
3204 struct sset_node *node;
3206 if (state->has_port) {
3207 ofproto_port_destroy(&state->port);
3208 state->has_port = false;
3210 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3211 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3214 error = port_query_by_name(ofproto_, node->name, &state->port);
3216 *port = state->port;
3217 state->has_port = true;
3219 } else if (error != ENODEV) {
3224 if (!state->ghost) {
3225 state->ghost = true;
3228 return port_dump_next(ofproto_, state_, port);
3235 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3237 struct port_dump_state *state = state_;
3239 if (state->has_port) {
3240 ofproto_port_destroy(&state->port);
3247 port_poll(const struct ofproto *ofproto_, char **devnamep)
3249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3251 if (ofproto->port_poll_errno) {
3252 int error = ofproto->port_poll_errno;
3253 ofproto->port_poll_errno = 0;
3257 if (sset_is_empty(&ofproto->port_poll_set)) {
3261 *devnamep = sset_pop(&ofproto->port_poll_set);
3266 port_poll_wait(const struct ofproto *ofproto_)
3268 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3269 dpif_port_poll_wait(ofproto->backer->dpif);
3273 port_is_lacp_current(const struct ofport *ofport_)
3275 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3276 return (ofport->bundle && ofport->bundle->lacp
3277 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3281 /* Upcall handling. */
3283 /* Flow miss batching.
3285 * Some dpifs implement operations faster when you hand them off in a batch.
3286 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3287 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3288 * more packets, plus possibly installing the flow in the dpif.
3290 * So far we only batch the operations that affect flow setup time the most.
3291 * It's possible to batch more than that, but the benefit might be minimal. */
3293 struct hmap_node hmap_node;
3294 struct ofproto_dpif *ofproto;
3296 enum odp_key_fitness key_fitness;
3297 const struct nlattr *key;
3299 struct list packets;
3300 enum dpif_upcall_type upcall_type;
3303 struct flow_miss_op {
3304 struct dpif_op dpif_op;
3306 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3307 struct xlate_out xout;
3308 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3310 struct ofpbuf mask; /* Flow mask for "put" ops. */
3311 struct odputil_keybuf maskbuf;
3313 /* If this is a "put" op, then a pointer to the subfacet that should
3314 * be marked as uninstalled if the operation fails. */
3315 struct subfacet *subfacet;
3318 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3319 * OpenFlow controller as necessary according to their individual
3320 * configurations. */
3322 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3323 const struct flow *flow)
3325 struct ofputil_packet_in pin;
3327 pin.packet = packet->data;
3328 pin.packet_len = packet->size;
3329 pin.reason = OFPR_NO_MATCH;
3330 pin.controller_id = 0;
3335 pin.send_len = 0; /* not used for flow table misses */
3337 flow_get_metadata(flow, &pin.fmd);
3339 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3342 static struct flow_miss *
3343 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3344 const struct flow *flow, uint32_t hash)
3346 struct flow_miss *miss;
3348 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3349 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3357 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3358 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3359 * 'miss' is associated with a subfacet the caller must also initialize the
3360 * returned op->subfacet, and if anything needs to be freed after processing
3361 * the op, the caller must initialize op->garbage also. */
3363 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3364 struct flow_miss_op *op)
3366 if (miss->flow.in_port.ofp_port
3367 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3368 miss->flow.vlan_tci)) {
3369 /* This packet was received on a VLAN splinter port. We
3370 * added a VLAN to the packet to make the packet resemble
3371 * the flow, but the actions were composed assuming that
3372 * the packet contained no VLAN. So, we must remove the
3373 * VLAN header from the packet before trying to execute the
3375 eth_pop_vlan(packet);
3378 op->subfacet = NULL;
3379 op->xout_garbage = false;
3380 op->dpif_op.type = DPIF_OP_EXECUTE;
3381 op->dpif_op.u.execute.key = miss->key;
3382 op->dpif_op.u.execute.key_len = miss->key_len;
3383 op->dpif_op.u.execute.packet = packet;
3384 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3387 /* Helper for handle_flow_miss_without_facet() and
3388 * handle_flow_miss_with_facet(). */
3390 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3391 const struct flow *flow, bool fail_open)
3395 * Extra-special case for fail-open mode.
3397 * We are in fail-open mode and the packet matched the fail-open
3398 * rule, but we are connected to a controller too. We should send
3399 * the packet up to the controller in the hope that it will try to
3400 * set up a flow and thereby allow us to exit fail-open.
3402 * See the top-level comment in fail-open.c for more information.
3404 send_packet_in_miss(ofproto, packet, flow);
3408 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3409 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3410 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3411 * return value of true). However, for short flows the cost of bookkeeping is
3412 * much higher than the benefits, so when the datapath holds a large number of
3413 * flows we impose some heuristics to decide which flows are likely to be worth
3416 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3418 struct dpif_backer *backer = miss->ofproto->backer;
3421 switch (flow_miss_model) {
3422 case OFPROTO_HANDLE_MISS_AUTO:
3424 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3426 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3430 if (!backer->governor) {
3433 n_subfacets = hmap_count(&backer->subfacets);
3434 if (n_subfacets * 2 <= flow_eviction_threshold) {
3438 backer->governor = governor_create();
3441 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3442 return governor_should_install_flow(backer->governor, hash,
3443 list_size(&miss->packets));
3446 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3447 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3448 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3450 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3451 struct flow_miss *miss,
3452 struct flow_miss_op *ops, size_t *n_ops)
3454 struct ofpbuf *packet;
3456 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3458 COVERAGE_INC(facet_suppress);
3460 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3461 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3464 struct xlate_in xin;
3466 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3467 xlate_actions_for_side_effects(&xin);
3470 if (xout->odp_actions.size) {
3471 struct flow_miss_op *op = &ops[*n_ops];
3472 struct dpif_execute *execute = &op->dpif_op.u.execute;
3474 init_flow_miss_execute_op(miss, packet, op);
3475 xlate_out_copy(&op->xout, xout);
3476 execute->actions = op->xout.odp_actions.data;
3477 execute->actions_len = op->xout.odp_actions.size;
3478 op->xout_garbage = true;
3485 /* Handles 'miss', which matches 'facet'. May add any required datapath
3486 * operations to 'ops', incrementing '*n_ops' for each new op.
3488 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3489 * This is really important only for new facets: if we just called time_msec()
3490 * here, then the new subfacet or its packets could look (occasionally) as
3491 * though it was used some time after the facet was used. That can make a
3492 * one-packet flow look like it has a nonzero duration, which looks odd in
3493 * e.g. NetFlow statistics.
3495 * If non-null, 'stats' will be folded into 'facet'. */
3497 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3498 long long int now, struct dpif_flow_stats *stats,
3499 struct flow_miss_op *ops, size_t *n_ops)
3501 enum subfacet_path want_path;
3502 struct subfacet *subfacet;
3503 struct ofpbuf *packet;
3505 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3507 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3508 struct flow_miss_op *op = &ops[*n_ops];
3510 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3513 if (want_path != SF_FAST_PATH) {
3514 struct rule_dpif *rule;
3515 struct xlate_in xin;
3517 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3518 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3519 xlate_actions_for_side_effects(&xin);
3522 if (facet->xout.odp_actions.size) {
3523 struct dpif_execute *execute = &op->dpif_op.u.execute;
3525 init_flow_miss_execute_op(miss, packet, op);
3526 execute->actions = facet->xout.odp_actions.data,
3527 execute->actions_len = facet->xout.odp_actions.size;
3532 /* Don't install the flow if it's the result of the "userspace"
3533 * action for an already installed facet. This can occur when a
3534 * datapath flow with wildcards has a "userspace" action and flows
3535 * sent to userspace result in a different subfacet, which will then
3536 * be rejected as overlapping by the datapath. */
3537 if (miss->upcall_type == DPIF_UC_ACTION
3538 && !list_is_empty(&facet->subfacets)) {
3540 facet->used = MAX(facet->used, stats->used);
3541 facet->packet_count += stats->n_packets;
3542 facet->byte_count += stats->n_bytes;
3543 facet->tcp_flags |= stats->tcp_flags;
3548 subfacet = subfacet_create(facet, miss, now);
3550 subfacet_update_stats(subfacet, stats);
3553 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3554 struct flow_miss_op *op = &ops[(*n_ops)++];
3555 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3557 subfacet->path = want_path;
3559 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3560 if (enable_megaflows) {
3561 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3562 &miss->flow, UINT32_MAX);
3565 op->xout_garbage = false;
3566 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3567 op->subfacet = subfacet;
3568 put->flags = DPIF_FP_CREATE;
3569 put->key = miss->key;
3570 put->key_len = miss->key_len;
3571 put->mask = op->mask.data;
3572 put->mask_len = op->mask.size;
3574 if (want_path == SF_FAST_PATH) {
3575 put->actions = facet->xout.odp_actions.data;
3576 put->actions_len = facet->xout.odp_actions.size;
3578 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3579 op->slow_stub, sizeof op->slow_stub,
3580 &put->actions, &put->actions_len);
3586 /* Handles flow miss 'miss'. May add any required datapath operations
3587 * to 'ops', incrementing '*n_ops' for each new op. */
3589 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3592 struct ofproto_dpif *ofproto = miss->ofproto;
3593 struct dpif_flow_stats stats__;
3594 struct dpif_flow_stats *stats = &stats__;
3595 struct ofpbuf *packet;
3596 struct facet *facet;
3600 memset(stats, 0, sizeof *stats);
3602 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3603 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3604 stats->n_bytes += packet->size;
3608 facet = facet_lookup_valid(ofproto, &miss->flow);
3610 struct flow_wildcards wc;
3611 struct rule_dpif *rule;
3612 struct xlate_out xout;
3613 struct xlate_in xin;
3615 flow_wildcards_init_catchall(&wc);
3616 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3617 rule_credit_stats(rule, stats);
3619 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3621 xin.resubmit_stats = stats;
3622 xin.may_learn = true;
3623 xlate_actions(&xin, &xout);
3624 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3626 /* There does not exist a bijection between 'struct flow' and datapath
3627 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3628 * assumption used throughout the facet and subfacet handling code.
3629 * Since we have to handle these misses in userspace anyway, we simply
3630 * skip facet creation, avoiding the problem altogether. */
3631 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3632 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3633 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3637 facet = facet_create(miss, rule, &xout, stats);
3640 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3643 static struct drop_key *
3644 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3647 struct drop_key *drop_key;
3649 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3650 &backer->drop_keys) {
3651 if (drop_key->key_len == key_len
3652 && !memcmp(drop_key->key, key, key_len)) {
3660 drop_key_clear(struct dpif_backer *backer)
3662 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3663 struct drop_key *drop_key, *next;
3665 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3668 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3670 if (error && !VLOG_DROP_WARN(&rl)) {
3671 struct ds ds = DS_EMPTY_INITIALIZER;
3672 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3673 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3674 ovs_strerror(error), ds_cstr(&ds));
3678 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3679 free(drop_key->key);
3684 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3685 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3686 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3687 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3688 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3689 * 'packet' ingressed.
3691 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3692 * 'flow''s in_port to OFPP_NONE.
3694 * This function does post-processing on data returned from
3695 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3696 * of the upcall processing logic. In particular, if the extracted in_port is
3697 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3698 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3699 * a VLAN header onto 'packet' (if it is nonnull).
3701 * Similarly, this function also includes some logic to help with tunnels. It
3702 * may modify 'flow' as necessary to make the tunneling implementation
3703 * transparent to the upcall processing logic.
3705 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3706 * or some other positive errno if there are other problems. */
3708 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3709 const struct nlattr *key, size_t key_len,
3710 struct flow *flow, enum odp_key_fitness *fitnessp,
3711 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3713 const struct ofport_dpif *port;
3714 enum odp_key_fitness fitness;
3717 fitness = odp_flow_key_to_flow(key, key_len, flow);
3718 if (fitness == ODP_FIT_ERROR) {
3724 *odp_in_port = flow->in_port.odp_port;
3727 port = (tnl_port_should_receive(flow)
3728 ? tnl_port_receive(flow)
3729 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3730 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3735 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3736 * it's theoretically possible that we'll receive an ofport belonging to an
3737 * entirely different datapath. In practice, this can't happen because no
3738 * platforms has two separate datapaths which each support tunneling. */
3739 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3741 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3743 /* Make the packet resemble the flow, so that it gets sent to
3744 * an OpenFlow controller properly, so that it looks correct
3745 * for sFlow, and so that flow_extract() will get the correct
3746 * vlan_tci if it is called on 'packet'.
3748 * The allocated space inside 'packet' probably also contains
3749 * 'key', that is, both 'packet' and 'key' are probably part of
3750 * a struct dpif_upcall (see the large comment on that
3751 * structure definition), so pushing data on 'packet' is in
3752 * general not a good idea since it could overwrite 'key' or
3753 * free it as a side effect. However, it's OK in this special
3754 * case because we know that 'packet' is inside a Netlink
3755 * attribute: pushing 4 bytes will just overwrite the 4-byte
3756 * "struct nlattr", which is fine since we don't need that
3757 * header anymore. */
3758 eth_push_vlan(packet, flow->vlan_tci);
3760 /* We can't reproduce 'key' from 'flow'. */
3761 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3766 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3771 *fitnessp = fitness;
3777 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3780 struct dpif_upcall *upcall;
3781 struct flow_miss *miss;
3782 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3783 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3784 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3794 /* Construct the to-do list.
3796 * This just amounts to extracting the flow from each packet and sticking
3797 * the packets that have the same flow in the same "flow_miss" structure so
3798 * that we can process them together. */
3801 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3802 struct flow_miss *miss = &misses[n_misses];
3803 struct flow_miss *existing_miss;
3804 struct ofproto_dpif *ofproto;
3805 odp_port_t odp_in_port;
3810 error = ofproto_receive(backer, upcall->packet, upcall->key,
3811 upcall->key_len, &flow, &miss->key_fitness,
3812 &ofproto, &odp_in_port);
3813 if (error == ENODEV) {
3814 struct drop_key *drop_key;
3816 /* Received packet on datapath port for which we couldn't
3817 * associate an ofproto. This can happen if a port is removed
3818 * while traffic is being received. Print a rate-limited message
3819 * in case it happens frequently. Install a drop flow so
3820 * that future packets of the flow are inexpensively dropped
3822 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3823 "%"PRIu32, odp_in_port);
3825 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3828 ret = dpif_flow_put(backer->dpif,
3829 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3830 upcall->key, upcall->key_len,
3831 NULL, 0, NULL, 0, NULL);
3834 drop_key = xmalloc(sizeof *drop_key);
3835 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3836 drop_key->key_len = upcall->key_len;
3838 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3839 hash_bytes(drop_key->key, drop_key->key_len, 0));
3848 ofproto->n_missed++;
3849 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3850 &flow.tunnel, &flow.in_port, &miss->flow);
3852 /* Add other packets to a to-do list. */
3853 hash = flow_hash(&miss->flow, 0);
3854 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3855 if (!existing_miss) {
3856 hmap_insert(&todo, &miss->hmap_node, hash);
3857 miss->ofproto = ofproto;
3858 miss->key = upcall->key;
3859 miss->key_len = upcall->key_len;
3860 miss->upcall_type = upcall->type;
3861 list_init(&miss->packets);
3865 miss = existing_miss;
3867 list_push_back(&miss->packets, &upcall->packet->list_node);
3870 /* Process each element in the to-do list, constructing the set of
3871 * operations to batch. */
3873 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3874 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3876 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3878 /* Execute batch. */
3879 for (i = 0; i < n_ops; i++) {
3880 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3882 dpif_operate(backer->dpif, dpif_ops, n_ops);
3884 for (i = 0; i < n_ops; i++) {
3885 if (dpif_ops[i]->error != 0
3886 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3887 && flow_miss_ops[i].subfacet) {
3888 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3890 COVERAGE_INC(subfacet_install_fail);
3892 /* Zero-out subfacet counters when installation failed, but
3893 * datapath reported hits. This should not happen and
3894 * indicates a bug, since if the datapath flow exists, we
3895 * should not be attempting to create a new subfacet. A
3896 * buggy datapath could trigger this, so just zero out the
3897 * counters and log an error. */
3898 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3899 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3900 "datapath reported hits");
3901 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3904 subfacet->path = SF_NOT_INSTALLED;
3908 if (flow_miss_ops[i].xout_garbage) {
3909 xlate_out_uninit(&flow_miss_ops[i].xout);
3912 hmap_destroy(&todo);
3915 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3917 classify_upcall(const struct dpif_upcall *upcall)
3919 size_t userdata_len;
3920 union user_action_cookie cookie;
3922 /* First look at the upcall type. */
3923 switch (upcall->type) {
3924 case DPIF_UC_ACTION:
3930 case DPIF_N_UC_TYPES:
3932 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3936 /* "action" upcalls need a closer look. */
3937 if (!upcall->userdata) {
3938 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3941 userdata_len = nl_attr_get_size(upcall->userdata);
3942 if (userdata_len < sizeof cookie.type
3943 || userdata_len > sizeof cookie) {
3944 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3948 memset(&cookie, 0, sizeof cookie);
3949 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3950 if (userdata_len == sizeof cookie.sflow
3951 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3952 return SFLOW_UPCALL;
3953 } else if (userdata_len == sizeof cookie.slow_path
3954 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3956 } else if (userdata_len == sizeof cookie.flow_sample
3957 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3958 return FLOW_SAMPLE_UPCALL;
3959 } else if (userdata_len == sizeof cookie.ipfix
3960 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3961 return IPFIX_UPCALL;
3963 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3964 " and size %zu", cookie.type, userdata_len);
3970 handle_sflow_upcall(struct dpif_backer *backer,
3971 const struct dpif_upcall *upcall)
3973 struct ofproto_dpif *ofproto;
3974 union user_action_cookie cookie;
3976 odp_port_t odp_in_port;
3978 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3979 &flow, NULL, &ofproto, &odp_in_port)
3980 || !ofproto->sflow) {
3984 memset(&cookie, 0, sizeof cookie);
3985 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3986 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3987 odp_in_port, &cookie);
3991 handle_flow_sample_upcall(struct dpif_backer *backer,
3992 const struct dpif_upcall *upcall)
3994 struct ofproto_dpif *ofproto;
3995 union user_action_cookie cookie;
3998 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3999 &flow, NULL, &ofproto, NULL)
4000 || !ofproto->ipfix) {
4004 memset(&cookie, 0, sizeof cookie);
4005 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4007 /* The flow reflects exactly the contents of the packet. Sample
4008 * the packet using it. */
4009 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4010 cookie.flow_sample.collector_set_id,
4011 cookie.flow_sample.probability,
4012 cookie.flow_sample.obs_domain_id,
4013 cookie.flow_sample.obs_point_id);
4017 handle_ipfix_upcall(struct dpif_backer *backer,
4018 const struct dpif_upcall *upcall)
4020 struct ofproto_dpif *ofproto;
4023 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4024 &flow, NULL, &ofproto, NULL)
4025 || !ofproto->ipfix) {
4029 /* The flow reflects exactly the contents of the packet. Sample
4030 * the packet using it. */
4031 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4035 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4037 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4038 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4039 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4044 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4047 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4048 struct dpif_upcall *upcall = &misses[n_misses];
4049 struct ofpbuf *buf = &miss_bufs[n_misses];
4052 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4053 sizeof miss_buf_stubs[n_misses]);
4054 error = dpif_recv(backer->dpif, upcall, buf);
4060 switch (classify_upcall(upcall)) {
4062 /* Handle it later. */
4067 handle_sflow_upcall(backer, upcall);
4071 case FLOW_SAMPLE_UPCALL:
4072 handle_flow_sample_upcall(backer, upcall);
4077 handle_ipfix_upcall(backer, upcall);
4087 /* Handle deferred MISS_UPCALL processing. */
4088 handle_miss_upcalls(backer, misses, n_misses);
4089 for (i = 0; i < n_misses; i++) {
4090 ofpbuf_uninit(&miss_bufs[i]);
4096 /* Flow expiration. */
4098 static int subfacet_max_idle(const struct dpif_backer *);
4099 static void update_stats(struct dpif_backer *);
4100 static void rule_expire(struct rule_dpif *);
4101 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4103 /* This function is called periodically by run(). Its job is to collect
4104 * updates for the flows that have been installed into the datapath, most
4105 * importantly when they last were used, and then use that information to
4106 * expire flows that have not been used recently.
4108 * Returns the number of milliseconds after which it should be called again. */
4110 expire(struct dpif_backer *backer)
4112 struct ofproto_dpif *ofproto;
4116 /* Periodically clear out the drop keys in an effort to keep them
4117 * relatively few. */
4118 drop_key_clear(backer);
4120 /* Update stats for each flow in the backer. */
4121 update_stats(backer);
4123 n_subfacets = hmap_count(&backer->subfacets);
4125 struct subfacet *subfacet;
4126 long long int total, now;
4130 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4131 total += now - subfacet->created;
4133 backer->avg_subfacet_life += total / n_subfacets;
4135 backer->avg_subfacet_life /= 2;
4137 backer->avg_n_subfacet += n_subfacets;
4138 backer->avg_n_subfacet /= 2;
4140 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4142 max_idle = subfacet_max_idle(backer);
4143 expire_subfacets(backer, max_idle);
4145 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4146 struct rule *rule, *next_rule;
4148 if (ofproto->backer != backer) {
4152 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4154 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4155 &ofproto->up.expirable) {
4156 rule_expire(rule_dpif_cast(rule));
4159 /* All outstanding data in existing flows has been accounted, so it's a
4160 * good time to do bond rebalancing. */
4161 if (ofproto->has_bonded_bundles) {
4162 struct ofbundle *bundle;
4164 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4166 bond_rebalance(bundle->bond, &backer->revalidate_set);
4172 return MIN(max_idle, 1000);
4175 /* Updates flow table statistics given that the datapath just reported 'stats'
4176 * as 'subfacet''s statistics. */
4178 update_subfacet_stats(struct subfacet *subfacet,
4179 const struct dpif_flow_stats *stats)
4181 struct facet *facet = subfacet->facet;
4182 struct dpif_flow_stats diff;
4184 diff.tcp_flags = stats->tcp_flags;
4185 diff.used = stats->used;
4187 if (stats->n_packets >= subfacet->dp_packet_count) {
4188 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4190 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4194 if (stats->n_bytes >= subfacet->dp_byte_count) {
4195 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4197 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4201 facet->ofproto->n_hit += diff.n_packets;
4202 subfacet->dp_packet_count = stats->n_packets;
4203 subfacet->dp_byte_count = stats->n_bytes;
4204 subfacet_update_stats(subfacet, &diff);
4206 if (facet->accounted_bytes < facet->byte_count) {
4208 facet_account(facet);
4209 facet->accounted_bytes = facet->byte_count;
4213 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4214 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4216 delete_unexpected_flow(struct dpif_backer *backer,
4217 const struct nlattr *key, size_t key_len)
4219 if (!VLOG_DROP_WARN(&rl)) {
4223 odp_flow_key_format(key, key_len, &s);
4224 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4228 COVERAGE_INC(facet_unexpected);
4229 dpif_flow_del(backer->dpif, key, key_len, NULL);
4232 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4234 * This function also pushes statistics updates to rules which each facet
4235 * resubmits into. Generally these statistics will be accurate. However, if a
4236 * facet changes the rule it resubmits into at some time in between
4237 * update_stats() runs, it is possible that statistics accrued to the
4238 * old rule will be incorrectly attributed to the new rule. This could be
4239 * avoided by calling update_stats() whenever rules are created or
4240 * deleted. However, the performance impact of making so many calls to the
4241 * datapath do not justify the benefit of having perfectly accurate statistics.
4243 * In addition, this function maintains per ofproto flow hit counts. The patch
4244 * port is not treated specially. e.g. A packet ingress from br0 patched into
4245 * br1 will increase the hit count of br0 by 1, however, does not affect
4246 * the hit or miss counts of br1.
4249 update_stats(struct dpif_backer *backer)
4251 const struct dpif_flow_stats *stats;
4252 struct dpif_flow_dump dump;
4253 const struct nlattr *key, *mask;
4254 size_t key_len, mask_len;
4256 dpif_flow_dump_start(&dump, backer->dpif);
4257 while (dpif_flow_dump_next(&dump, &key, &key_len,
4258 &mask, &mask_len, NULL, NULL, &stats)) {
4259 struct subfacet *subfacet;
4262 key_hash = odp_flow_key_hash(key, key_len);
4263 subfacet = subfacet_find(backer, key, key_len, key_hash);
4264 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4266 update_subfacet_stats(subfacet, stats);
4270 /* Stats are updated per-packet. */
4273 case SF_NOT_INSTALLED:
4275 delete_unexpected_flow(backer, key, key_len);
4280 dpif_flow_dump_done(&dump);
4282 update_moving_averages(backer);
4285 /* Calculates and returns the number of milliseconds of idle time after which
4286 * subfacets should expire from the datapath. When a subfacet expires, we fold
4287 * its statistics into its facet, and when a facet's last subfacet expires, we
4288 * fold its statistic into its rule. */
4290 subfacet_max_idle(const struct dpif_backer *backer)
4293 * Idle time histogram.
4295 * Most of the time a switch has a relatively small number of subfacets.
4296 * When this is the case we might as well keep statistics for all of them
4297 * in userspace and to cache them in the kernel datapath for performance as
4300 * As the number of subfacets increases, the memory required to maintain
4301 * statistics about them in userspace and in the kernel becomes
4302 * significant. However, with a large number of subfacets it is likely
4303 * that only a few of them are "heavy hitters" that consume a large amount
4304 * of bandwidth. At this point, only heavy hitters are worth caching in
4305 * the kernel and maintaining in userspaces; other subfacets we can
4308 * The technique used to compute the idle time is to build a histogram with
4309 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4310 * that is installed in the kernel gets dropped in the appropriate bucket.
4311 * After the histogram has been built, we compute the cutoff so that only
4312 * the most-recently-used 1% of subfacets (but at least
4313 * flow_eviction_threshold flows) are kept cached. At least
4314 * the most-recently-used bucket of subfacets is kept, so actually an
4315 * arbitrary number of subfacets can be kept in any given expiration run
4316 * (though the next run will delete most of those unless they receive
4319 * This requires a second pass through the subfacets, in addition to the
4320 * pass made by update_stats(), because the former function never looks at
4321 * uninstallable subfacets.
4323 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4324 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4325 int buckets[N_BUCKETS] = { 0 };
4326 int total, subtotal, bucket;
4327 struct subfacet *subfacet;
4331 total = hmap_count(&backer->subfacets);
4332 if (total <= flow_eviction_threshold) {
4333 return N_BUCKETS * BUCKET_WIDTH;
4336 /* Build histogram. */
4338 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4339 long long int idle = now - subfacet->used;
4340 int bucket = (idle <= 0 ? 0
4341 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4342 : (unsigned int) idle / BUCKET_WIDTH);
4346 /* Find the first bucket whose flows should be expired. */
4347 subtotal = bucket = 0;
4349 subtotal += buckets[bucket++];
4350 } while (bucket < N_BUCKETS &&
4351 subtotal < MAX(flow_eviction_threshold, total / 100));
4353 if (VLOG_IS_DBG_ENABLED()) {
4357 ds_put_cstr(&s, "keep");
4358 for (i = 0; i < N_BUCKETS; i++) {
4360 ds_put_cstr(&s, ", drop");
4363 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4366 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4370 return bucket * BUCKET_WIDTH;
4374 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4376 /* Cutoff time for most flows. */
4377 long long int normal_cutoff = time_msec() - dp_max_idle;
4379 /* We really want to keep flows for special protocols around, so use a more
4380 * conservative cutoff. */
4381 long long int special_cutoff = time_msec() - 10000;
4383 struct subfacet *subfacet, *next_subfacet;
4384 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4388 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4389 &backer->subfacets) {
4390 long long int cutoff;
4392 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4396 if (subfacet->used < cutoff) {
4397 if (subfacet->path != SF_NOT_INSTALLED) {
4398 batch[n_batch++] = subfacet;
4399 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4400 subfacet_destroy_batch(backer, batch, n_batch);
4404 subfacet_destroy(subfacet);
4410 subfacet_destroy_batch(backer, batch, n_batch);
4414 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4415 * then delete it entirely. */
4417 rule_expire(struct rule_dpif *rule)
4422 if (rule->up.pending) {
4423 /* We'll have to expire it later. */
4427 /* Has 'rule' expired? */
4429 if (rule->up.hard_timeout
4430 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4431 reason = OFPRR_HARD_TIMEOUT;
4432 } else if (rule->up.idle_timeout
4433 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4434 reason = OFPRR_IDLE_TIMEOUT;
4439 COVERAGE_INC(ofproto_dpif_expired);
4441 /* Get rid of the rule. */
4442 ofproto_rule_expire(&rule->up, reason);
4447 /* Creates and returns a new facet based on 'miss'.
4449 * The caller must already have determined that no facet with an identical
4450 * 'miss->flow' exists in 'miss->ofproto'.
4452 * 'rule' and 'xout' must have been created based on 'miss'.
4454 * 'facet'' statistics are initialized based on 'stats'.
4456 * The facet will initially have no subfacets. The caller should create (at
4457 * least) one subfacet with subfacet_create(). */
4458 static struct facet *
4459 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4460 struct xlate_out *xout, struct dpif_flow_stats *stats)
4462 struct ofproto_dpif *ofproto = miss->ofproto;
4463 struct facet *facet;
4466 facet = xzalloc(sizeof *facet);
4467 facet->ofproto = miss->ofproto;
4468 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4469 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4470 facet->tcp_flags = stats->tcp_flags;
4471 facet->used = stats->used;
4472 facet->flow = miss->flow;
4473 facet->learn_rl = time_msec() + 500;
4475 list_init(&facet->subfacets);
4476 netflow_flow_init(&facet->nf_flow);
4477 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4479 xlate_out_copy(&facet->xout, xout);
4481 match_init(&match, &facet->flow, &facet->xout.wc);
4482 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4483 classifier_insert(&ofproto->facets, &facet->cr);
4485 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4486 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4492 facet_free(struct facet *facet)
4495 xlate_out_uninit(&facet->xout);
4500 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4501 * 'packet', which arrived on 'in_port'. */
4503 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4504 const struct nlattr *odp_actions, size_t actions_len,
4505 struct ofpbuf *packet)
4507 struct odputil_keybuf keybuf;
4511 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4512 odp_flow_key_from_flow(&key, flow,
4513 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4515 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4516 odp_actions, actions_len, packet);
4520 /* Remove 'facet' from its ofproto and free up the associated memory:
4522 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4523 * rule's statistics, via subfacet_uninstall().
4525 * - Removes 'facet' from its rule and from ofproto->facets.
4528 facet_remove(struct facet *facet)
4530 struct subfacet *subfacet, *next_subfacet;
4532 ovs_assert(!list_is_empty(&facet->subfacets));
4534 /* First uninstall all of the subfacets to get final statistics. */
4535 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4536 subfacet_uninstall(subfacet);
4539 /* Flush the final stats to the rule.
4541 * This might require us to have at least one subfacet around so that we
4542 * can use its actions for accounting in facet_account(), which is why we
4543 * have uninstalled but not yet destroyed the subfacets. */
4544 facet_flush_stats(facet);
4546 /* Now we're really all done so destroy everything. */
4547 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4548 &facet->subfacets) {
4549 subfacet_destroy__(subfacet);
4551 classifier_remove(&facet->ofproto->facets, &facet->cr);
4552 cls_rule_destroy(&facet->cr);
4556 /* Feed information from 'facet' back into the learning table to keep it in
4557 * sync with what is actually flowing through the datapath. */
4559 facet_learn(struct facet *facet)
4561 long long int now = time_msec();
4563 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4567 facet->learn_rl = now + 500;
4569 if (!facet->xout.has_learn
4570 && !facet->xout.has_normal
4571 && (!facet->xout.has_fin_timeout
4572 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4576 facet_push_stats(facet, true);
4580 facet_account(struct facet *facet)
4582 const struct nlattr *a;
4587 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4590 n_bytes = facet->byte_count - facet->accounted_bytes;
4592 /* This loop feeds byte counters to bond_account() for rebalancing to use
4593 * as a basis. We also need to track the actual VLAN on which the packet
4594 * is going to be sent to ensure that it matches the one passed to
4595 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4598 * We use the actions from an arbitrary subfacet because they should all
4599 * be equally valid for our purpose. */
4600 vlan_tci = facet->flow.vlan_tci;
4601 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4602 facet->xout.odp_actions.size) {
4603 const struct ovs_action_push_vlan *vlan;
4604 struct ofport_dpif *port;
4606 switch (nl_attr_type(a)) {
4607 case OVS_ACTION_ATTR_OUTPUT:
4608 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4609 if (port && port->bundle && port->bundle->bond) {
4610 bond_account(port->bundle->bond, &facet->flow,
4611 vlan_tci_to_vid(vlan_tci), n_bytes);
4615 case OVS_ACTION_ATTR_POP_VLAN:
4616 vlan_tci = htons(0);
4619 case OVS_ACTION_ATTR_PUSH_VLAN:
4620 vlan = nl_attr_get(a);
4621 vlan_tci = vlan->vlan_tci;
4627 /* Returns true if the only action for 'facet' is to send to the controller.
4628 * (We don't report NetFlow expiration messages for such facets because they
4629 * are just part of the control logic for the network, not real traffic). */
4631 facet_is_controller_flow(struct facet *facet)
4634 struct ofproto_dpif *ofproto = facet->ofproto;
4635 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4637 const struct ofpact *ofpacts = rule->up.ofpacts;
4638 size_t ofpacts_len = rule->up.ofpacts_len;
4640 if (ofpacts_len > 0 &&
4641 ofpacts->type == OFPACT_CONTROLLER &&
4642 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4649 /* Folds all of 'facet''s statistics into its rule. Also updates the
4650 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4651 * 'facet''s statistics in the datapath should have been zeroed and folded into
4652 * its packet and byte counts before this function is called. */
4654 facet_flush_stats(struct facet *facet)
4656 struct ofproto_dpif *ofproto = facet->ofproto;
4657 struct subfacet *subfacet;
4659 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4660 ovs_assert(!subfacet->dp_byte_count);
4661 ovs_assert(!subfacet->dp_packet_count);
4664 facet_push_stats(facet, false);
4665 if (facet->accounted_bytes < facet->byte_count) {
4666 facet_account(facet);
4667 facet->accounted_bytes = facet->byte_count;
4670 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4671 struct ofexpired expired;
4672 expired.flow = facet->flow;
4673 expired.packet_count = facet->packet_count;
4674 expired.byte_count = facet->byte_count;
4675 expired.used = facet->used;
4676 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4679 /* Reset counters to prevent double counting if 'facet' ever gets
4681 facet_reset_counters(facet);
4683 netflow_flow_clear(&facet->nf_flow);
4684 facet->tcp_flags = 0;
4687 /* Searches 'ofproto''s table of facets for one which would be responsible for
4688 * 'flow'. Returns it if found, otherwise a null pointer.
4690 * The returned facet might need revalidation; use facet_lookup_valid()
4691 * instead if that is important. */
4692 static struct facet *
4693 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4695 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4696 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4699 /* Searches 'ofproto''s table of facets for one capable that covers
4700 * 'flow'. Returns it if found, otherwise a null pointer.
4702 * The returned facet is guaranteed to be valid. */
4703 static struct facet *
4704 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4706 struct facet *facet;
4708 facet = facet_find(ofproto, flow);
4710 && (ofproto->backer->need_revalidate
4711 || tag_set_intersects(&ofproto->backer->revalidate_set,
4713 && !facet_revalidate(facet)) {
4721 facet_check_consistency(struct facet *facet)
4723 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4725 struct xlate_out xout;
4726 struct xlate_in xin;
4728 struct rule_dpif *rule;
4731 /* Check the datapath actions for consistency. */
4732 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4733 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4734 xlate_actions(&xin, &xout);
4736 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4737 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4738 && facet->xout.slow == xout.slow
4739 && facet->fail_open == fail_open;
4740 if (!ok && !VLOG_DROP_WARN(&rl)) {
4741 struct ds s = DS_EMPTY_INITIALIZER;
4743 flow_format(&s, &facet->flow);
4744 ds_put_cstr(&s, ": inconsistency in facet");
4746 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4747 ds_put_cstr(&s, " (actions were: ");
4748 format_odp_actions(&s, facet->xout.odp_actions.data,
4749 facet->xout.odp_actions.size);
4750 ds_put_cstr(&s, ") (correct actions: ");
4751 format_odp_actions(&s, xout.odp_actions.data,
4752 xout.odp_actions.size);
4753 ds_put_char(&s, ')');
4756 if (facet->xout.slow != xout.slow) {
4757 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4760 if (facet->fail_open != fail_open) {
4761 ds_put_format(&s, " fail open incorrect. should be %s",
4762 fail_open ? "true" : "false");
4766 xlate_out_uninit(&xout);
4771 /* Re-searches the classifier for 'facet':
4773 * - If the rule found is different from 'facet''s current rule, moves
4774 * 'facet' to the new rule and recompiles its actions.
4776 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4777 * where it is and recompiles its actions anyway.
4779 * - If any of 'facet''s subfacets correspond to a new flow according to
4780 * ofproto_receive(), 'facet' is removed.
4782 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4784 facet_revalidate(struct facet *facet)
4786 struct ofproto_dpif *ofproto = facet->ofproto;
4787 struct rule_dpif *new_rule;
4788 struct subfacet *subfacet;
4789 struct flow_wildcards wc;
4790 struct xlate_out xout;
4791 struct xlate_in xin;
4793 COVERAGE_INC(facet_revalidate);
4795 /* Check that child subfacets still correspond to this facet. Tunnel
4796 * configuration changes could cause a subfacet's OpenFlow in_port to
4798 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4799 struct ofproto_dpif *recv_ofproto;
4800 struct flow recv_flow;
4803 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4804 subfacet->key_len, &recv_flow, NULL,
4805 &recv_ofproto, NULL);
4807 || recv_ofproto != ofproto
4808 || facet != facet_find(ofproto, &recv_flow)) {
4809 facet_remove(facet);
4814 flow_wildcards_init_catchall(&wc);
4815 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4817 /* Calculate new datapath actions.
4819 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4820 * emit a NetFlow expiration and, if so, we need to have the old state
4821 * around to properly compose it. */
4822 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4823 xlate_actions(&xin, &xout);
4824 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4826 /* A facet's slow path reason should only change under dramatic
4827 * circumstances. Rather than try to update everything, it's simpler to
4828 * remove the facet and start over.
4830 * More importantly, if a facet's wildcards change, it will be relatively
4831 * difficult to figure out if its subfacets still belong to it, and if not
4832 * which facet they may belong to. Again, to avoid the complexity, we
4833 * simply give up instead. */
4834 if (facet->xout.slow != xout.slow
4835 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4836 facet_remove(facet);
4837 xlate_out_uninit(&xout);
4841 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4842 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4843 if (subfacet->path == SF_FAST_PATH) {
4844 struct dpif_flow_stats stats;
4846 subfacet_install(subfacet, &xout.odp_actions, &stats);
4847 subfacet_update_stats(subfacet, &stats);
4851 facet_flush_stats(facet);
4853 ofpbuf_clear(&facet->xout.odp_actions);
4854 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4855 xout.odp_actions.size);
4858 /* Update 'facet' now that we've taken care of all the old state. */
4859 facet->xout.tags = xout.tags;
4860 facet->xout.slow = xout.slow;
4861 facet->xout.has_learn = xout.has_learn;
4862 facet->xout.has_normal = xout.has_normal;
4863 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4864 facet->xout.nf_output_iface = xout.nf_output_iface;
4865 facet->xout.mirrors = xout.mirrors;
4866 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4867 facet->used = MAX(facet->used, new_rule->up.created);
4868 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4870 xlate_out_uninit(&xout);
4875 facet_reset_counters(struct facet *facet)
4877 facet->packet_count = 0;
4878 facet->byte_count = 0;
4879 facet->prev_packet_count = 0;
4880 facet->prev_byte_count = 0;
4881 facet->accounted_bytes = 0;
4885 facet_push_stats(struct facet *facet, bool may_learn)
4887 struct dpif_flow_stats stats;
4889 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4890 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4891 ovs_assert(facet->used >= facet->prev_used);
4893 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4894 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4895 stats.used = facet->used;
4896 stats.tcp_flags = facet->tcp_flags;
4898 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4899 struct ofproto_dpif *ofproto = facet->ofproto;
4900 struct ofport_dpif *in_port;
4901 struct rule_dpif *rule;
4902 struct xlate_in xin;
4904 facet->prev_packet_count = facet->packet_count;
4905 facet->prev_byte_count = facet->byte_count;
4906 facet->prev_used = facet->used;
4908 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4909 if (in_port && in_port->is_tunnel) {
4910 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4913 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4914 rule_credit_stats(rule, &stats);
4915 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4917 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4918 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4919 stats.n_packets, stats.n_bytes);
4921 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4923 xin.resubmit_stats = &stats;
4924 xin.may_learn = may_learn;
4925 xlate_actions_for_side_effects(&xin);
4930 push_all_stats__(bool run_fast)
4932 static long long int rl = LLONG_MIN;
4933 struct ofproto_dpif *ofproto;
4935 if (time_msec() < rl) {
4939 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4940 struct cls_cursor cursor;
4941 struct facet *facet;
4943 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4944 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4945 facet_push_stats(facet, false);
4952 rl = time_msec() + 100;
4956 push_all_stats(void)
4958 push_all_stats__(true);
4962 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4964 rule->packet_count += stats->n_packets;
4965 rule->byte_count += stats->n_bytes;
4966 ofproto_rule_update_used(&rule->up, stats->used);
4971 static struct subfacet *
4972 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4973 size_t key_len, uint32_t key_hash)
4975 struct subfacet *subfacet;
4977 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4978 &backer->subfacets) {
4979 if (subfacet->key_len == key_len
4980 && !memcmp(key, subfacet->key, key_len)) {
4988 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4989 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4990 * existing subfacet if there is one, otherwise creates and returns a
4992 static struct subfacet *
4993 subfacet_create(struct facet *facet, struct flow_miss *miss,
4996 struct dpif_backer *backer = miss->ofproto->backer;
4997 enum odp_key_fitness key_fitness = miss->key_fitness;
4998 const struct nlattr *key = miss->key;
4999 size_t key_len = miss->key_len;
5001 struct subfacet *subfacet;
5003 key_hash = odp_flow_key_hash(key, key_len);
5005 if (list_is_empty(&facet->subfacets)) {
5006 subfacet = &facet->one_subfacet;
5008 subfacet = subfacet_find(backer, key, key_len, key_hash);
5010 if (subfacet->facet == facet) {
5014 /* This shouldn't happen. */
5015 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5016 subfacet_destroy(subfacet);
5019 subfacet = xmalloc(sizeof *subfacet);
5022 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
5023 list_push_back(&facet->subfacets, &subfacet->list_node);
5024 subfacet->facet = facet;
5025 subfacet->key_fitness = key_fitness;
5026 subfacet->key = xmemdup(key, key_len);
5027 subfacet->key_len = key_len;
5028 subfacet->used = now;
5029 subfacet->created = now;
5030 subfacet->dp_packet_count = 0;
5031 subfacet->dp_byte_count = 0;
5032 subfacet->path = SF_NOT_INSTALLED;
5033 subfacet->backer = backer;
5035 backer->subfacet_add_count++;
5039 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5040 * its facet within 'ofproto', and frees it. */
5042 subfacet_destroy__(struct subfacet *subfacet)
5044 struct facet *facet = subfacet->facet;
5045 struct ofproto_dpif *ofproto = facet->ofproto;
5047 /* Update ofproto stats before uninstall the subfacet. */
5048 ofproto->backer->subfacet_del_count++;
5050 subfacet_uninstall(subfacet);
5051 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
5052 list_remove(&subfacet->list_node);
5053 free(subfacet->key);
5054 if (subfacet != &facet->one_subfacet) {
5059 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5060 * last remaining subfacet in its facet destroys the facet too. */
5062 subfacet_destroy(struct subfacet *subfacet)
5064 struct facet *facet = subfacet->facet;
5066 if (list_is_singleton(&facet->subfacets)) {
5067 /* facet_remove() needs at least one subfacet (it will remove it). */
5068 facet_remove(facet);
5070 subfacet_destroy__(subfacet);
5075 subfacet_destroy_batch(struct dpif_backer *backer,
5076 struct subfacet **subfacets, int n)
5078 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5079 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5080 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5083 for (i = 0; i < n; i++) {
5084 ops[i].type = DPIF_OP_FLOW_DEL;
5085 ops[i].u.flow_del.key = subfacets[i]->key;
5086 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5087 ops[i].u.flow_del.stats = &stats[i];
5091 dpif_operate(backer->dpif, opsp, n);
5092 for (i = 0; i < n; i++) {
5093 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5094 subfacets[i]->path = SF_NOT_INSTALLED;
5095 subfacet_destroy(subfacets[i]);
5100 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5101 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5102 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5103 * since 'subfacet' was last updated.
5105 * Returns 0 if successful, otherwise a positive errno value. */
5107 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5108 struct dpif_flow_stats *stats)
5110 struct facet *facet = subfacet->facet;
5111 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5112 const struct nlattr *actions = odp_actions->data;
5113 size_t actions_len = odp_actions->size;
5114 struct odputil_keybuf maskbuf;
5117 uint64_t slow_path_stub[128 / 8];
5118 enum dpif_flow_put_flags flags;
5121 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5124 flags |= DPIF_FP_ZERO_STATS;
5127 if (path == SF_SLOW_PATH) {
5128 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5129 slow_path_stub, sizeof slow_path_stub,
5130 &actions, &actions_len);
5133 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5134 if (enable_megaflows) {
5135 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5136 &facet->flow, UINT32_MAX);
5139 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5140 subfacet->key_len, mask.data, mask.size,
5141 actions, actions_len, stats);
5144 subfacet_reset_dp_stats(subfacet, stats);
5148 COVERAGE_INC(subfacet_install_fail);
5150 subfacet->path = path;
5155 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5157 subfacet_uninstall(struct subfacet *subfacet)
5159 if (subfacet->path != SF_NOT_INSTALLED) {
5160 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5161 struct dpif_flow_stats stats;
5164 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5165 subfacet->key_len, &stats);
5166 subfacet_reset_dp_stats(subfacet, &stats);
5168 subfacet_update_stats(subfacet, &stats);
5170 subfacet->path = SF_NOT_INSTALLED;
5172 ovs_assert(subfacet->dp_packet_count == 0);
5173 ovs_assert(subfacet->dp_byte_count == 0);
5177 /* Resets 'subfacet''s datapath statistics counters. This should be called
5178 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5179 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5180 * was reset in the datapath. 'stats' will be modified to include only
5181 * statistics new since 'subfacet' was last updated. */
5183 subfacet_reset_dp_stats(struct subfacet *subfacet,
5184 struct dpif_flow_stats *stats)
5187 && subfacet->dp_packet_count <= stats->n_packets
5188 && subfacet->dp_byte_count <= stats->n_bytes) {
5189 stats->n_packets -= subfacet->dp_packet_count;
5190 stats->n_bytes -= subfacet->dp_byte_count;
5193 subfacet->dp_packet_count = 0;
5194 subfacet->dp_byte_count = 0;
5197 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5199 * Because of the meaning of a subfacet's counters, it only makes sense to do
5200 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5201 * represents a packet that was sent by hand or if it represents statistics
5202 * that have been cleared out of the datapath. */
5204 subfacet_update_stats(struct subfacet *subfacet,
5205 const struct dpif_flow_stats *stats)
5207 if (stats->n_packets || stats->used > subfacet->used) {
5208 struct facet *facet = subfacet->facet;
5210 subfacet->used = MAX(subfacet->used, stats->used);
5211 facet->used = MAX(facet->used, stats->used);
5212 facet->packet_count += stats->n_packets;
5213 facet->byte_count += stats->n_bytes;
5214 facet->tcp_flags |= stats->tcp_flags;
5220 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5221 * the fields that were relevant as part of the lookup. */
5222 static struct rule_dpif *
5223 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5224 struct flow_wildcards *wc)
5226 struct rule_dpif *rule;
5228 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5233 return rule_dpif_miss_rule(ofproto, flow);
5237 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5238 const struct flow *flow, struct flow_wildcards *wc,
5241 struct cls_rule *cls_rule;
5242 struct classifier *cls;
5245 if (table_id >= N_TABLES) {
5250 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5251 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5254 cls = &ofproto->up.tables[table_id].cls;
5255 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5256 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5257 /* We must pretend that transport ports are unavailable. */
5258 struct flow ofpc_normal_flow = *flow;
5259 ofpc_normal_flow.tp_src = htons(0);
5260 ofpc_normal_flow.tp_dst = htons(0);
5261 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5262 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5263 cls_rule = &ofproto->drop_frags_rule->up.cr;
5265 flow_wildcards_init_exact(wc);
5268 cls_rule = classifier_lookup(cls, flow, wc);
5270 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5274 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5276 struct ofport_dpif *port;
5278 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5280 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5281 flow->in_port.ofp_port);
5282 return ofproto->miss_rule;
5285 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5286 return ofproto->no_packet_in_rule;
5288 return ofproto->miss_rule;
5292 complete_operation(struct rule_dpif *rule)
5294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5296 rule_invalidate(rule);
5298 struct dpif_completion *c = xmalloc(sizeof *c);
5299 c->op = rule->up.pending;
5300 list_push_back(&ofproto->completions, &c->list_node);
5302 ofoperation_complete(rule->up.pending, 0);
5306 static struct rule *
5309 struct rule_dpif *rule = xmalloc(sizeof *rule);
5314 rule_dealloc(struct rule *rule_)
5316 struct rule_dpif *rule = rule_dpif_cast(rule_);
5321 rule_construct(struct rule *rule_)
5323 struct rule_dpif *rule = rule_dpif_cast(rule_);
5324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5325 struct rule_dpif *victim;
5328 rule->packet_count = 0;
5329 rule->byte_count = 0;
5331 table_id = rule->up.table_id;
5332 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5334 rule->tag = victim->tag;
5335 } else if (table_id == 0) {
5340 miniflow_expand(&rule->up.cr.match.flow, &flow);
5341 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5342 ofproto->tables[table_id].basis);
5345 complete_operation(rule);
5350 rule_destruct(struct rule *rule)
5352 complete_operation(rule_dpif_cast(rule));
5356 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5358 struct rule_dpif *rule = rule_dpif_cast(rule_);
5360 /* push_all_stats() can handle flow misses which, when using the learn
5361 * action, can cause rules to be added and deleted. This can corrupt our
5362 * caller's datastructures which assume that rule_get_stats() doesn't have
5363 * an impact on the flow table. To be safe, we disable miss handling. */
5364 push_all_stats__(false);
5366 /* Start from historical data for 'rule' itself that are no longer tracked
5367 * in facets. This counts, for example, facets that have expired. */
5368 *packets = rule->packet_count;
5369 *bytes = rule->byte_count;
5373 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5374 struct ofpbuf *packet)
5376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5377 struct dpif_flow_stats stats;
5378 struct xlate_out xout;
5379 struct xlate_in xin;
5381 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5382 rule_credit_stats(rule, &stats);
5384 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5385 xin.resubmit_stats = &stats;
5386 xlate_actions(&xin, &xout);
5388 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5389 xout.odp_actions.size, packet);
5391 xlate_out_uninit(&xout);
5395 rule_execute(struct rule *rule, const struct flow *flow,
5396 struct ofpbuf *packet)
5398 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5399 ofpbuf_delete(packet);
5404 rule_modify_actions(struct rule *rule_)
5406 struct rule_dpif *rule = rule_dpif_cast(rule_);
5408 complete_operation(rule);
5411 /* Sends 'packet' out 'ofport'.
5412 * May modify 'packet'.
5413 * Returns 0 if successful, otherwise a positive errno value. */
5415 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5418 uint64_t odp_actions_stub[1024 / 8];
5419 struct ofpbuf key, odp_actions;
5420 struct dpif_flow_stats stats;
5421 struct odputil_keybuf keybuf;
5422 struct ofpact_output output;
5423 struct xlate_out xout;
5424 struct xlate_in xin;
5426 union flow_in_port in_port_;
5429 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5430 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5432 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5433 in_port_.ofp_port = OFPP_NONE;
5434 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5435 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5437 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5439 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5440 output.port = ofport->up.ofp_port;
5443 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5444 xin.ofpacts_len = sizeof output;
5445 xin.ofpacts = &output.ofpact;
5446 xin.resubmit_stats = &stats;
5447 xlate_actions(&xin, &xout);
5449 error = dpif_execute(ofproto->backer->dpif,
5451 xout.odp_actions.data, xout.odp_actions.size,
5453 xlate_out_uninit(&xout);
5456 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5457 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5458 ovs_strerror(error));
5461 ofproto->stats.tx_packets++;
5462 ofproto->stats.tx_bytes += packet->size;
5466 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5467 * The action will state 'slow' as the reason that the action is in the slow
5468 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5469 * dump-flows" output to see why a flow is in the slow path.)
5471 * The 'stub_size' bytes in 'stub' will be used to store the action.
5472 * 'stub_size' must be large enough for the action.
5474 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5477 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5478 enum slow_path_reason slow,
5479 uint64_t *stub, size_t stub_size,
5480 const struct nlattr **actionsp, size_t *actions_lenp)
5482 union user_action_cookie cookie;
5485 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5486 cookie.slow_path.unused = 0;
5487 cookie.slow_path.reason = slow;
5489 ofpbuf_use_stack(&buf, stub, stub_size);
5490 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5491 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5493 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5495 put_userspace_action(ofproto, &buf, flow, &cookie,
5496 sizeof cookie.slow_path);
5498 *actionsp = buf.data;
5499 *actions_lenp = buf.size;
5503 put_userspace_action(const struct ofproto_dpif *ofproto,
5504 struct ofpbuf *odp_actions,
5505 const struct flow *flow,
5506 const union user_action_cookie *cookie,
5507 const size_t cookie_size)
5511 pid = dpif_port_get_pid(ofproto->backer->dpif,
5512 ofp_port_to_odp_port(ofproto,
5513 flow->in_port.ofp_port));
5515 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5519 calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
5520 uint8_t table_id, struct rule_dpif *rule)
5522 if (table_id > 0 && table_id < N_TABLES) {
5523 struct table_dpif *table = &ofproto->tables[table_id];
5524 if (table->other_table) {
5525 return (rule && rule->tag
5527 : rule_calculate_tag(flow, &table->other_table->mask,
5535 /* Optimized flow revalidation.
5537 * It's a difficult problem, in general, to tell which facets need to have
5538 * their actions recalculated whenever the OpenFlow flow table changes. We
5539 * don't try to solve that general problem: for most kinds of OpenFlow flow
5540 * table changes, we recalculate the actions for every facet. This is
5541 * relatively expensive, but it's good enough if the OpenFlow flow table
5542 * doesn't change very often.
5544 * However, we can expect one particular kind of OpenFlow flow table change to
5545 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5546 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5547 * table, we add a special case that applies to flow tables in which every rule
5548 * has the same form (that is, the same wildcards), except that the table is
5549 * also allowed to have a single "catch-all" flow that matches all packets. We
5550 * optimize this case by tagging all of the facets that resubmit into the table
5551 * and invalidating the same tag whenever a flow changes in that table. The
5552 * end result is that we revalidate just the facets that need it (and sometimes
5553 * a few more, but not all of the facets or even all of the facets that
5554 * resubmit to the table modified by MAC learning). */
5556 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5557 * into an OpenFlow table with the given 'basis'. */
5559 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5562 if (minimask_is_catchall(mask)) {
5565 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5566 return tag_create_deterministic(hash);
5570 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5571 * taggability of that table.
5573 * This function must be called after *each* change to a flow table. If you
5574 * skip calling it on some changes then the pointer comparisons at the end can
5575 * be invalid if you get unlucky. For example, if a flow removal causes a
5576 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5577 * different wildcards to be created with the same address, then this function
5578 * will incorrectly skip revalidation. */
5580 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5582 struct table_dpif *table = &ofproto->tables[table_id];
5583 const struct oftable *oftable = &ofproto->up.tables[table_id];
5584 struct cls_table *catchall, *other;
5585 struct cls_table *t;
5587 catchall = other = NULL;
5589 switch (hmap_count(&oftable->cls.tables)) {
5591 /* We could tag this OpenFlow table but it would make the logic a
5592 * little harder and it's a corner case that doesn't seem worth it
5598 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5599 if (cls_table_is_catchall(t)) {
5601 } else if (!other) {
5604 /* Indicate that we can't tag this by setting both tables to
5605 * NULL. (We know that 'catchall' is already NULL.) */
5612 /* Can't tag this table. */
5616 if (table->catchall_table != catchall || table->other_table != other) {
5617 table->catchall_table = catchall;
5618 table->other_table = other;
5619 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5623 /* Given 'rule' that has changed in some way (either it is a rule being
5624 * inserted, a rule being deleted, or a rule whose actions are being
5625 * modified), marks facets for revalidation to ensure that packets will be
5626 * forwarded correctly according to the new state of the flow table.
5628 * This function must be called after *each* change to a flow table. See
5629 * the comment on table_update_taggable() for more information. */
5631 rule_invalidate(const struct rule_dpif *rule)
5633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5635 table_update_taggable(ofproto, rule->up.table_id);
5637 if (!ofproto->backer->need_revalidate) {
5638 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5640 if (table->other_table && rule->tag) {
5641 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5643 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5649 set_frag_handling(struct ofproto *ofproto_,
5650 enum ofp_config_flags frag_handling)
5652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5653 if (frag_handling != OFPC_FRAG_REASM) {
5654 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5662 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5663 const struct flow *flow,
5664 const struct ofpact *ofpacts, size_t ofpacts_len)
5666 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5667 struct odputil_keybuf keybuf;
5668 struct dpif_flow_stats stats;
5669 struct xlate_out xout;
5670 struct xlate_in xin;
5674 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5675 odp_flow_key_from_flow(&key, flow,
5676 ofp_port_to_odp_port(ofproto,
5677 flow->in_port.ofp_port));
5679 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5681 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5682 xin.resubmit_stats = &stats;
5683 xin.ofpacts_len = ofpacts_len;
5684 xin.ofpacts = ofpacts;
5686 xlate_actions(&xin, &xout);
5687 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5688 xout.odp_actions.data, xout.odp_actions.size, packet);
5689 xlate_out_uninit(&xout);
5697 set_netflow(struct ofproto *ofproto_,
5698 const struct netflow_options *netflow_options)
5700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5702 if (netflow_options) {
5703 if (!ofproto->netflow) {
5704 ofproto->netflow = netflow_create();
5705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5707 return netflow_set_options(ofproto->netflow, netflow_options);
5708 } else if (ofproto->netflow) {
5709 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5710 netflow_destroy(ofproto->netflow);
5711 ofproto->netflow = NULL;
5718 get_netflow_ids(const struct ofproto *ofproto_,
5719 uint8_t *engine_type, uint8_t *engine_id)
5721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5723 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5727 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5729 if (!facet_is_controller_flow(facet) &&
5730 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5731 struct subfacet *subfacet;
5732 struct ofexpired expired;
5734 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5735 if (subfacet->path == SF_FAST_PATH) {
5736 struct dpif_flow_stats stats;
5738 subfacet_install(subfacet, &facet->xout.odp_actions,
5740 subfacet_update_stats(subfacet, &stats);
5744 expired.flow = facet->flow;
5745 expired.packet_count = facet->packet_count;
5746 expired.byte_count = facet->byte_count;
5747 expired.used = facet->used;
5748 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5753 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5755 struct cls_cursor cursor;
5756 struct facet *facet;
5758 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5759 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5760 send_active_timeout(ofproto, facet);
5764 static struct ofproto_dpif *
5765 ofproto_dpif_lookup(const char *name)
5767 struct ofproto_dpif *ofproto;
5769 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5770 hash_string(name, 0), &all_ofproto_dpifs) {
5771 if (!strcmp(ofproto->up.name, name)) {
5779 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5780 const char *argv[], void *aux OVS_UNUSED)
5782 struct ofproto_dpif *ofproto;
5785 ofproto = ofproto_dpif_lookup(argv[1]);
5787 unixctl_command_reply_error(conn, "no such bridge");
5790 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5791 mac_learning_flush(ofproto->ml);
5792 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5794 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5795 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5796 mac_learning_flush(ofproto->ml);
5797 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5801 unixctl_command_reply(conn, "table successfully flushed");
5804 static struct ofport_dpif *
5805 ofbundle_get_a_port(const struct ofbundle *bundle)
5807 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5812 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5813 const char *argv[], void *aux OVS_UNUSED)
5815 struct ds ds = DS_EMPTY_INITIALIZER;
5816 const struct ofproto_dpif *ofproto;
5817 const struct mac_entry *e;
5819 ofproto = ofproto_dpif_lookup(argv[1]);
5821 unixctl_command_reply_error(conn, "no such bridge");
5825 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5826 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5827 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5828 struct ofbundle *bundle = e->port.p;
5829 char name[OFP_MAX_PORT_NAME_LEN];
5831 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5833 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5834 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5835 mac_entry_age(ofproto->ml, e));
5837 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5838 unixctl_command_reply(conn, ds_cstr(&ds));
5843 struct xlate_out xout;
5844 struct xlate_in xin;
5850 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5852 ds_put_char_multiple(result, '\t', level);
5854 ds_put_cstr(result, "No match\n");
5858 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5859 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5860 cls_rule_format(&rule->up.cr, result);
5861 ds_put_char(result, '\n');
5863 ds_put_char_multiple(result, '\t', level);
5864 ds_put_cstr(result, "OpenFlow ");
5865 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5866 ds_put_char(result, '\n');
5870 trace_format_flow(struct ds *result, int level, const char *title,
5871 struct trace_ctx *trace)
5873 ds_put_char_multiple(result, '\t', level);
5874 ds_put_format(result, "%s: ", title);
5875 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5876 ds_put_cstr(result, "unchanged");
5878 flow_format(result, &trace->xin.flow);
5879 trace->flow = trace->xin.flow;
5881 ds_put_char(result, '\n');
5885 trace_format_regs(struct ds *result, int level, const char *title,
5886 struct trace_ctx *trace)
5890 ds_put_char_multiple(result, '\t', level);
5891 ds_put_format(result, "%s:", title);
5892 for (i = 0; i < FLOW_N_REGS; i++) {
5893 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5895 ds_put_char(result, '\n');
5899 trace_format_odp(struct ds *result, int level, const char *title,
5900 struct trace_ctx *trace)
5902 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5904 ds_put_char_multiple(result, '\t', level);
5905 ds_put_format(result, "%s: ", title);
5906 format_odp_actions(result, odp_actions->data, odp_actions->size);
5907 ds_put_char(result, '\n');
5911 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5913 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5914 struct ds *result = trace->result;
5916 ds_put_char(result, '\n');
5917 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5918 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5919 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5920 trace_format_rule(result, recurse + 1, rule);
5924 trace_report(struct xlate_in *xin, const char *s, int recurse)
5926 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5927 struct ds *result = trace->result;
5929 ds_put_char_multiple(result, '\t', recurse);
5930 ds_put_cstr(result, s);
5931 ds_put_char(result, '\n');
5935 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5936 void *aux OVS_UNUSED)
5938 const struct dpif_backer *backer;
5939 struct ofproto_dpif *ofproto;
5940 struct ofpbuf odp_key, odp_mask;
5941 struct ofpbuf *packet;
5949 ofpbuf_init(&odp_key, 0);
5950 ofpbuf_init(&odp_mask, 0);
5952 /* Handle "-generate" or a hex string as the last argument. */
5953 if (!strcmp(argv[argc - 1], "-generate")) {
5954 packet = ofpbuf_new(0);
5957 const char *error = eth_from_hex(argv[argc - 1], &packet);
5960 } else if (argc == 4) {
5961 /* The 3-argument form must end in "-generate' or a hex string. */
5962 unixctl_command_reply_error(conn, error);
5967 /* Parse the flow and determine whether a datapath or
5968 * bridge is specified. If function odp_flow_key_from_string()
5969 * returns 0, the flow is a odp_flow. If function
5970 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5971 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5972 /* If the odp_flow is the second argument,
5973 * the datapath name is the first argument. */
5975 const char *dp_type;
5976 if (!strncmp(argv[1], "ovs-", 4)) {
5977 dp_type = argv[1] + 4;
5981 backer = shash_find_data(&all_dpif_backers, dp_type);
5983 unixctl_command_reply_error(conn, "Cannot find datapath "
5988 /* No datapath name specified, so there should be only one
5990 struct shash_node *node;
5991 if (shash_count(&all_dpif_backers) != 1) {
5992 unixctl_command_reply_error(conn, "Must specify datapath "
5993 "name, there is more than one type of datapath");
5996 node = shash_first(&all_dpif_backers);
5997 backer = node->data;
6000 /* Extract the ofproto_dpif object from the ofproto_receive()
6002 if (ofproto_receive(backer, NULL, odp_key.data,
6003 odp_key.size, &flow, NULL, &ofproto, NULL)) {
6004 unixctl_command_reply_error(conn, "Invalid datapath flow");
6007 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
6008 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
6010 unixctl_command_reply_error(conn, "Must specify bridge name");
6014 ofproto = ofproto_dpif_lookup(argv[1]);
6016 unixctl_command_reply_error(conn, "Unknown bridge name");
6020 unixctl_command_reply_error(conn, "Bad flow syntax");
6024 /* Generate a packet, if requested. */
6026 if (!packet->size) {
6027 flow_compose(packet, &flow);
6029 union flow_in_port in_port_;
6031 in_port_ = flow.in_port;
6032 ds_put_cstr(&result, "Packet: ");
6033 s = ofp_packet_to_string(packet->data, packet->size);
6034 ds_put_cstr(&result, s);
6037 /* Use the metadata from the flow and the packet argument
6038 * to reconstruct the flow. */
6039 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
6044 ofproto_trace(ofproto, &flow, packet, &result);
6045 unixctl_command_reply(conn, ds_cstr(&result));
6048 ds_destroy(&result);
6049 ofpbuf_delete(packet);
6050 ofpbuf_uninit(&odp_key);
6051 ofpbuf_uninit(&odp_mask);
6055 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6056 const struct ofpbuf *packet, struct ds *ds)
6058 struct rule_dpif *rule;
6060 ds_put_cstr(ds, "Flow: ");
6061 flow_format(ds, flow);
6062 ds_put_char(ds, '\n');
6064 rule = rule_dpif_lookup(ofproto, flow, NULL);
6066 trace_format_rule(ds, 0, rule);
6067 if (rule == ofproto->miss_rule) {
6068 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6069 } else if (rule == ofproto->no_packet_in_rule) {
6070 ds_put_cstr(ds, "\nNo match, packets dropped because "
6071 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6072 } else if (rule == ofproto->drop_frags_rule) {
6073 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
6074 "and the fragment handling mode is \"drop\".\n");
6078 uint64_t odp_actions_stub[1024 / 8];
6079 struct ofpbuf odp_actions;
6080 struct trace_ctx trace;
6084 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6087 ofpbuf_use_stub(&odp_actions,
6088 odp_actions_stub, sizeof odp_actions_stub);
6089 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
6090 trace.xin.resubmit_hook = trace_resubmit;
6091 trace.xin.report_hook = trace_report;
6093 xlate_actions(&trace.xin, &trace.xout);
6095 ds_put_char(ds, '\n');
6096 trace_format_flow(ds, 0, "Final flow", &trace);
6098 match_init(&match, flow, &trace.xout.wc);
6099 ds_put_cstr(ds, "Relevant fields: ");
6100 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
6101 ds_put_char(ds, '\n');
6103 ds_put_cstr(ds, "Datapath actions: ");
6104 format_odp_actions(ds, trace.xout.odp_actions.data,
6105 trace.xout.odp_actions.size);
6107 if (trace.xout.slow) {
6108 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6109 "slow path because it:");
6110 switch (trace.xout.slow) {
6112 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6115 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6118 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6121 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
6123 case SLOW_CONTROLLER:
6124 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6125 "to the OpenFlow controller.");
6132 xlate_out_uninit(&trace.xout);
6137 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6138 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6141 unixctl_command_reply(conn, NULL);
6145 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6146 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6149 unixctl_command_reply(conn, NULL);
6152 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6153 * 'reply' describing the results. */
6155 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6157 struct cls_cursor cursor;
6158 struct facet *facet;
6162 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6163 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6164 if (!facet_check_consistency(facet)) {
6169 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6173 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6174 ofproto->up.name, errors);
6176 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6181 ofproto_dpif_self_check(struct unixctl_conn *conn,
6182 int argc, const char *argv[], void *aux OVS_UNUSED)
6184 struct ds reply = DS_EMPTY_INITIALIZER;
6185 struct ofproto_dpif *ofproto;
6188 ofproto = ofproto_dpif_lookup(argv[1]);
6190 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6191 "ofproto/list for help)");
6194 ofproto_dpif_self_check__(ofproto, &reply);
6196 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6197 ofproto_dpif_self_check__(ofproto, &reply);
6201 unixctl_command_reply(conn, ds_cstr(&reply));
6205 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6206 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6207 * to destroy 'ofproto_shash' and free the returned value. */
6208 static const struct shash_node **
6209 get_ofprotos(struct shash *ofproto_shash)
6211 const struct ofproto_dpif *ofproto;
6213 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6214 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6215 shash_add_nocopy(ofproto_shash, name, ofproto);
6218 return shash_sort(ofproto_shash);
6222 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6223 const char *argv[] OVS_UNUSED,
6224 void *aux OVS_UNUSED)
6226 struct ds ds = DS_EMPTY_INITIALIZER;
6227 struct shash ofproto_shash;
6228 const struct shash_node **sorted_ofprotos;
6231 shash_init(&ofproto_shash);
6232 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6233 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6234 const struct shash_node *node = sorted_ofprotos[i];
6235 ds_put_format(&ds, "%s\n", node->name);
6238 shash_destroy(&ofproto_shash);
6239 free(sorted_ofprotos);
6241 unixctl_command_reply(conn, ds_cstr(&ds));
6246 show_dp_rates(struct ds *ds, const char *heading,
6247 const struct avg_subfacet_rates *rates)
6249 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6250 heading, rates->add_rate, rates->del_rate);
6254 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6256 const struct shash_node **ofprotos;
6257 struct ofproto_dpif *ofproto;
6258 struct shash ofproto_shash;
6259 uint64_t n_hit, n_missed;
6260 long long int minutes;
6263 n_hit = n_missed = 0;
6264 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6265 if (ofproto->backer == backer) {
6266 n_missed += ofproto->n_missed;
6267 n_hit += ofproto->n_hit;
6271 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6272 dpif_name(backer->dpif), n_hit, n_missed);
6273 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6274 " life span: %lldms\n", hmap_count(&backer->subfacets),
6275 backer->avg_n_subfacet, backer->max_n_subfacet,
6276 backer->avg_subfacet_life);
6278 minutes = (time_msec() - backer->created) / (1000 * 60);
6279 if (minutes >= 60) {
6280 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6282 if (minutes >= 60 * 24) {
6283 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6285 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6287 shash_init(&ofproto_shash);
6288 ofprotos = get_ofprotos(&ofproto_shash);
6289 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6290 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6291 const struct shash_node **ports;
6294 if (ofproto->backer != backer) {
6298 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6299 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6301 ports = shash_sort(&ofproto->up.port_by_name);
6302 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6303 const struct shash_node *node = ports[j];
6304 struct ofport *ofport = node->data;
6306 odp_port_t odp_port;
6308 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6311 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6312 if (odp_port != ODPP_NONE) {
6313 ds_put_format(ds, "%"PRIu32":", odp_port);
6315 ds_put_cstr(ds, "none:");
6318 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6321 if (!netdev_get_config(ofport->netdev, &config)) {
6322 const struct smap_node **nodes;
6325 nodes = smap_sort(&config);
6326 for (i = 0; i < smap_count(&config); i++) {
6327 const struct smap_node *node = nodes[i];
6328 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6329 node->key, node->value);
6333 smap_destroy(&config);
6335 ds_put_char(ds, ')');
6336 ds_put_char(ds, '\n');
6340 shash_destroy(&ofproto_shash);
6345 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6346 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6348 struct ds ds = DS_EMPTY_INITIALIZER;
6349 const struct shash_node **backers;
6352 backers = shash_sort(&all_dpif_backers);
6353 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6354 dpif_show_backer(backers[i]->data, &ds);
6358 unixctl_command_reply(conn, ds_cstr(&ds));
6362 /* Dump the megaflow (facet) cache. This is useful to check the
6363 * correctness of flow wildcarding, since the same mechanism is used for
6364 * both xlate caching and kernel wildcarding.
6366 * It's important to note that in the output the flow description uses
6367 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6369 * This command is only needed for advanced debugging, so it's not
6370 * documented in the man page. */
6372 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6373 int argc OVS_UNUSED, const char *argv[],
6374 void *aux OVS_UNUSED)
6376 struct ds ds = DS_EMPTY_INITIALIZER;
6377 const struct ofproto_dpif *ofproto;
6378 long long int now = time_msec();
6379 struct cls_cursor cursor;
6380 struct facet *facet;
6382 ofproto = ofproto_dpif_lookup(argv[1]);
6384 unixctl_command_reply_error(conn, "no such bridge");
6388 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6389 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6390 cls_rule_format(&facet->cr, &ds);
6391 ds_put_cstr(&ds, ", ");
6392 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6393 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6394 ds_put_cstr(&ds, "Datapath actions: ");
6395 if (facet->xout.slow) {
6396 uint64_t slow_path_stub[128 / 8];
6397 const struct nlattr *actions;
6400 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6401 slow_path_stub, sizeof slow_path_stub,
6402 &actions, &actions_len);
6403 format_odp_actions(&ds, actions, actions_len);
6405 format_odp_actions(&ds, facet->xout.odp_actions.data,
6406 facet->xout.odp_actions.size);
6408 ds_put_cstr(&ds, "\n");
6411 ds_chomp(&ds, '\n');
6412 unixctl_command_reply(conn, ds_cstr(&ds));
6416 /* Disable using the megaflows.
6418 * This command is only needed for advanced debugging, so it's not
6419 * documented in the man page. */
6421 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6422 int argc OVS_UNUSED,
6423 const char *argv[] OVS_UNUSED,
6424 void *aux OVS_UNUSED)
6426 struct ofproto_dpif *ofproto;
6428 enable_megaflows = false;
6430 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6431 flush(&ofproto->up);
6434 unixctl_command_reply(conn, "megaflows disabled");
6437 /* Re-enable using megaflows.
6439 * This command is only needed for advanced debugging, so it's not
6440 * documented in the man page. */
6442 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6443 int argc OVS_UNUSED,
6444 const char *argv[] OVS_UNUSED,
6445 void *aux OVS_UNUSED)
6447 struct ofproto_dpif *ofproto;
6449 enable_megaflows = true;
6451 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6452 flush(&ofproto->up);
6455 unixctl_command_reply(conn, "megaflows enabled");
6459 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6460 int argc OVS_UNUSED, const char *argv[],
6461 void *aux OVS_UNUSED)
6463 struct ds ds = DS_EMPTY_INITIALIZER;
6464 const struct ofproto_dpif *ofproto;
6465 struct subfacet *subfacet;
6467 ofproto = ofproto_dpif_lookup(argv[1]);
6469 unixctl_command_reply_error(conn, "no such bridge");
6473 update_stats(ofproto->backer);
6475 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6476 struct facet *facet = subfacet->facet;
6477 struct odputil_keybuf maskbuf;
6480 if (facet->ofproto != ofproto) {
6484 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6485 if (enable_megaflows) {
6486 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6487 &facet->flow, UINT32_MAX);
6490 odp_flow_format(subfacet->key, subfacet->key_len,
6491 mask.data, mask.size, &ds);
6493 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6494 subfacet->dp_packet_count, subfacet->dp_byte_count);
6495 if (subfacet->used) {
6496 ds_put_format(&ds, "%.3fs",
6497 (time_msec() - subfacet->used) / 1000.0);
6499 ds_put_format(&ds, "never");
6501 if (subfacet->facet->tcp_flags) {
6502 ds_put_cstr(&ds, ", flags:");
6503 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6506 ds_put_cstr(&ds, ", actions:");
6507 if (facet->xout.slow) {
6508 uint64_t slow_path_stub[128 / 8];
6509 const struct nlattr *actions;
6512 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6513 slow_path_stub, sizeof slow_path_stub,
6514 &actions, &actions_len);
6515 format_odp_actions(&ds, actions, actions_len);
6517 format_odp_actions(&ds, facet->xout.odp_actions.data,
6518 facet->xout.odp_actions.size);
6520 ds_put_char(&ds, '\n');
6523 unixctl_command_reply(conn, ds_cstr(&ds));
6528 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6529 int argc OVS_UNUSED, const char *argv[],
6530 void *aux OVS_UNUSED)
6532 struct ds ds = DS_EMPTY_INITIALIZER;
6533 struct ofproto_dpif *ofproto;
6535 ofproto = ofproto_dpif_lookup(argv[1]);
6537 unixctl_command_reply_error(conn, "no such bridge");
6541 flush(&ofproto->up);
6543 unixctl_command_reply(conn, ds_cstr(&ds));
6548 ofproto_dpif_unixctl_init(void)
6550 static bool registered;
6556 unixctl_command_register(
6558 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6559 1, 3, ofproto_unixctl_trace, NULL);
6560 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6561 ofproto_unixctl_fdb_flush, NULL);
6562 unixctl_command_register("fdb/show", "bridge", 1, 1,
6563 ofproto_unixctl_fdb_show, NULL);
6564 unixctl_command_register("ofproto/clog", "", 0, 0,
6565 ofproto_dpif_clog, NULL);
6566 unixctl_command_register("ofproto/unclog", "", 0, 0,
6567 ofproto_dpif_unclog, NULL);
6568 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6569 ofproto_dpif_self_check, NULL);
6570 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6571 ofproto_unixctl_dpif_dump_dps, NULL);
6572 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6574 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6575 ofproto_unixctl_dpif_dump_flows, NULL);
6576 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6577 ofproto_unixctl_dpif_del_flows, NULL);
6578 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6579 ofproto_unixctl_dpif_dump_megaflows, NULL);
6580 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6581 ofproto_unixctl_dpif_disable_megaflows, NULL);
6582 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6583 ofproto_unixctl_dpif_enable_megaflows, NULL);
6586 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6588 * This is deprecated. It is only for compatibility with broken device drivers
6589 * in old versions of Linux that do not properly support VLANs when VLAN
6590 * devices are not used. When broken device drivers are no longer in
6591 * widespread use, we will delete these interfaces. */
6594 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6597 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6599 if (realdev_ofp_port == ofport->realdev_ofp_port
6600 && vid == ofport->vlandev_vid) {
6604 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6606 if (ofport->realdev_ofp_port) {
6609 if (realdev_ofp_port && ofport->bundle) {
6610 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6611 * themselves be part of a bundle. */
6612 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6615 ofport->realdev_ofp_port = realdev_ofp_port;
6616 ofport->vlandev_vid = vid;
6618 if (realdev_ofp_port) {
6619 vsp_add(ofport, realdev_ofp_port, vid);
6626 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6628 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6632 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6634 return !hmap_is_empty(&ofproto->realdev_vid_map);
6637 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6638 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6639 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6640 * 'vlan_tci' 9, it would return the port number of eth0.9.
6642 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6643 * function just returns its 'realdev_ofp_port' argument. */
6645 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6646 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6648 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6649 int vid = vlan_tci_to_vid(vlan_tci);
6650 const struct vlan_splinter *vsp;
6652 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6653 hash_realdev_vid(realdev_ofp_port, vid),
6654 &ofproto->realdev_vid_map) {
6655 if (vsp->realdev_ofp_port == realdev_ofp_port
6656 && vsp->vid == vid) {
6657 return vsp->vlandev_ofp_port;
6661 return realdev_ofp_port;
6664 static struct vlan_splinter *
6665 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6667 struct vlan_splinter *vsp;
6669 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6670 hash_ofp_port(vlandev_ofp_port),
6671 &ofproto->vlandev_map) {
6672 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6680 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6681 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6682 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6683 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6684 * eth0 and store 9 in '*vid'.
6686 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6687 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6690 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6691 ofp_port_t vlandev_ofp_port, int *vid)
6693 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6694 const struct vlan_splinter *vsp;
6696 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6701 return vsp->realdev_ofp_port;
6707 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6708 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6709 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6710 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6711 * always the case unless VLAN splinters are enabled), returns false without
6712 * making any changes. */
6714 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6719 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6724 /* Cause the flow to be processed as if it came in on the real device with
6725 * the VLAN device's VLAN ID. */
6726 flow->in_port.ofp_port = realdev;
6727 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6732 vsp_remove(struct ofport_dpif *port)
6734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6735 struct vlan_splinter *vsp;
6737 vsp = vlandev_find(ofproto, port->up.ofp_port);
6739 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6740 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6743 port->realdev_ofp_port = 0;
6745 VLOG_ERR("missing vlan device record");
6750 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6754 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6755 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6756 == realdev_ofp_port)) {
6757 struct vlan_splinter *vsp;
6759 vsp = xmalloc(sizeof *vsp);
6760 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6761 hash_ofp_port(port->up.ofp_port));
6762 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6763 hash_realdev_vid(realdev_ofp_port, vid));
6764 vsp->realdev_ofp_port = realdev_ofp_port;
6765 vsp->vlandev_ofp_port = port->up.ofp_port;
6768 port->realdev_ofp_port = realdev_ofp_port;
6770 VLOG_ERR("duplicate vlan device record");
6775 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6777 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6778 return ofport ? ofport->odp_port : ODPP_NONE;
6781 static struct ofport_dpif *
6782 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6784 struct ofport_dpif *port;
6786 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6787 &backer->odp_to_ofport_map) {
6788 if (port->odp_port == odp_port) {
6797 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6799 struct ofport_dpif *port;
6801 port = odp_port_to_ofport(ofproto->backer, odp_port);
6802 if (port && &ofproto->up == port->up.ofproto) {
6803 return port->up.ofp_port;
6809 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6810 * most heavily weighted element. 'base' designates the rate of decay: after
6811 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6814 exp_mavg(double *avg, int base, double new)
6816 *avg = (*avg * (base - 1) + new) / base;
6820 update_moving_averages(struct dpif_backer *backer)
6822 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6823 long long int minutes = (time_msec() - backer->created) / min_ms;
6826 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6828 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6831 backer->lifetime.add_rate = 0.0;
6832 backer->lifetime.del_rate = 0.0;
6835 /* Update hourly averages on the minute boundaries. */
6836 if (time_msec() - backer->last_minute >= min_ms) {
6837 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6838 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6840 /* Update daily averages on the hour boundaries. */
6841 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6842 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6843 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6846 backer->total_subfacet_add_count += backer->subfacet_add_count;
6847 backer->total_subfacet_del_count += backer->subfacet_del_count;
6848 backer->subfacet_add_count = 0;
6849 backer->subfacet_del_count = 0;
6850 backer->last_minute += min_ms;
6854 const struct ofproto_class ofproto_dpif_class = {
6889 port_is_lacp_current,
6890 NULL, /* rule_choose_table */
6897 rule_modify_actions,
6911 get_stp_port_status,
6918 is_mirror_output_bundle,
6919 forward_bpdu_changed,
6920 set_mac_table_config,
6922 NULL, /* meter_get_features */
6923 NULL, /* meter_set */
6924 NULL, /* meter_get */
6925 NULL, /* meter_del */