2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
34 #include "guarded-list.h"
38 #include "mac-learning.h"
39 #include "meta-flow.h"
40 #include "multipath.h"
41 #include "netdev-vport.h"
46 #include "odp-execute.h"
49 #include "ofp-actions.h"
50 #include "ofp-parse.h"
51 #include "ofp-print.h"
52 #include "ofproto-dpif-governor.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-sflow.h"
56 #include "ofproto-dpif-upcall.h"
57 #include "ofproto-dpif-xlate.h"
58 #include "poll-loop.h"
63 #include "unaligned.h"
65 #include "vlan-bitmap.h"
68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
70 COVERAGE_DEFINE(ofproto_dpif_expired);
71 COVERAGE_DEFINE(facet_changed_rule);
72 COVERAGE_DEFINE(facet_revalidate);
73 COVERAGE_DEFINE(facet_unexpected);
74 COVERAGE_DEFINE(facet_suppress);
75 COVERAGE_DEFINE(subfacet_install_fail);
76 COVERAGE_DEFINE(packet_in_overflow);
77 COVERAGE_DEFINE(flow_mod_overflow);
79 /* Number of implemented OpenFlow tables. */
80 enum { N_TABLES = 255 };
81 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
82 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
92 * - Do include packets and bytes from facets that have been deleted or
93 * whose own statistics have been folded into the rule.
95 * - Do include packets and bytes sent "by hand" that were accounted to
96 * the rule without any facet being involved (this is a rare corner
97 * case in rule_execute()).
99 * - Do not include packet or bytes that can be obtained from any facet's
100 * packet_count or byte_count member or that can be obtained from the
101 * datapath by, e.g., dpif_flow_get() for any subfacet.
103 struct ovs_mutex stats_mutex;
104 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
105 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
108 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
110 static struct rule_dpif *rule_dpif_cast(const struct rule *);
113 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 void *aux; /* Key supplied by ofproto's client. */
116 char *name; /* Identifier for log messages. */
119 struct list ports; /* Contains "struct ofport"s. */
120 enum port_vlan_mode vlan_mode; /* VLAN mode */
121 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
122 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
123 * NULL if all VLANs are trunked. */
124 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
125 struct bond *bond; /* Nonnull iff more than one port. */
126 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
129 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
132 static void bundle_remove(struct ofport *);
133 static void bundle_update(struct ofbundle *);
134 static void bundle_destroy(struct ofbundle *);
135 static void bundle_del_port(struct ofport_dpif *);
136 static void bundle_run(struct ofbundle *);
137 static void bundle_wait(struct ofbundle *);
139 static void stp_run(struct ofproto_dpif *ofproto);
140 static void stp_wait(struct ofproto_dpif *ofproto);
141 static int set_stp_port(struct ofport *,
142 const struct ofproto_port_stp_settings *);
144 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
145 enum slow_path_reason,
146 uint64_t *stub, size_t stub_size,
147 const struct nlattr **actionsp,
148 size_t *actions_lenp);
150 /* A subfacet (see "struct subfacet" below) has three possible installation
153 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
154 * case just after the subfacet is created, just before the subfacet is
155 * destroyed, or if the datapath returns an error when we try to install a
158 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
160 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
161 * ofproto_dpif is installed in the datapath.
164 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
165 SF_FAST_PATH, /* Full actions are installed. */
166 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
169 /* A dpif flow and actions associated with a facet.
171 * See also the large comment on struct facet. */
174 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
175 struct list list_node; /* In struct facet's 'facets' list. */
176 struct facet *facet; /* Owning facet. */
177 struct dpif_backer *backer; /* Owning backer. */
179 enum odp_key_fitness key_fitness;
183 long long int used; /* Time last used; time created if not used. */
184 long long int created; /* Time created. */
186 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
187 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
189 enum subfacet_path path; /* Installed in datapath? */
192 #define SUBFACET_DESTROY_MAX_BATCH 50
194 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
195 static struct subfacet *subfacet_find(struct dpif_backer *,
196 const struct nlattr *key, size_t key_len,
198 static void subfacet_destroy(struct subfacet *);
199 static void subfacet_destroy__(struct subfacet *);
200 static void subfacet_destroy_batch(struct dpif_backer *,
201 struct subfacet **, int n);
202 static void subfacet_reset_dp_stats(struct subfacet *,
203 struct dpif_flow_stats *);
204 static void subfacet_update_stats(struct subfacet *,
205 const struct dpif_flow_stats *);
206 static int subfacet_install(struct subfacet *,
207 const struct ofpbuf *odp_actions,
208 struct dpif_flow_stats *);
209 static void subfacet_uninstall(struct subfacet *);
211 /* A unique, non-overlapping instantiation of an OpenFlow flow.
213 * A facet associates a "struct flow", which represents the Open vSwitch
214 * userspace idea of an exact-match flow, with one or more subfacets.
215 * While the facet is created based on an exact-match flow, it is stored
216 * within the ofproto based on the wildcards that could be expressed
217 * based on the flow table and other configuration. (See the 'wc'
218 * description in "struct xlate_out" for more details.)
220 * Each subfacet tracks the datapath's idea of the flow equivalent to
221 * the facet. When the kernel module (or other dpif implementation) and
222 * Open vSwitch userspace agree on the definition of a flow key, there
223 * is exactly one subfacet per facet. If the dpif implementation
224 * supports more-specific flow matching than userspace, however, a facet
225 * can have more than one subfacet. Examples include the dpif
226 * implementation not supporting the same wildcards as userspace or some
227 * distinction in flow that userspace simply doesn't understand.
229 * Flow expiration works in terms of subfacets, so a facet must have at
230 * least one subfacet or it will never expire, leaking memory. */
233 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
234 struct ofproto_dpif *ofproto;
237 struct list subfacets;
238 long long int used; /* Time last used; time created if not used. */
241 struct flow flow; /* Flow of the creating subfacet. */
242 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
246 * - Do include packets and bytes sent "by hand", e.g. with
249 * - Do include packets and bytes that were obtained from the datapath
250 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
251 * DPIF_FP_ZERO_STATS).
253 * - Do not include packets or bytes that can be obtained from the
254 * datapath for any existing subfacet.
256 uint64_t packet_count; /* Number of packets received. */
257 uint64_t byte_count; /* Number of bytes received. */
259 /* Resubmit statistics. */
260 uint64_t prev_packet_count; /* Number of packets from last stats push. */
261 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
262 long long int prev_used; /* Used time from last stats push. */
265 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
266 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
267 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
269 struct xlate_out xout;
271 /* Storage for a single subfacet, to reduce malloc() time and space
272 * overhead. (A facet always has at least one subfacet and in the common
273 * case has exactly one subfacet. However, 'one_subfacet' may not
274 * always be valid, since it could have been removed after newer
275 * subfacets were pushed onto the 'subfacets' list.) */
276 struct subfacet one_subfacet;
278 long long int learn_rl; /* Rate limiter for facet_learn(). */
281 static struct facet *facet_create(const struct flow_miss *);
282 static void facet_remove(struct facet *);
283 static void facet_free(struct facet *);
285 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
286 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
287 const struct flow *);
288 static bool facet_revalidate(struct facet *);
289 static bool facet_check_consistency(struct facet *);
291 static void facet_flush_stats(struct facet *);
293 static void facet_reset_counters(struct facet *);
294 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
295 struct dpif_flow_stats *, bool may_learn);
296 static void facet_push_stats(struct facet *, bool may_learn);
297 static void facet_learn(struct facet *);
298 static void facet_account(struct facet *);
299 static void push_all_stats(void);
301 static bool facet_is_controller_flow(struct facet *);
304 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
308 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
309 struct list bundle_node; /* In struct ofbundle's "ports" list. */
310 struct cfm *cfm; /* Connectivity Fault Management, if any. */
311 struct bfd *bfd; /* BFD, if any. */
312 bool may_enable; /* May be enabled in bonds. */
313 bool is_tunnel; /* This port is a tunnel. */
314 long long int carrier_seq; /* Carrier status changes. */
315 struct ofport_dpif *peer; /* Peer if patch port. */
318 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
319 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
320 long long int stp_state_entered;
322 /* Queue to DSCP mapping. */
323 struct ofproto_port_queue *qdscp;
326 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
328 * This is deprecated. It is only for compatibility with broken device
329 * drivers in old versions of Linux that do not properly support VLANs when
330 * VLAN devices are not used. When broken device drivers are no longer in
331 * widespread use, we will delete these interfaces. */
332 ofp_port_t realdev_ofp_port;
336 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
338 * This is deprecated. It is only for compatibility with broken device drivers
339 * in old versions of Linux that do not properly support VLANs when VLAN
340 * devices are not used. When broken device drivers are no longer in
341 * widespread use, we will delete these interfaces. */
342 struct vlan_splinter {
343 struct hmap_node realdev_vid_node;
344 struct hmap_node vlandev_node;
345 ofp_port_t realdev_ofp_port;
346 ofp_port_t vlandev_ofp_port;
350 static void vsp_remove(struct ofport_dpif *);
351 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
353 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
356 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
359 static struct ofport_dpif *
360 ofport_dpif_cast(const struct ofport *ofport)
362 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
365 static void port_run(struct ofport_dpif *);
366 static void port_run_fast(struct ofport_dpif *);
367 static void port_wait(struct ofport_dpif *);
368 static int set_bfd(struct ofport *, const struct smap *);
369 static int set_cfm(struct ofport *, const struct cfm_settings *);
370 static void ofport_update_peer(struct ofport_dpif *);
371 static void run_fast_rl(void);
372 static int run_fast(struct ofproto *);
374 struct dpif_completion {
375 struct list list_node;
376 struct ofoperation *op;
379 /* Reasons that we might need to revalidate every facet, and corresponding
382 * A value of 0 means that there is no need to revalidate.
384 * It would be nice to have some cleaner way to integrate with coverage
385 * counters, but with only a few reasons I guess this is good enough for
387 enum revalidate_reason {
388 REV_RECONFIGURE = 1, /* Switch configuration changed. */
389 REV_STP, /* Spanning tree protocol port status change. */
390 REV_BOND, /* Bonding changed. */
391 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
392 REV_FLOW_TABLE, /* Flow table changed. */
393 REV_MAC_LEARNING, /* Mac learning changed. */
394 REV_INCONSISTENCY /* Facet self-check failed. */
396 COVERAGE_DEFINE(rev_reconfigure);
397 COVERAGE_DEFINE(rev_stp);
398 COVERAGE_DEFINE(rev_bond);
399 COVERAGE_DEFINE(rev_port_toggled);
400 COVERAGE_DEFINE(rev_flow_table);
401 COVERAGE_DEFINE(rev_mac_learning);
402 COVERAGE_DEFINE(rev_inconsistency);
404 struct avg_subfacet_rates {
405 double add_rate; /* Moving average of new flows created per minute. */
406 double del_rate; /* Moving average of flows deleted per minute. */
409 /* All datapaths of a given type share a single dpif backer instance. */
415 struct timer next_expiration;
417 struct ovs_rwlock odp_to_ofport_lock;
418 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
420 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
422 /* Facet revalidation flags applying to facets which use this backer. */
423 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
425 struct hmap drop_keys; /* Set of dropped odp keys. */
426 bool recv_set_enable; /* Enables or disables receiving packets. */
428 struct hmap subfacets;
429 struct governor *governor;
431 /* Subfacet statistics.
433 * These keep track of the total number of subfacets added and deleted and
434 * flow life span. They are useful for computing the flow rates stats
435 * exposed via "ovs-appctl dpif/show". The goal is to learn about
436 * traffic patterns in ways that we can use later to improve Open vSwitch
437 * performance in new situations. */
438 long long int created; /* Time when it is created. */
439 unsigned max_n_subfacet; /* Maximum number of flows */
440 unsigned avg_n_subfacet; /* Average number of flows. */
441 long long int avg_subfacet_life; /* Average life span of subfacets. */
443 /* The average number of subfacets... */
444 struct avg_subfacet_rates hourly; /* ...over the last hour. */
445 struct avg_subfacet_rates daily; /* ...over the last day. */
446 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
447 long long int last_minute; /* Last time 'hourly' was updated. */
449 /* Number of subfacets added or deleted since 'last_minute'. */
450 unsigned subfacet_add_count;
451 unsigned subfacet_del_count;
453 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
454 unsigned long long int total_subfacet_add_count;
455 unsigned long long int total_subfacet_del_count;
457 /* Number of upcall handling threads. */
458 unsigned int n_handler_threads;
461 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
462 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
464 static void drop_key_clear(struct dpif_backer *);
465 static void update_moving_averages(struct dpif_backer *backer);
467 struct ofproto_dpif {
468 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
470 struct dpif_backer *backer;
472 /* Special OpenFlow rules. */
473 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
474 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
475 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
478 struct netflow *netflow;
479 struct dpif_sflow *sflow;
480 struct dpif_ipfix *ipfix;
481 struct hmap bundles; /* Contains "struct ofbundle"s. */
482 struct mac_learning *ml;
483 bool has_bonded_bundles;
484 struct mbridge *mbridge;
487 struct classifier facets; /* Contains 'struct facet's. */
488 long long int consistency_rl;
490 struct netdev_stats stats; /* To account packets generated and consumed in
495 long long int stp_last_tick;
497 /* VLAN splinters. */
498 struct ovs_mutex vsp_mutex;
499 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
500 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
503 struct sset ports; /* Set of standard port names. */
504 struct sset ghost_ports; /* Ports with no datapath port. */
505 struct sset port_poll_set; /* Queued names for port_poll() reply. */
506 int port_poll_errno; /* Last errno for port_poll() reply. */
508 /* Per ofproto's dpif stats. */
513 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
516 /* By default, flows in the datapath are wildcarded (megaflows). They
517 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
518 static bool enable_megaflows = true;
520 /* All existing ofproto_dpif instances, indexed by ->up.name. */
521 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
523 static void ofproto_dpif_unixctl_init(void);
525 static inline struct ofproto_dpif *
526 ofproto_dpif_cast(const struct ofproto *ofproto)
528 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
529 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
532 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
533 ofp_port_t ofp_port);
534 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
535 const struct ofpbuf *packet, struct ds *);
538 static void handle_upcalls(struct dpif_backer *);
540 /* Flow expiration. */
541 static int expire(struct dpif_backer *);
544 static void send_netflow_active_timeouts(struct ofproto_dpif *);
547 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
549 /* Global variables. */
550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
552 /* Initial mappings of port to bridge mappings. */
553 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
555 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
558 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
559 struct ofputil_flow_mod *fm)
561 ofproto_flow_mod(&ofproto->up, fm);
564 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
565 * Takes ownership of 'pin' and pin->packet. */
567 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
568 struct ofputil_packet_in *pin)
570 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
571 COVERAGE_INC(packet_in_overflow);
572 free(CONST_CAST(void *, pin->packet));
577 /* Factory functions. */
580 init(const struct shash *iface_hints)
582 struct shash_node *node;
584 /* Make a local copy, since we don't own 'iface_hints' elements. */
585 SHASH_FOR_EACH(node, iface_hints) {
586 const struct iface_hint *orig_hint = node->data;
587 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
589 new_hint->br_name = xstrdup(orig_hint->br_name);
590 new_hint->br_type = xstrdup(orig_hint->br_type);
591 new_hint->ofp_port = orig_hint->ofp_port;
593 shash_add(&init_ofp_ports, node->name, new_hint);
598 enumerate_types(struct sset *types)
600 dp_enumerate_types(types);
604 enumerate_names(const char *type, struct sset *names)
606 struct ofproto_dpif *ofproto;
609 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
610 if (strcmp(type, ofproto->up.type)) {
613 sset_add(names, ofproto->up.name);
620 del(const char *type, const char *name)
625 error = dpif_open(name, type, &dpif);
627 error = dpif_delete(dpif);
634 port_open_type(const char *datapath_type, const char *port_type)
636 return dpif_port_open_type(datapath_type, port_type);
639 /* Type functions. */
641 static void process_dpif_port_changes(struct dpif_backer *);
642 static void process_dpif_all_ports_changed(struct dpif_backer *);
643 static void process_dpif_port_change(struct dpif_backer *,
644 const char *devname);
645 static void process_dpif_port_error(struct dpif_backer *, int error);
647 static struct ofproto_dpif *
648 lookup_ofproto_dpif_by_port_name(const char *name)
650 struct ofproto_dpif *ofproto;
652 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
653 if (sset_contains(&ofproto->ports, name)) {
662 type_run(const char *type)
664 static long long int push_timer = LLONG_MIN;
665 struct dpif_backer *backer;
667 backer = shash_find_data(&all_dpif_backers, type);
669 /* This is not necessarily a problem, since backers are only
670 * created on demand. */
674 dpif_run(backer->dpif);
676 /* The most natural place to push facet statistics is when they're pulled
677 * from the datapath. However, when there are many flows in the datapath,
678 * this expensive operation can occur so frequently, that it reduces our
679 * ability to quickly set up flows. To reduce the cost, we push statistics
681 if (time_msec() > push_timer) {
682 push_timer = time_msec() + 2000;
686 /* If vswitchd started with other_config:flow_restore_wait set as "true",
687 * and the configuration has now changed to "false", enable receiving
688 * packets from the datapath. */
689 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
692 backer->recv_set_enable = true;
694 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
696 udpif_recv_set(backer->udpif, 0, false);
697 VLOG_ERR("Failed to enable receiving packets in dpif.");
700 udpif_recv_set(backer->udpif, n_handler_threads,
701 backer->recv_set_enable);
702 dpif_flow_flush(backer->dpif);
703 backer->need_revalidate = REV_RECONFIGURE;
706 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
707 * to reset the handler threads. */
708 if (backer->n_handler_threads != n_handler_threads) {
709 udpif_recv_set(backer->udpif, n_handler_threads,
710 backer->recv_set_enable);
711 backer->n_handler_threads = n_handler_threads;
714 if (backer->need_revalidate) {
715 struct ofproto_dpif *ofproto;
716 struct simap_node *node;
717 struct simap tmp_backers;
719 /* Handle tunnel garbage collection. */
720 simap_init(&tmp_backers);
721 simap_swap(&backer->tnl_backers, &tmp_backers);
723 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
724 struct ofport_dpif *iter;
726 if (backer != ofproto->backer) {
730 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
731 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
734 if (!iter->is_tunnel) {
738 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
739 namebuf, sizeof namebuf);
740 node = simap_find(&tmp_backers, dp_port);
742 simap_put(&backer->tnl_backers, dp_port, node->data);
743 simap_delete(&tmp_backers, node);
744 node = simap_find(&backer->tnl_backers, dp_port);
746 node = simap_find(&backer->tnl_backers, dp_port);
748 odp_port_t odp_port = ODPP_NONE;
750 if (!dpif_port_add(backer->dpif, iter->up.netdev,
752 simap_put(&backer->tnl_backers, dp_port,
753 odp_to_u32(odp_port));
754 node = simap_find(&backer->tnl_backers, dp_port);
759 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
760 if (tnl_port_reconfigure(iter, iter->up.netdev,
762 backer->need_revalidate = REV_RECONFIGURE;
767 SIMAP_FOR_EACH (node, &tmp_backers) {
768 dpif_port_del(backer->dpif, u32_to_odp(node->data));
770 simap_destroy(&tmp_backers);
772 switch (backer->need_revalidate) {
773 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
774 case REV_STP: COVERAGE_INC(rev_stp); break;
775 case REV_BOND: COVERAGE_INC(rev_bond); break;
776 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
777 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
778 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
779 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
781 backer->need_revalidate = 0;
783 /* Clear the drop_keys in case we should now be accepting some
784 * formerly dropped flows. */
785 drop_key_clear(backer);
787 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
788 struct facet *facet, *next;
789 struct ofport_dpif *ofport;
790 struct cls_cursor cursor;
791 struct ofbundle *bundle;
793 if (ofproto->backer != backer) {
797 ovs_rwlock_wrlock(&xlate_rwlock);
798 xlate_ofproto_set(ofproto, ofproto->up.name,
799 ofproto->backer->dpif, ofproto->miss_rule,
800 ofproto->no_packet_in_rule, ofproto->ml,
801 ofproto->stp, ofproto->mbridge,
802 ofproto->sflow, ofproto->ipfix,
803 ofproto->up.frag_handling,
804 ofproto->up.forward_bpdu,
805 connmgr_has_in_band(ofproto->up.connmgr),
806 ofproto->netflow != NULL);
808 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
809 xlate_bundle_set(ofproto, bundle, bundle->name,
810 bundle->vlan_mode, bundle->vlan,
811 bundle->trunks, bundle->use_priority_tags,
812 bundle->bond, bundle->lacp,
816 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
817 int stp_port = ofport->stp_port
818 ? stp_port_no(ofport->stp_port)
820 xlate_ofport_set(ofproto, ofport->bundle, ofport,
821 ofport->up.ofp_port, ofport->odp_port,
822 ofport->up.netdev, ofport->cfm,
823 ofport->bfd, ofport->peer, stp_port,
824 ofport->qdscp, ofport->n_qdscp,
825 ofport->up.pp.config, ofport->is_tunnel,
828 ovs_rwlock_unlock(&xlate_rwlock);
830 /* Only ofproto-dpif cares about the facet classifier so we just
831 * lock cls_cursor_init() to appease the thread safety analysis. */
832 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
833 cls_cursor_init(&cursor, &ofproto->facets, NULL);
834 ovs_rwlock_unlock(&ofproto->facets.rwlock);
835 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
836 facet_revalidate(facet);
841 udpif_revalidate(backer->udpif);
844 if (!backer->recv_set_enable) {
845 /* Wake up before a max of 1000ms. */
846 timer_set_duration(&backer->next_expiration, 1000);
847 } else if (timer_expired(&backer->next_expiration)) {
848 int delay = expire(backer);
849 timer_set_duration(&backer->next_expiration, delay);
852 process_dpif_port_changes(backer);
854 if (backer->governor) {
857 governor_run(backer->governor);
859 /* If the governor has shrunk to its minimum size and the number of
860 * subfacets has dwindled, then drop the governor entirely.
862 * For hysteresis, the number of subfacets to drop the governor is
863 * smaller than the number needed to trigger its creation. */
864 n_subfacets = hmap_count(&backer->subfacets);
865 if (n_subfacets * 4 < flow_eviction_threshold
866 && governor_is_idle(backer->governor)) {
867 governor_destroy(backer->governor);
868 backer->governor = NULL;
875 /* Check for and handle port changes in 'backer''s dpif. */
877 process_dpif_port_changes(struct dpif_backer *backer)
883 error = dpif_port_poll(backer->dpif, &devname);
889 process_dpif_all_ports_changed(backer);
893 process_dpif_port_change(backer, devname);
898 process_dpif_port_error(backer, error);
905 process_dpif_all_ports_changed(struct dpif_backer *backer)
907 struct ofproto_dpif *ofproto;
908 struct dpif_port dpif_port;
909 struct dpif_port_dump dump;
910 struct sset devnames;
913 sset_init(&devnames);
914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
915 if (ofproto->backer == backer) {
916 struct ofport *ofport;
918 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
919 sset_add(&devnames, netdev_get_name(ofport->netdev));
923 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
924 sset_add(&devnames, dpif_port.name);
927 SSET_FOR_EACH (devname, &devnames) {
928 process_dpif_port_change(backer, devname);
930 sset_destroy(&devnames);
934 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
936 struct ofproto_dpif *ofproto;
937 struct dpif_port port;
939 /* Don't report on the datapath's device. */
940 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
944 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
945 &all_ofproto_dpifs) {
946 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
951 ofproto = lookup_ofproto_dpif_by_port_name(devname);
952 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
953 /* The port was removed. If we know the datapath,
954 * report it through poll_set(). If we don't, it may be
955 * notifying us of a removal we initiated, so ignore it.
956 * If there's a pending ENOBUFS, let it stand, since
957 * everything will be reevaluated. */
958 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
959 sset_add(&ofproto->port_poll_set, devname);
960 ofproto->port_poll_errno = 0;
962 } else if (!ofproto) {
963 /* The port was added, but we don't know with which
964 * ofproto we should associate it. Delete it. */
965 dpif_port_del(backer->dpif, port.port_no);
967 struct ofport_dpif *ofport;
969 ofport = ofport_dpif_cast(shash_find_data(
970 &ofproto->up.port_by_name, devname));
972 && ofport->odp_port != port.port_no
973 && !odp_port_to_ofport(backer, port.port_no))
975 /* 'ofport''s datapath port number has changed from
976 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
977 * structures to match. */
978 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
979 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
980 ofport->odp_port = port.port_no;
981 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
982 hash_odp_port(port.port_no));
983 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
984 backer->need_revalidate = REV_RECONFIGURE;
987 dpif_port_destroy(&port);
990 /* Propagate 'error' to all ofprotos based on 'backer'. */
992 process_dpif_port_error(struct dpif_backer *backer, int error)
994 struct ofproto_dpif *ofproto;
996 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
997 if (ofproto->backer == backer) {
998 sset_clear(&ofproto->port_poll_set);
999 ofproto->port_poll_errno = error;
1005 dpif_backer_run_fast(struct dpif_backer *backer)
1007 handle_upcalls(backer);
1013 type_run_fast(const char *type)
1015 struct dpif_backer *backer;
1017 backer = shash_find_data(&all_dpif_backers, type);
1019 /* This is not necessarily a problem, since backers are only
1020 * created on demand. */
1024 return dpif_backer_run_fast(backer);
1030 static long long int port_rl = LLONG_MIN;
1032 if (time_msec() >= port_rl) {
1033 struct ofproto_dpif *ofproto;
1035 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1036 run_fast(&ofproto->up);
1038 port_rl = time_msec() + 200;
1043 type_wait(const char *type)
1045 struct dpif_backer *backer;
1047 backer = shash_find_data(&all_dpif_backers, type);
1049 /* This is not necessarily a problem, since backers are only
1050 * created on demand. */
1054 if (backer->governor) {
1055 governor_wait(backer->governor);
1058 timer_wait(&backer->next_expiration);
1059 dpif_wait(backer->dpif);
1060 udpif_wait(backer->udpif);
1063 /* Basic life-cycle. */
1065 static int add_internal_flows(struct ofproto_dpif *);
1067 static struct ofproto *
1070 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1071 return &ofproto->up;
1075 dealloc(struct ofproto *ofproto_)
1077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1082 close_dpif_backer(struct dpif_backer *backer)
1084 struct shash_node *node;
1086 ovs_assert(backer->refcount > 0);
1088 if (--backer->refcount) {
1092 drop_key_clear(backer);
1093 hmap_destroy(&backer->drop_keys);
1095 udpif_destroy(backer->udpif);
1097 simap_destroy(&backer->tnl_backers);
1098 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1099 hmap_destroy(&backer->odp_to_ofport_map);
1100 node = shash_find(&all_dpif_backers, backer->type);
1102 shash_delete(&all_dpif_backers, node);
1103 dpif_close(backer->dpif);
1105 ovs_assert(hmap_is_empty(&backer->subfacets));
1106 hmap_destroy(&backer->subfacets);
1107 governor_destroy(backer->governor);
1112 /* Datapath port slated for removal from datapath. */
1113 struct odp_garbage {
1114 struct list list_node;
1115 odp_port_t odp_port;
1119 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1121 struct dpif_backer *backer;
1122 struct dpif_port_dump port_dump;
1123 struct dpif_port port;
1124 struct shash_node *node;
1125 struct list garbage_list;
1126 struct odp_garbage *garbage, *next;
1132 backer = shash_find_data(&all_dpif_backers, type);
1139 backer_name = xasprintf("ovs-%s", type);
1141 /* Remove any existing datapaths, since we assume we're the only
1142 * userspace controlling the datapath. */
1144 dp_enumerate_names(type, &names);
1145 SSET_FOR_EACH(name, &names) {
1146 struct dpif *old_dpif;
1148 /* Don't remove our backer if it exists. */
1149 if (!strcmp(name, backer_name)) {
1153 if (dpif_open(name, type, &old_dpif)) {
1154 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1156 dpif_delete(old_dpif);
1157 dpif_close(old_dpif);
1160 sset_destroy(&names);
1162 backer = xmalloc(sizeof *backer);
1164 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1167 VLOG_ERR("failed to open datapath of type %s: %s", type,
1168 ovs_strerror(error));
1172 backer->udpif = udpif_create(backer, backer->dpif);
1174 backer->type = xstrdup(type);
1175 backer->governor = NULL;
1176 backer->refcount = 1;
1177 hmap_init(&backer->odp_to_ofport_map);
1178 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1179 hmap_init(&backer->drop_keys);
1180 hmap_init(&backer->subfacets);
1181 timer_set_duration(&backer->next_expiration, 1000);
1182 backer->need_revalidate = 0;
1183 simap_init(&backer->tnl_backers);
1184 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1187 if (backer->recv_set_enable) {
1188 dpif_flow_flush(backer->dpif);
1191 /* Loop through the ports already on the datapath and remove any
1192 * that we don't need anymore. */
1193 list_init(&garbage_list);
1194 dpif_port_dump_start(&port_dump, backer->dpif);
1195 while (dpif_port_dump_next(&port_dump, &port)) {
1196 node = shash_find(&init_ofp_ports, port.name);
1197 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1198 garbage = xmalloc(sizeof *garbage);
1199 garbage->odp_port = port.port_no;
1200 list_push_front(&garbage_list, &garbage->list_node);
1203 dpif_port_dump_done(&port_dump);
1205 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1206 dpif_port_del(backer->dpif, garbage->odp_port);
1207 list_remove(&garbage->list_node);
1211 shash_add(&all_dpif_backers, type, backer);
1213 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1215 VLOG_ERR("failed to listen on datapath of type %s: %s",
1216 type, ovs_strerror(error));
1217 close_dpif_backer(backer);
1220 udpif_recv_set(backer->udpif, n_handler_threads,
1221 backer->recv_set_enable);
1222 backer->n_handler_threads = n_handler_threads;
1224 backer->max_n_subfacet = 0;
1225 backer->created = time_msec();
1226 backer->last_minute = backer->created;
1227 memset(&backer->hourly, 0, sizeof backer->hourly);
1228 memset(&backer->daily, 0, sizeof backer->daily);
1229 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1230 backer->subfacet_add_count = 0;
1231 backer->subfacet_del_count = 0;
1232 backer->total_subfacet_add_count = 0;
1233 backer->total_subfacet_del_count = 0;
1234 backer->avg_n_subfacet = 0;
1235 backer->avg_subfacet_life = 0;
1241 construct(struct ofproto *ofproto_)
1243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1244 struct shash_node *node, *next;
1248 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1253 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1254 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1256 ofproto->netflow = NULL;
1257 ofproto->sflow = NULL;
1258 ofproto->ipfix = NULL;
1259 ofproto->stp = NULL;
1260 hmap_init(&ofproto->bundles);
1261 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1262 ofproto->mbridge = mbridge_create();
1263 ofproto->has_bonded_bundles = false;
1264 ovs_mutex_init(&ofproto->vsp_mutex);
1266 classifier_init(&ofproto->facets);
1267 ofproto->consistency_rl = LLONG_MIN;
1269 guarded_list_init(&ofproto->pins);
1271 ofproto_dpif_unixctl_init();
1273 hmap_init(&ofproto->vlandev_map);
1274 hmap_init(&ofproto->realdev_vid_map);
1276 sset_init(&ofproto->ports);
1277 sset_init(&ofproto->ghost_ports);
1278 sset_init(&ofproto->port_poll_set);
1279 ofproto->port_poll_errno = 0;
1281 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1282 struct iface_hint *iface_hint = node->data;
1284 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1285 /* Check if the datapath already has this port. */
1286 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1287 sset_add(&ofproto->ports, node->name);
1290 free(iface_hint->br_name);
1291 free(iface_hint->br_type);
1293 shash_delete(&init_ofp_ports, node);
1297 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1298 hash_string(ofproto->up.name, 0));
1299 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1301 ofproto_init_tables(ofproto_, N_TABLES);
1302 error = add_internal_flows(ofproto);
1303 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1306 ofproto->n_missed = 0;
1312 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1313 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1315 struct ofputil_flow_mod fm;
1318 match_init_catchall(&fm.match);
1320 match_set_reg(&fm.match, 0, id);
1321 fm.new_cookie = htonll(0);
1322 fm.cookie = htonll(0);
1323 fm.cookie_mask = htonll(0);
1324 fm.modify_cookie = false;
1325 fm.table_id = TBL_INTERNAL;
1326 fm.command = OFPFC_ADD;
1327 fm.idle_timeout = 0;
1328 fm.hard_timeout = 0;
1332 fm.ofpacts = ofpacts->data;
1333 fm.ofpacts_len = ofpacts->size;
1335 error = ofproto_flow_mod(&ofproto->up, &fm);
1337 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1338 id, ofperr_to_string(error));
1342 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1344 rule_dpif_unref(*rulep);
1353 add_internal_flows(struct ofproto_dpif *ofproto)
1355 struct ofpact_controller *controller;
1356 uint64_t ofpacts_stub[128 / 8];
1357 struct ofpbuf ofpacts;
1361 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1364 controller = ofpact_put_CONTROLLER(&ofpacts);
1365 controller->max_len = UINT16_MAX;
1366 controller->controller_id = 0;
1367 controller->reason = OFPR_NO_MATCH;
1368 ofpact_pad(&ofpacts);
1370 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1375 ofpbuf_clear(&ofpacts);
1376 error = add_internal_flow(ofproto, id++, &ofpacts,
1377 &ofproto->no_packet_in_rule);
1382 error = add_internal_flow(ofproto, id++, &ofpacts,
1383 &ofproto->drop_frags_rule);
1388 destruct(struct ofproto *ofproto_)
1390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1391 struct rule_dpif *rule, *next_rule;
1392 struct ofputil_packet_in *pin, *next_pin;
1393 struct facet *facet, *next_facet;
1394 struct cls_cursor cursor;
1395 struct oftable *table;
1398 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1399 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1400 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1401 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1402 facet_remove(facet);
1405 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1406 ovs_rwlock_wrlock(&xlate_rwlock);
1407 xlate_remove_ofproto(ofproto);
1408 ovs_rwlock_unlock(&xlate_rwlock);
1410 /* Ensure that the upcall processing threads have no remaining references
1411 * to the ofproto or anything in it. */
1412 udpif_synchronize(ofproto->backer->udpif);
1414 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1415 * use-after-free error. */
1416 udpif_revalidate(ofproto->backer->udpif);
1418 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1420 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1421 struct cls_cursor cursor;
1423 ovs_rwlock_rdlock(&table->cls.rwlock);
1424 cls_cursor_init(&cursor, &table->cls, NULL);
1425 ovs_rwlock_unlock(&table->cls.rwlock);
1426 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1427 ofproto_rule_delete(&ofproto->up, &rule->up);
1431 guarded_list_pop_all(&ofproto->pins, &pins);
1432 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1433 list_remove(&pin->list_node);
1434 free(CONST_CAST(void *, pin->packet));
1437 guarded_list_destroy(&ofproto->pins);
1439 mbridge_unref(ofproto->mbridge);
1441 netflow_destroy(ofproto->netflow);
1442 dpif_sflow_unref(ofproto->sflow);
1443 hmap_destroy(&ofproto->bundles);
1444 mac_learning_unref(ofproto->ml);
1446 classifier_destroy(&ofproto->facets);
1448 hmap_destroy(&ofproto->vlandev_map);
1449 hmap_destroy(&ofproto->realdev_vid_map);
1451 sset_destroy(&ofproto->ports);
1452 sset_destroy(&ofproto->ghost_ports);
1453 sset_destroy(&ofproto->port_poll_set);
1455 ovs_mutex_destroy(&ofproto->vsp_mutex);
1457 close_dpif_backer(ofproto->backer);
1461 run_fast(struct ofproto *ofproto_)
1463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1464 struct ofputil_packet_in *pin, *next_pin;
1465 struct ofport_dpif *ofport;
1468 /* Do not perform any periodic activity required by 'ofproto' while
1469 * waiting for flow restore to complete. */
1470 if (ofproto_get_flow_restore_wait()) {
1474 guarded_list_pop_all(&ofproto->pins, &pins);
1475 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1476 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1477 list_remove(&pin->list_node);
1478 free(CONST_CAST(void *, pin->packet));
1482 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1483 port_run_fast(ofport);
1490 run(struct ofproto *ofproto_)
1492 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1493 struct ofport_dpif *ofport;
1494 struct ofbundle *bundle;
1497 if (mbridge_need_revalidate(ofproto->mbridge)) {
1498 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1499 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1500 mac_learning_flush(ofproto->ml);
1501 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1504 /* Do not perform any periodic activity below required by 'ofproto' while
1505 * waiting for flow restore to complete. */
1506 if (ofproto_get_flow_restore_wait()) {
1510 error = run_fast(ofproto_);
1515 if (ofproto->netflow) {
1516 if (netflow_run(ofproto->netflow)) {
1517 send_netflow_active_timeouts(ofproto);
1520 if (ofproto->sflow) {
1521 dpif_sflow_run(ofproto->sflow);
1523 if (ofproto->ipfix) {
1524 dpif_ipfix_run(ofproto->ipfix);
1527 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1530 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1535 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1536 if (mac_learning_run(ofproto->ml)) {
1537 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1539 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1541 /* Check the consistency of a random facet, to aid debugging. */
1542 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1543 if (time_msec() >= ofproto->consistency_rl
1544 && !classifier_is_empty(&ofproto->facets)
1545 && !ofproto->backer->need_revalidate) {
1546 struct cls_table *table;
1547 struct cls_rule *cr;
1548 struct facet *facet;
1550 ofproto->consistency_rl = time_msec() + 250;
1552 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1553 struct cls_table, hmap_node);
1554 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1556 facet = CONTAINER_OF(cr, struct facet, cr);
1558 if (!facet_check_consistency(facet)) {
1559 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1562 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1568 wait(struct ofproto *ofproto_)
1570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1571 struct ofport_dpif *ofport;
1572 struct ofbundle *bundle;
1574 if (ofproto_get_flow_restore_wait()) {
1578 if (ofproto->sflow) {
1579 dpif_sflow_wait(ofproto->sflow);
1581 if (ofproto->ipfix) {
1582 dpif_ipfix_wait(ofproto->ipfix);
1584 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1587 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1588 bundle_wait(bundle);
1590 if (ofproto->netflow) {
1591 netflow_wait(ofproto->netflow);
1593 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1594 mac_learning_wait(ofproto->ml);
1595 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1597 if (ofproto->backer->need_revalidate) {
1598 /* Shouldn't happen, but if it does just go around again. */
1599 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1600 poll_immediate_wake();
1605 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1607 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1608 struct cls_cursor cursor;
1609 size_t n_subfacets = 0;
1610 struct facet *facet;
1612 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1613 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1614 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1616 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1617 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1618 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1619 n_subfacets += list_size(&facet->subfacets);
1621 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1622 simap_increase(usage, "subfacets", n_subfacets);
1626 flush(struct ofproto *ofproto_)
1628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1629 struct subfacet *subfacet, *next_subfacet;
1630 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1634 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1635 &ofproto->backer->subfacets) {
1636 if (subfacet->facet->ofproto != ofproto) {
1640 if (subfacet->path != SF_NOT_INSTALLED) {
1641 batch[n_batch++] = subfacet;
1642 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1643 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1647 subfacet_destroy(subfacet);
1652 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1657 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1658 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1660 *arp_match_ip = true;
1661 *actions = (OFPUTIL_A_OUTPUT |
1662 OFPUTIL_A_SET_VLAN_VID |
1663 OFPUTIL_A_SET_VLAN_PCP |
1664 OFPUTIL_A_STRIP_VLAN |
1665 OFPUTIL_A_SET_DL_SRC |
1666 OFPUTIL_A_SET_DL_DST |
1667 OFPUTIL_A_SET_NW_SRC |
1668 OFPUTIL_A_SET_NW_DST |
1669 OFPUTIL_A_SET_NW_TOS |
1670 OFPUTIL_A_SET_TP_SRC |
1671 OFPUTIL_A_SET_TP_DST |
1676 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1679 struct dpif_dp_stats s;
1680 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1683 strcpy(ots->name, "classifier");
1685 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1686 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, true);
1687 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1689 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1692 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1693 ots->lookup_count = htonll(n_lookup);
1694 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1697 static struct ofport *
1700 struct ofport_dpif *port = xmalloc(sizeof *port);
1705 port_dealloc(struct ofport *port_)
1707 struct ofport_dpif *port = ofport_dpif_cast(port_);
1712 port_construct(struct ofport *port_)
1714 struct ofport_dpif *port = ofport_dpif_cast(port_);
1715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1716 const struct netdev *netdev = port->up.netdev;
1717 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1718 struct dpif_port dpif_port;
1721 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1722 port->bundle = NULL;
1725 port->may_enable = true;
1726 port->stp_port = NULL;
1727 port->stp_state = STP_DISABLED;
1728 port->is_tunnel = false;
1732 port->realdev_ofp_port = 0;
1733 port->vlandev_vid = 0;
1734 port->carrier_seq = netdev_get_carrier_resets(netdev);
1736 if (netdev_vport_is_patch(netdev)) {
1737 /* By bailing out here, we don't submit the port to the sFlow module
1738 * to be considered for counter polling export. This is correct
1739 * because the patch port represents an interface that sFlow considers
1740 * to be "internal" to the switch as a whole, and therefore not an
1741 * candidate for counter polling. */
1742 port->odp_port = ODPP_NONE;
1743 ofport_update_peer(port);
1747 error = dpif_port_query_by_name(ofproto->backer->dpif,
1748 netdev_vport_get_dpif_port(netdev, namebuf,
1755 port->odp_port = dpif_port.port_no;
1757 if (netdev_get_tunnel_config(netdev)) {
1758 tnl_port_add(port, port->up.netdev, port->odp_port);
1759 port->is_tunnel = true;
1761 /* Sanity-check that a mapping doesn't already exist. This
1762 * shouldn't happen for non-tunnel ports. */
1763 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1764 VLOG_ERR("port %s already has an OpenFlow port number",
1766 dpif_port_destroy(&dpif_port);
1770 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1771 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1772 hash_odp_port(port->odp_port));
1773 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1775 dpif_port_destroy(&dpif_port);
1777 if (ofproto->sflow) {
1778 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1785 port_destruct(struct ofport *port_)
1787 struct ofport_dpif *port = ofport_dpif_cast(port_);
1788 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1789 const char *devname = netdev_get_name(port->up.netdev);
1790 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1791 const char *dp_port_name;
1793 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1794 ovs_rwlock_wrlock(&xlate_rwlock);
1795 xlate_ofport_remove(port);
1796 ovs_rwlock_unlock(&xlate_rwlock);
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 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1817 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1818 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1822 sset_find_and_delete(&ofproto->ports, devname);
1823 sset_find_and_delete(&ofproto->ghost_ports, devname);
1824 bundle_remove(port_);
1825 set_cfm(port_, NULL);
1826 set_bfd(port_, NULL);
1827 if (port->stp_port) {
1828 stp_port_disable(port->stp_port);
1830 if (ofproto->sflow) {
1831 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1838 port_modified(struct ofport *port_)
1840 struct ofport_dpif *port = ofport_dpif_cast(port_);
1842 if (port->bundle && port->bundle->bond) {
1843 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1847 cfm_set_netdev(port->cfm, port->up.netdev);
1851 bfd_set_netdev(port->bfd, port->up.netdev);
1854 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1856 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1860 ofport_update_peer(port);
1864 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1866 struct ofport_dpif *port = ofport_dpif_cast(port_);
1867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1868 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1870 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1871 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1872 OFPUTIL_PC_NO_PACKET_IN)) {
1873 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1875 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1876 bundle_update(port->bundle);
1882 set_sflow(struct ofproto *ofproto_,
1883 const struct ofproto_sflow_options *sflow_options)
1885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1886 struct dpif_sflow *ds = ofproto->sflow;
1888 if (sflow_options) {
1890 struct ofport_dpif *ofport;
1892 ds = ofproto->sflow = dpif_sflow_create();
1893 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1894 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1896 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1898 dpif_sflow_set_options(ds, sflow_options);
1901 dpif_sflow_unref(ds);
1902 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1903 ofproto->sflow = NULL;
1911 struct ofproto *ofproto_,
1912 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1913 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1914 size_t n_flow_exporters_options)
1916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1917 struct dpif_ipfix *di = ofproto->ipfix;
1918 bool has_options = bridge_exporter_options || flow_exporters_options;
1920 if (has_options && !di) {
1921 di = ofproto->ipfix = dpif_ipfix_create();
1925 /* Call set_options in any case to cleanly flush the flow
1926 * caches in the last exporters that are to be destroyed. */
1927 dpif_ipfix_set_options(
1928 di, bridge_exporter_options, flow_exporters_options,
1929 n_flow_exporters_options);
1932 dpif_ipfix_unref(di);
1933 ofproto->ipfix = NULL;
1941 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1943 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1950 struct ofproto_dpif *ofproto;
1952 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1953 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1954 ofport->cfm = cfm_create(ofport->up.netdev);
1957 if (cfm_configure(ofport->cfm, s)) {
1963 cfm_unref(ofport->cfm);
1969 get_cfm_status(const struct ofport *ofport_,
1970 struct ofproto_cfm_status *status)
1972 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1975 status->faults = cfm_get_fault(ofport->cfm);
1976 status->remote_opstate = cfm_get_opup(ofport->cfm);
1977 status->health = cfm_get_health(ofport->cfm);
1978 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1986 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1989 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1993 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1994 cfg, ofport->up.netdev);
1995 if (ofport->bfd != old) {
1996 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2003 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2005 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2008 bfd_get_status(ofport->bfd, smap);
2015 /* Spanning Tree. */
2018 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2020 struct ofproto_dpif *ofproto = ofproto_;
2021 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2022 struct ofport_dpif *ofport;
2024 ofport = stp_port_get_aux(sp);
2026 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2027 ofproto->up.name, port_num);
2029 struct eth_header *eth = pkt->l2;
2031 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2032 if (eth_addr_is_zero(eth->eth_src)) {
2033 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2034 "with unknown MAC", ofproto->up.name, port_num);
2036 send_packet(ofport, pkt);
2042 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2044 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2048 /* Only revalidate flows if the configuration changed. */
2049 if (!s != !ofproto->stp) {
2050 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2054 if (!ofproto->stp) {
2055 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2056 send_bpdu_cb, ofproto);
2057 ofproto->stp_last_tick = time_msec();
2060 stp_set_bridge_id(ofproto->stp, s->system_id);
2061 stp_set_bridge_priority(ofproto->stp, s->priority);
2062 stp_set_hello_time(ofproto->stp, s->hello_time);
2063 stp_set_max_age(ofproto->stp, s->max_age);
2064 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2066 struct ofport *ofport;
2068 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2069 set_stp_port(ofport, NULL);
2072 stp_unref(ofproto->stp);
2073 ofproto->stp = NULL;
2080 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2086 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2087 s->designated_root = stp_get_designated_root(ofproto->stp);
2088 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2097 update_stp_port_state(struct ofport_dpif *ofport)
2099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2100 enum stp_state state;
2102 /* Figure out new state. */
2103 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2107 if (ofport->stp_state != state) {
2108 enum ofputil_port_state of_state;
2111 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2112 netdev_get_name(ofport->up.netdev),
2113 stp_state_name(ofport->stp_state),
2114 stp_state_name(state));
2115 if (stp_learn_in_state(ofport->stp_state)
2116 != stp_learn_in_state(state)) {
2117 /* xxx Learning action flows should also be flushed. */
2118 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2119 mac_learning_flush(ofproto->ml);
2120 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2122 fwd_change = stp_forward_in_state(ofport->stp_state)
2123 != stp_forward_in_state(state);
2125 ofproto->backer->need_revalidate = REV_STP;
2126 ofport->stp_state = state;
2127 ofport->stp_state_entered = time_msec();
2129 if (fwd_change && ofport->bundle) {
2130 bundle_update(ofport->bundle);
2133 /* Update the STP state bits in the OpenFlow port description. */
2134 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2135 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2136 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2137 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2138 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2140 ofproto_port_set_state(&ofport->up, of_state);
2144 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2145 * caller is responsible for assigning STP port numbers and ensuring
2146 * there are no duplicates. */
2148 set_stp_port(struct ofport *ofport_,
2149 const struct ofproto_port_stp_settings *s)
2151 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2153 struct stp_port *sp = ofport->stp_port;
2155 if (!s || !s->enable) {
2157 ofport->stp_port = NULL;
2158 stp_port_disable(sp);
2159 update_stp_port_state(ofport);
2162 } else if (sp && stp_port_no(sp) != s->port_num
2163 && ofport == stp_port_get_aux(sp)) {
2164 /* The port-id changed, so disable the old one if it's not
2165 * already in use by another port. */
2166 stp_port_disable(sp);
2169 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2170 stp_port_enable(sp);
2172 stp_port_set_aux(sp, ofport);
2173 stp_port_set_priority(sp, s->priority);
2174 stp_port_set_path_cost(sp, s->path_cost);
2176 update_stp_port_state(ofport);
2182 get_stp_port_status(struct ofport *ofport_,
2183 struct ofproto_port_stp_status *s)
2185 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2187 struct stp_port *sp = ofport->stp_port;
2189 if (!ofproto->stp || !sp) {
2195 s->port_id = stp_port_get_id(sp);
2196 s->state = stp_port_get_state(sp);
2197 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2198 s->role = stp_port_get_role(sp);
2199 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2205 stp_run(struct ofproto_dpif *ofproto)
2208 long long int now = time_msec();
2209 long long int elapsed = now - ofproto->stp_last_tick;
2210 struct stp_port *sp;
2213 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2214 ofproto->stp_last_tick = now;
2216 while (stp_get_changed_port(ofproto->stp, &sp)) {
2217 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2220 update_stp_port_state(ofport);
2224 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2225 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2226 mac_learning_flush(ofproto->ml);
2227 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2233 stp_wait(struct ofproto_dpif *ofproto)
2236 poll_timer_wait(1000);
2241 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2244 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2247 if (ofport->n_qdscp != n_qdscp
2248 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2249 n_qdscp * sizeof *qdscp))) {
2250 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2251 free(ofport->qdscp);
2252 ofport->qdscp = n_qdscp
2253 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2255 ofport->n_qdscp = n_qdscp;
2263 /* Expires all MAC learning entries associated with 'bundle' and forces its
2264 * ofproto to revalidate every flow.
2266 * Normally MAC learning entries are removed only from the ofproto associated
2267 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2268 * are removed from every ofproto. When patch ports and SLB bonds are in use
2269 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2270 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2271 * with the host from which it migrated. */
2273 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2275 struct ofproto_dpif *ofproto = bundle->ofproto;
2276 struct mac_learning *ml = ofproto->ml;
2277 struct mac_entry *mac, *next_mac;
2279 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2280 ovs_rwlock_wrlock(&ml->rwlock);
2281 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2282 if (mac->port.p == bundle) {
2284 struct ofproto_dpif *o;
2286 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2288 struct mac_entry *e;
2290 ovs_rwlock_wrlock(&o->ml->rwlock);
2291 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2293 mac_learning_expire(o->ml, e);
2295 ovs_rwlock_unlock(&o->ml->rwlock);
2300 mac_learning_expire(ml, mac);
2303 ovs_rwlock_unlock(&ml->rwlock);
2306 static struct ofbundle *
2307 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2309 struct ofbundle *bundle;
2311 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2312 &ofproto->bundles) {
2313 if (bundle->aux == aux) {
2321 bundle_update(struct ofbundle *bundle)
2323 struct ofport_dpif *port;
2325 bundle->floodable = true;
2326 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2327 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2328 || !stp_forward_in_state(port->stp_state)) {
2329 bundle->floodable = false;
2336 bundle_del_port(struct ofport_dpif *port)
2338 struct ofbundle *bundle = port->bundle;
2340 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2342 list_remove(&port->bundle_node);
2343 port->bundle = NULL;
2346 lacp_slave_unregister(bundle->lacp, port);
2349 bond_slave_unregister(bundle->bond, port);
2352 bundle_update(bundle);
2356 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2357 struct lacp_slave_settings *lacp)
2359 struct ofport_dpif *port;
2361 port = get_ofp_port(bundle->ofproto, ofp_port);
2366 if (port->bundle != bundle) {
2367 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2369 bundle_remove(&port->up);
2372 port->bundle = bundle;
2373 list_push_back(&bundle->ports, &port->bundle_node);
2374 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2375 || !stp_forward_in_state(port->stp_state)) {
2376 bundle->floodable = false;
2380 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2381 lacp_slave_register(bundle->lacp, port, lacp);
2388 bundle_destroy(struct ofbundle *bundle)
2390 struct ofproto_dpif *ofproto;
2391 struct ofport_dpif *port, *next_port;
2397 ofproto = bundle->ofproto;
2398 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2400 ovs_rwlock_wrlock(&xlate_rwlock);
2401 xlate_bundle_remove(bundle);
2402 ovs_rwlock_unlock(&xlate_rwlock);
2404 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2405 bundle_del_port(port);
2408 bundle_flush_macs(bundle, true);
2409 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2411 free(bundle->trunks);
2412 lacp_unref(bundle->lacp);
2413 bond_unref(bundle->bond);
2418 bundle_set(struct ofproto *ofproto_, void *aux,
2419 const struct ofproto_bundle_settings *s)
2421 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2422 bool need_flush = false;
2423 struct ofport_dpif *port;
2424 struct ofbundle *bundle;
2425 unsigned long *trunks;
2431 bundle_destroy(bundle_lookup(ofproto, aux));
2435 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2436 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2438 bundle = bundle_lookup(ofproto, aux);
2440 bundle = xmalloc(sizeof *bundle);
2442 bundle->ofproto = ofproto;
2443 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2444 hash_pointer(aux, 0));
2446 bundle->name = NULL;
2448 list_init(&bundle->ports);
2449 bundle->vlan_mode = PORT_VLAN_TRUNK;
2451 bundle->trunks = NULL;
2452 bundle->use_priority_tags = s->use_priority_tags;
2453 bundle->lacp = NULL;
2454 bundle->bond = NULL;
2456 bundle->floodable = true;
2457 mbridge_register_bundle(ofproto->mbridge, bundle);
2460 if (!bundle->name || strcmp(s->name, bundle->name)) {
2462 bundle->name = xstrdup(s->name);
2467 if (!bundle->lacp) {
2468 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2469 bundle->lacp = lacp_create();
2471 lacp_configure(bundle->lacp, s->lacp);
2473 lacp_unref(bundle->lacp);
2474 bundle->lacp = NULL;
2477 /* Update set of ports. */
2479 for (i = 0; i < s->n_slaves; i++) {
2480 if (!bundle_add_port(bundle, s->slaves[i],
2481 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2485 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2486 struct ofport_dpif *next_port;
2488 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2489 for (i = 0; i < s->n_slaves; i++) {
2490 if (s->slaves[i] == port->up.ofp_port) {
2495 bundle_del_port(port);
2499 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2501 if (list_is_empty(&bundle->ports)) {
2502 bundle_destroy(bundle);
2506 /* Set VLAN tagging mode */
2507 if (s->vlan_mode != bundle->vlan_mode
2508 || s->use_priority_tags != bundle->use_priority_tags) {
2509 bundle->vlan_mode = s->vlan_mode;
2510 bundle->use_priority_tags = s->use_priority_tags;
2515 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2516 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2518 if (vlan != bundle->vlan) {
2519 bundle->vlan = vlan;
2523 /* Get trunked VLANs. */
2524 switch (s->vlan_mode) {
2525 case PORT_VLAN_ACCESS:
2529 case PORT_VLAN_TRUNK:
2530 trunks = CONST_CAST(unsigned long *, s->trunks);
2533 case PORT_VLAN_NATIVE_UNTAGGED:
2534 case PORT_VLAN_NATIVE_TAGGED:
2535 if (vlan != 0 && (!s->trunks
2536 || !bitmap_is_set(s->trunks, vlan)
2537 || bitmap_is_set(s->trunks, 0))) {
2538 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2540 trunks = bitmap_clone(s->trunks, 4096);
2542 trunks = bitmap_allocate1(4096);
2544 bitmap_set1(trunks, vlan);
2545 bitmap_set0(trunks, 0);
2547 trunks = CONST_CAST(unsigned long *, s->trunks);
2554 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2555 free(bundle->trunks);
2556 if (trunks == s->trunks) {
2557 bundle->trunks = vlan_bitmap_clone(trunks);
2559 bundle->trunks = trunks;
2564 if (trunks != s->trunks) {
2569 if (!list_is_short(&bundle->ports)) {
2570 bundle->ofproto->has_bonded_bundles = true;
2572 if (bond_reconfigure(bundle->bond, s->bond)) {
2573 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2576 bundle->bond = bond_create(s->bond);
2577 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2580 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2581 bond_slave_register(bundle->bond, port, port->up.netdev);
2584 bond_unref(bundle->bond);
2585 bundle->bond = NULL;
2588 /* If we changed something that would affect MAC learning, un-learn
2589 * everything on this port and force flow revalidation. */
2591 bundle_flush_macs(bundle, false);
2598 bundle_remove(struct ofport *port_)
2600 struct ofport_dpif *port = ofport_dpif_cast(port_);
2601 struct ofbundle *bundle = port->bundle;
2604 bundle_del_port(port);
2605 if (list_is_empty(&bundle->ports)) {
2606 bundle_destroy(bundle);
2607 } else if (list_is_short(&bundle->ports)) {
2608 bond_unref(bundle->bond);
2609 bundle->bond = NULL;
2615 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2617 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2618 struct ofport_dpif *port = port_;
2619 uint8_t ea[ETH_ADDR_LEN];
2622 error = netdev_get_etheraddr(port->up.netdev, ea);
2624 struct ofpbuf packet;
2627 ofpbuf_init(&packet, 0);
2628 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2630 memcpy(packet_pdu, pdu, pdu_size);
2632 send_packet(port, &packet);
2633 ofpbuf_uninit(&packet);
2635 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2636 "%s (%s)", port->bundle->name,
2637 netdev_get_name(port->up.netdev), ovs_strerror(error));
2642 bundle_send_learning_packets(struct ofbundle *bundle)
2644 struct ofproto_dpif *ofproto = bundle->ofproto;
2645 struct ofpbuf *learning_packet;
2646 int error, n_packets, n_errors;
2647 struct mac_entry *e;
2648 struct list packets;
2650 list_init(&packets);
2651 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2652 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2653 if (e->port.p != bundle) {
2656 learning_packet = bond_compose_learning_packet(bundle->bond,
2659 learning_packet->private_p = port_void;
2660 list_push_back(&packets, &learning_packet->list_node);
2663 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2665 error = n_packets = n_errors = 0;
2666 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2669 ret = send_packet(learning_packet->private_p, learning_packet);
2676 ofpbuf_list_delete(&packets);
2679 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2680 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2681 "packets, last error was: %s",
2682 bundle->name, n_errors, n_packets, ovs_strerror(error));
2684 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2685 bundle->name, n_packets);
2690 bundle_run(struct ofbundle *bundle)
2693 lacp_run(bundle->lacp, send_pdu_cb);
2696 struct ofport_dpif *port;
2698 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2699 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2702 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2703 bundle->ofproto->backer->need_revalidate = REV_BOND;
2706 if (bond_should_send_learning_packets(bundle->bond)) {
2707 bundle_send_learning_packets(bundle);
2713 bundle_wait(struct ofbundle *bundle)
2716 lacp_wait(bundle->lacp);
2719 bond_wait(bundle->bond);
2726 mirror_set__(struct ofproto *ofproto_, void *aux,
2727 const struct ofproto_mirror_settings *s)
2729 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2730 struct ofbundle **srcs, **dsts;
2735 mirror_destroy(ofproto->mbridge, aux);
2739 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2740 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2742 for (i = 0; i < s->n_srcs; i++) {
2743 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2746 for (i = 0; i < s->n_dsts; i++) {
2747 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2750 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2751 s->n_dsts, s->src_vlans,
2752 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2759 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2760 uint64_t *packets, uint64_t *bytes)
2763 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2768 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2771 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2772 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2773 mac_learning_flush(ofproto->ml);
2775 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2780 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2783 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2784 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2788 forward_bpdu_changed(struct ofproto *ofproto_)
2790 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2791 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2795 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2799 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2800 mac_learning_set_idle_time(ofproto->ml, idle_time);
2801 mac_learning_set_max_entries(ofproto->ml, max_entries);
2802 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2807 static struct ofport_dpif *
2808 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2810 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2811 return ofport ? ofport_dpif_cast(ofport) : NULL;
2814 static struct ofport_dpif *
2815 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2817 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2818 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2822 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2823 struct ofproto_port *ofproto_port,
2824 struct dpif_port *dpif_port)
2826 ofproto_port->name = dpif_port->name;
2827 ofproto_port->type = dpif_port->type;
2828 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2832 ofport_update_peer(struct ofport_dpif *ofport)
2834 const struct ofproto_dpif *ofproto;
2835 struct dpif_backer *backer;
2838 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2842 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2843 backer->need_revalidate = REV_RECONFIGURE;
2846 ofport->peer->peer = NULL;
2847 ofport->peer = NULL;
2850 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2855 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2856 struct ofport *peer_ofport;
2857 struct ofport_dpif *peer;
2860 if (ofproto->backer != backer) {
2864 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2869 peer = ofport_dpif_cast(peer_ofport);
2870 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2871 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2873 ofport->peer = peer;
2874 ofport->peer->peer = ofport;
2884 port_run_fast(struct ofport_dpif *ofport)
2886 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2887 struct ofpbuf packet;
2889 ofpbuf_init(&packet, 0);
2890 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2891 send_packet(ofport, &packet);
2892 ofpbuf_uninit(&packet);
2895 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2896 struct ofpbuf packet;
2898 ofpbuf_init(&packet, 0);
2899 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2900 send_packet(ofport, &packet);
2901 ofpbuf_uninit(&packet);
2906 port_run(struct ofport_dpif *ofport)
2908 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2909 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2910 bool enable = netdev_get_carrier(ofport->up.netdev);
2911 bool cfm_enable = false;
2912 bool bfd_enable = false;
2914 ofport->carrier_seq = carrier_seq;
2916 port_run_fast(ofport);
2919 int cfm_opup = cfm_get_opup(ofport->cfm);
2921 cfm_run(ofport->cfm);
2922 cfm_enable = !cfm_get_fault(ofport->cfm);
2924 if (cfm_opup >= 0) {
2925 cfm_enable = cfm_enable && cfm_opup;
2930 bfd_run(ofport->bfd);
2931 bfd_enable = bfd_forwarding(ofport->bfd);
2934 if (ofport->bfd || ofport->cfm) {
2935 enable = enable && (cfm_enable || bfd_enable);
2938 if (ofport->bundle) {
2939 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2940 if (carrier_changed) {
2941 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2945 if (ofport->may_enable != enable) {
2946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2947 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2950 ofport->may_enable = enable;
2954 port_wait(struct ofport_dpif *ofport)
2957 cfm_wait(ofport->cfm);
2961 bfd_wait(ofport->bfd);
2966 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2967 struct ofproto_port *ofproto_port)
2969 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2970 struct dpif_port dpif_port;
2973 if (sset_contains(&ofproto->ghost_ports, devname)) {
2974 const char *type = netdev_get_type_from_name(devname);
2976 /* We may be called before ofproto->up.port_by_name is populated with
2977 * the appropriate ofport. For this reason, we must get the name and
2978 * type from the netdev layer directly. */
2980 const struct ofport *ofport;
2982 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2983 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2984 ofproto_port->name = xstrdup(devname);
2985 ofproto_port->type = xstrdup(type);
2991 if (!sset_contains(&ofproto->ports, devname)) {
2994 error = dpif_port_query_by_name(ofproto->backer->dpif,
2995 devname, &dpif_port);
2997 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3003 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3006 const char *devname = netdev_get_name(netdev);
3007 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3008 const char *dp_port_name;
3010 if (netdev_vport_is_patch(netdev)) {
3011 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3015 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3016 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3017 odp_port_t port_no = ODPP_NONE;
3020 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3024 if (netdev_get_tunnel_config(netdev)) {
3025 simap_put(&ofproto->backer->tnl_backers,
3026 dp_port_name, odp_to_u32(port_no));
3030 if (netdev_get_tunnel_config(netdev)) {
3031 sset_add(&ofproto->ghost_ports, devname);
3033 sset_add(&ofproto->ports, devname);
3039 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3041 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3042 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3049 sset_find_and_delete(&ofproto->ghost_ports,
3050 netdev_get_name(ofport->up.netdev));
3051 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3052 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3053 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3055 /* The caller is going to close ofport->up.netdev. If this is a
3056 * bonded port, then the bond is using that netdev, so remove it
3057 * from the bond. The client will need to reconfigure everything
3058 * after deleting ports, so then the slave will get re-added. */
3059 bundle_remove(&ofport->up);
3066 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3068 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3073 error = netdev_get_stats(ofport->up.netdev, stats);
3075 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3078 /* ofproto->stats.tx_packets represents packets that we created
3079 * internally and sent to some port (e.g. packets sent with
3080 * send_packet()). Account for them as if they had come from
3081 * OFPP_LOCAL and got forwarded. */
3083 if (stats->rx_packets != UINT64_MAX) {
3084 stats->rx_packets += ofproto->stats.tx_packets;
3087 if (stats->rx_bytes != UINT64_MAX) {
3088 stats->rx_bytes += ofproto->stats.tx_bytes;
3091 /* ofproto->stats.rx_packets represents packets that were received on
3092 * some port and we processed internally and dropped (e.g. STP).
3093 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3095 if (stats->tx_packets != UINT64_MAX) {
3096 stats->tx_packets += ofproto->stats.rx_packets;
3099 if (stats->tx_bytes != UINT64_MAX) {
3100 stats->tx_bytes += ofproto->stats.rx_bytes;
3107 struct port_dump_state {
3112 struct ofproto_port port;
3117 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3119 *statep = xzalloc(sizeof(struct port_dump_state));
3124 port_dump_next(const struct ofproto *ofproto_, void *state_,
3125 struct ofproto_port *port)
3127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3128 struct port_dump_state *state = state_;
3129 const struct sset *sset;
3130 struct sset_node *node;
3132 if (state->has_port) {
3133 ofproto_port_destroy(&state->port);
3134 state->has_port = false;
3136 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3137 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3140 error = port_query_by_name(ofproto_, node->name, &state->port);
3142 *port = state->port;
3143 state->has_port = true;
3145 } else if (error != ENODEV) {
3150 if (!state->ghost) {
3151 state->ghost = true;
3154 return port_dump_next(ofproto_, state_, port);
3161 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3163 struct port_dump_state *state = state_;
3165 if (state->has_port) {
3166 ofproto_port_destroy(&state->port);
3173 port_poll(const struct ofproto *ofproto_, char **devnamep)
3175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3177 if (ofproto->port_poll_errno) {
3178 int error = ofproto->port_poll_errno;
3179 ofproto->port_poll_errno = 0;
3183 if (sset_is_empty(&ofproto->port_poll_set)) {
3187 *devnamep = sset_pop(&ofproto->port_poll_set);
3192 port_poll_wait(const struct ofproto *ofproto_)
3194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3195 dpif_port_poll_wait(ofproto->backer->dpif);
3199 port_is_lacp_current(const struct ofport *ofport_)
3201 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3202 return (ofport->bundle && ofport->bundle->lacp
3203 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3207 /* Upcall handling. */
3209 struct flow_miss_op {
3210 struct dpif_op dpif_op;
3212 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3213 struct xlate_out xout;
3214 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3216 struct ofpbuf mask; /* Flow mask for "put" ops. */
3217 struct odputil_keybuf maskbuf;
3219 /* If this is a "put" op, then a pointer to the subfacet that should
3220 * be marked as uninstalled if the operation fails. */
3221 struct subfacet *subfacet;
3224 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3225 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3226 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3227 * return value of true). However, for short flows the cost of bookkeeping is
3228 * much higher than the benefits, so when the datapath holds a large number of
3229 * flows we impose some heuristics to decide which flows are likely to be worth
3232 flow_miss_should_make_facet(struct flow_miss *miss)
3234 struct dpif_backer *backer = miss->ofproto->backer;
3237 switch (flow_miss_model) {
3238 case OFPROTO_HANDLE_MISS_AUTO:
3240 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3242 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3246 if (!backer->governor) {
3249 n_subfacets = hmap_count(&backer->subfacets);
3250 if (n_subfacets * 2 <= flow_eviction_threshold) {
3254 backer->governor = governor_create();
3257 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3258 return governor_should_install_flow(backer->governor, hash,
3259 list_size(&miss->packets));
3262 /* Handles 'miss', which matches 'facet'. May add any required datapath
3263 * operations to 'ops', incrementing '*n_ops' for each new op.
3265 * All of the packets in 'miss' are considered to have arrived at time
3266 * 'miss->stats.used'. This is really important only for new facets: if we
3267 * just called time_msec() here, then the new subfacet or its packets could
3268 * look (occasionally) as though it was used some time after the facet was
3269 * used. That can make a one-packet flow look like it has a nonzero duration,
3270 * which looks odd in e.g. NetFlow statistics. */
3272 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3273 struct flow_miss_op *ops, size_t *n_ops)
3275 enum subfacet_path want_path;
3276 struct subfacet *subfacet;
3278 facet->packet_count += miss->stats.n_packets;
3279 facet->prev_packet_count += miss->stats.n_packets;
3280 facet->byte_count += miss->stats.n_bytes;
3281 facet->prev_byte_count += miss->stats.n_bytes;
3283 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3285 /* Don't install the flow if it's the result of the "userspace"
3286 * action for an already installed facet. This can occur when a
3287 * datapath flow with wildcards has a "userspace" action and flows
3288 * sent to userspace result in a different subfacet, which will then
3289 * be rejected as overlapping by the datapath. */
3290 if (miss->upcall_type == DPIF_UC_ACTION
3291 && !list_is_empty(&facet->subfacets)) {
3295 subfacet = subfacet_create(facet, miss);
3296 if (subfacet->path != want_path) {
3297 struct flow_miss_op *op = &ops[(*n_ops)++];
3298 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3300 subfacet->path = want_path;
3302 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3303 if (enable_megaflows) {
3304 mask_to_netlink_attr(&op->mask, facet->ofproto,
3305 &facet->xout.wc.masks, &miss->flow);
3308 op->xout_garbage = false;
3309 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3310 op->subfacet = subfacet;
3311 put->flags = DPIF_FP_CREATE;
3312 put->key = miss->key;
3313 put->key_len = miss->key_len;
3314 put->mask = op->mask.data;
3315 put->mask_len = op->mask.size;
3317 if (want_path == SF_FAST_PATH) {
3318 put->actions = facet->xout.odp_actions.data;
3319 put->actions_len = facet->xout.odp_actions.size;
3321 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3322 op->slow_stub, sizeof op->slow_stub,
3323 &put->actions, &put->actions_len);
3329 /* Handles flow miss 'miss'. May add any required datapath operations
3330 * to 'ops', incrementing '*n_ops' for each new op. */
3332 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3335 struct facet *facet;
3337 miss->ofproto->n_missed += list_size(&miss->packets);
3339 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3341 /* There does not exist a bijection between 'struct flow' and datapath
3342 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3343 * assumption used throughout the facet and subfacet handling code.
3344 * Since we have to handle these misses in userspace anyway, we simply
3345 * skip facet creation, avoiding the problem altogether. */
3346 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3347 || !flow_miss_should_make_facet(miss)) {
3351 facet = facet_create(miss);
3353 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3356 static struct drop_key *
3357 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3360 struct drop_key *drop_key;
3362 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3363 &backer->drop_keys) {
3364 if (drop_key->key_len == key_len
3365 && !memcmp(drop_key->key, key, key_len)) {
3373 drop_key_clear(struct dpif_backer *backer)
3375 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3376 struct drop_key *drop_key, *next;
3378 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3381 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3383 if (error && !VLOG_DROP_WARN(&rl)) {
3384 struct ds ds = DS_EMPTY_INITIALIZER;
3385 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3386 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3387 ovs_strerror(error), ds_cstr(&ds));
3391 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3392 drop_key_destroy(drop_key);
3395 udpif_drop_key_clear(backer->udpif);
3399 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3401 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3402 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3403 struct flow_miss *miss;
3406 /* Process each element in the to-do list, constructing the set of
3407 * operations to batch. */
3409 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3410 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3412 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3414 /* Execute batch. */
3415 for (i = 0; i < n_ops; i++) {
3416 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3418 dpif_operate(backer->dpif, dpif_ops, n_ops);
3420 for (i = 0; i < n_ops; i++) {
3421 if (dpif_ops[i]->error != 0
3422 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3423 && flow_miss_ops[i].subfacet) {
3424 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3426 COVERAGE_INC(subfacet_install_fail);
3428 /* Zero-out subfacet counters when installation failed, but
3429 * datapath reported hits. This should not happen and
3430 * indicates a bug, since if the datapath flow exists, we
3431 * should not be attempting to create a new subfacet. A
3432 * buggy datapath could trigger this, so just zero out the
3433 * counters and log an error. */
3434 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3435 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3436 "datapath reported hits");
3437 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3440 subfacet->path = SF_NOT_INSTALLED;
3446 handle_sflow_upcall(struct dpif_backer *backer,
3447 const struct dpif_upcall *upcall)
3449 struct ofproto_dpif *ofproto;
3450 union user_action_cookie cookie;
3452 odp_port_t odp_in_port;
3454 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3455 &flow, NULL, &ofproto, &odp_in_port)
3456 || !ofproto->sflow) {
3460 memset(&cookie, 0, sizeof cookie);
3461 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3462 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3463 odp_in_port, &cookie);
3467 handle_flow_sample_upcall(struct dpif_backer *backer,
3468 const struct dpif_upcall *upcall)
3470 struct ofproto_dpif *ofproto;
3471 union user_action_cookie cookie;
3474 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3475 &flow, NULL, &ofproto, NULL)
3476 || !ofproto->ipfix) {
3480 memset(&cookie, 0, sizeof cookie);
3481 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3483 /* The flow reflects exactly the contents of the packet. Sample
3484 * the packet using it. */
3485 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3486 cookie.flow_sample.collector_set_id,
3487 cookie.flow_sample.probability,
3488 cookie.flow_sample.obs_domain_id,
3489 cookie.flow_sample.obs_point_id);
3493 handle_ipfix_upcall(struct dpif_backer *backer,
3494 const struct dpif_upcall *upcall)
3496 struct ofproto_dpif *ofproto;
3499 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3500 &flow, NULL, &ofproto, NULL)
3501 || !ofproto->ipfix) {
3505 /* The flow reflects exactly the contents of the packet. Sample
3506 * the packet using it. */
3507 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3511 handle_upcalls(struct dpif_backer *backer)
3513 struct flow_miss_batch *fmb;
3516 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3517 struct upcall *upcall = upcall_next(backer->udpif);
3523 switch (upcall->type) {
3525 handle_sflow_upcall(backer, &upcall->dpif_upcall);
3528 case FLOW_SAMPLE_UPCALL:
3529 handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3533 handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3543 upcall_destroy(upcall);
3546 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3547 struct drop_key *drop_key = drop_key_next(backer->udpif);
3552 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3553 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3554 hash_bytes(drop_key->key, drop_key->key_len, 0));
3555 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3556 drop_key->key, drop_key->key_len,
3557 NULL, 0, NULL, 0, NULL);
3559 drop_key_destroy(drop_key);
3563 fmb = flow_miss_batch_next(backer->udpif);
3565 handle_flow_misses(backer, fmb);
3566 flow_miss_batch_destroy(fmb);
3570 /* Flow expiration. */
3572 static int subfacet_max_idle(const struct dpif_backer *);
3573 static void update_stats(struct dpif_backer *);
3574 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3575 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3577 /* This function is called periodically by run(). Its job is to collect
3578 * updates for the flows that have been installed into the datapath, most
3579 * importantly when they last were used, and then use that information to
3580 * expire flows that have not been used recently.
3582 * Returns the number of milliseconds after which it should be called again. */
3584 expire(struct dpif_backer *backer)
3586 struct ofproto_dpif *ofproto;
3590 /* Periodically clear out the drop keys in an effort to keep them
3591 * relatively few. */
3592 drop_key_clear(backer);
3594 /* Update stats for each flow in the backer. */
3595 update_stats(backer);
3597 n_subfacets = hmap_count(&backer->subfacets);
3599 struct subfacet *subfacet;
3600 long long int total, now;
3604 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3605 total += now - subfacet->created;
3607 backer->avg_subfacet_life += total / n_subfacets;
3609 backer->avg_subfacet_life /= 2;
3611 backer->avg_n_subfacet += n_subfacets;
3612 backer->avg_n_subfacet /= 2;
3614 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3616 max_idle = subfacet_max_idle(backer);
3617 expire_subfacets(backer, max_idle);
3619 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3620 struct rule *rule, *next_rule;
3622 if (ofproto->backer != backer) {
3626 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3628 ovs_mutex_lock(&ofproto_mutex);
3629 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3630 &ofproto->up.expirable) {
3631 rule_expire(rule_dpif_cast(rule));
3633 ovs_mutex_unlock(&ofproto_mutex);
3635 /* All outstanding data in existing flows has been accounted, so it's a
3636 * good time to do bond rebalancing. */
3637 if (ofproto->has_bonded_bundles) {
3638 struct ofbundle *bundle;
3640 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3642 bond_rebalance(bundle->bond);
3648 return MIN(max_idle, 1000);
3651 /* Updates flow table statistics given that the datapath just reported 'stats'
3652 * as 'subfacet''s statistics. */
3654 update_subfacet_stats(struct subfacet *subfacet,
3655 const struct dpif_flow_stats *stats)
3657 struct facet *facet = subfacet->facet;
3658 struct dpif_flow_stats diff;
3660 diff.tcp_flags = stats->tcp_flags;
3661 diff.used = stats->used;
3663 if (stats->n_packets >= subfacet->dp_packet_count) {
3664 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3666 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3670 if (stats->n_bytes >= subfacet->dp_byte_count) {
3671 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3673 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3677 facet->ofproto->n_hit += diff.n_packets;
3678 subfacet->dp_packet_count = stats->n_packets;
3679 subfacet->dp_byte_count = stats->n_bytes;
3680 subfacet_update_stats(subfacet, &diff);
3682 if (facet->accounted_bytes < facet->byte_count) {
3684 facet_account(facet);
3685 facet->accounted_bytes = facet->byte_count;
3689 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3690 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3692 delete_unexpected_flow(struct dpif_backer *backer,
3693 const struct nlattr *key, size_t key_len)
3695 if (!VLOG_DROP_WARN(&rl)) {
3699 odp_flow_key_format(key, key_len, &s);
3700 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3704 COVERAGE_INC(facet_unexpected);
3705 dpif_flow_del(backer->dpif, key, key_len, NULL);
3708 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3710 * This function also pushes statistics updates to rules which each facet
3711 * resubmits into. Generally these statistics will be accurate. However, if a
3712 * facet changes the rule it resubmits into at some time in between
3713 * update_stats() runs, it is possible that statistics accrued to the
3714 * old rule will be incorrectly attributed to the new rule. This could be
3715 * avoided by calling update_stats() whenever rules are created or
3716 * deleted. However, the performance impact of making so many calls to the
3717 * datapath do not justify the benefit of having perfectly accurate statistics.
3719 * In addition, this function maintains per ofproto flow hit counts. The patch
3720 * port is not treated specially. e.g. A packet ingress from br0 patched into
3721 * br1 will increase the hit count of br0 by 1, however, does not affect
3722 * the hit or miss counts of br1.
3725 update_stats(struct dpif_backer *backer)
3727 const struct dpif_flow_stats *stats;
3728 struct dpif_flow_dump dump;
3729 const struct nlattr *key, *mask;
3730 size_t key_len, mask_len;
3732 dpif_flow_dump_start(&dump, backer->dpif);
3733 while (dpif_flow_dump_next(&dump, &key, &key_len,
3734 &mask, &mask_len, NULL, NULL, &stats)) {
3735 struct subfacet *subfacet;
3738 key_hash = odp_flow_key_hash(key, key_len);
3739 subfacet = subfacet_find(backer, key, key_len, key_hash);
3740 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3742 update_subfacet_stats(subfacet, stats);
3746 /* Stats are updated per-packet. */
3749 case SF_NOT_INSTALLED:
3751 delete_unexpected_flow(backer, key, key_len);
3756 dpif_flow_dump_done(&dump);
3758 update_moving_averages(backer);
3761 /* Calculates and returns the number of milliseconds of idle time after which
3762 * subfacets should expire from the datapath. When a subfacet expires, we fold
3763 * its statistics into its facet, and when a facet's last subfacet expires, we
3764 * fold its statistic into its rule. */
3766 subfacet_max_idle(const struct dpif_backer *backer)
3769 * Idle time histogram.
3771 * Most of the time a switch has a relatively small number of subfacets.
3772 * When this is the case we might as well keep statistics for all of them
3773 * in userspace and to cache them in the kernel datapath for performance as
3776 * As the number of subfacets increases, the memory required to maintain
3777 * statistics about them in userspace and in the kernel becomes
3778 * significant. However, with a large number of subfacets it is likely
3779 * that only a few of them are "heavy hitters" that consume a large amount
3780 * of bandwidth. At this point, only heavy hitters are worth caching in
3781 * the kernel and maintaining in userspaces; other subfacets we can
3784 * The technique used to compute the idle time is to build a histogram with
3785 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3786 * that is installed in the kernel gets dropped in the appropriate bucket.
3787 * After the histogram has been built, we compute the cutoff so that only
3788 * the most-recently-used 1% of subfacets (but at least
3789 * flow_eviction_threshold flows) are kept cached. At least
3790 * the most-recently-used bucket of subfacets is kept, so actually an
3791 * arbitrary number of subfacets can be kept in any given expiration run
3792 * (though the next run will delete most of those unless they receive
3795 * This requires a second pass through the subfacets, in addition to the
3796 * pass made by update_stats(), because the former function never looks at
3797 * uninstallable subfacets.
3799 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3800 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3801 int buckets[N_BUCKETS] = { 0 };
3802 int total, subtotal, bucket;
3803 struct subfacet *subfacet;
3807 /* If ofproto_max_idle is specified, uses it instead of doing the
3809 if (ofproto_max_idle) {
3810 return ofproto_max_idle;
3813 total = hmap_count(&backer->subfacets);
3814 if (total <= flow_eviction_threshold) {
3815 return N_BUCKETS * BUCKET_WIDTH;
3818 /* Build histogram. */
3820 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3821 long long int idle = now - subfacet->used;
3822 int bucket = (idle <= 0 ? 0
3823 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3824 : (unsigned int) idle / BUCKET_WIDTH);
3828 /* Find the first bucket whose flows should be expired. */
3829 subtotal = bucket = 0;
3831 subtotal += buckets[bucket++];
3832 } while (bucket < N_BUCKETS &&
3833 subtotal < MAX(flow_eviction_threshold, total / 100));
3835 if (VLOG_IS_DBG_ENABLED()) {
3839 ds_put_cstr(&s, "keep");
3840 for (i = 0; i < N_BUCKETS; i++) {
3842 ds_put_cstr(&s, ", drop");
3845 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3848 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3852 return bucket * BUCKET_WIDTH;
3856 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3858 /* Cutoff time for most flows. */
3859 long long int normal_cutoff = time_msec() - dp_max_idle;
3861 /* We really want to keep flows for special protocols around, so use a more
3862 * conservative cutoff. */
3863 long long int special_cutoff = time_msec() - 10000;
3865 struct subfacet *subfacet, *next_subfacet;
3866 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3870 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3871 &backer->subfacets) {
3872 long long int cutoff;
3874 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3878 if (subfacet->used < cutoff) {
3879 if (subfacet->path != SF_NOT_INSTALLED) {
3880 batch[n_batch++] = subfacet;
3881 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3882 subfacet_destroy_batch(backer, batch, n_batch);
3886 subfacet_destroy(subfacet);
3892 subfacet_destroy_batch(backer, batch, n_batch);
3896 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3897 * then delete it entirely. */
3899 rule_expire(struct rule_dpif *rule)
3900 OVS_REQUIRES(ofproto_mutex)
3902 uint16_t idle_timeout, hard_timeout;
3903 long long int now = time_msec();
3906 ovs_assert(!rule->up.pending);
3908 /* Has 'rule' expired? */
3909 ovs_mutex_lock(&rule->up.mutex);
3910 hard_timeout = rule->up.hard_timeout;
3911 idle_timeout = rule->up.idle_timeout;
3912 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3913 reason = OFPRR_HARD_TIMEOUT;
3914 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3915 reason = OFPRR_IDLE_TIMEOUT;
3919 ovs_mutex_unlock(&rule->up.mutex);
3922 COVERAGE_INC(ofproto_dpif_expired);
3923 ofproto_rule_expire(&rule->up, reason);
3929 /* Creates and returns a new facet based on 'miss'.
3931 * The caller must already have determined that no facet with an identical
3932 * 'miss->flow' exists in 'miss->ofproto'.
3934 * 'rule' and 'xout' must have been created based on 'miss'.
3936 * 'facet'' statistics are initialized based on 'stats'.
3938 * The facet will initially have no subfacets. The caller should create (at
3939 * least) one subfacet with subfacet_create(). */
3940 static struct facet *
3941 facet_create(const struct flow_miss *miss)
3943 struct ofproto_dpif *ofproto = miss->ofproto;
3944 struct facet *facet;
3947 facet = xzalloc(sizeof *facet);
3948 facet->ofproto = miss->ofproto;
3949 facet->used = miss->stats.used;
3950 facet->flow = miss->flow;
3951 facet->learn_rl = time_msec() + 500;
3953 list_init(&facet->subfacets);
3954 netflow_flow_init(&facet->nf_flow);
3955 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3957 xlate_out_copy(&facet->xout, &miss->xout);
3959 match_init(&match, &facet->flow, &facet->xout.wc);
3960 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3961 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3962 classifier_insert(&ofproto->facets, &facet->cr);
3963 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3965 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3970 facet_free(struct facet *facet)
3973 xlate_out_uninit(&facet->xout);
3978 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3979 * 'packet', which arrived on 'in_port'. */
3981 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3982 const struct nlattr *odp_actions, size_t actions_len,
3983 struct ofpbuf *packet)
3985 struct odputil_keybuf keybuf;
3989 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3990 odp_flow_key_from_flow(&key, flow,
3991 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3993 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3994 odp_actions, actions_len, packet);
3998 /* Remove 'facet' from its ofproto and free up the associated memory:
4000 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4001 * rule's statistics, via subfacet_uninstall().
4003 * - Removes 'facet' from its rule and from ofproto->facets.
4006 facet_remove(struct facet *facet)
4008 struct subfacet *subfacet, *next_subfacet;
4010 ovs_assert(!list_is_empty(&facet->subfacets));
4012 /* First uninstall all of the subfacets to get final statistics. */
4013 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4014 subfacet_uninstall(subfacet);
4017 /* Flush the final stats to the rule.
4019 * This might require us to have at least one subfacet around so that we
4020 * can use its actions for accounting in facet_account(), which is why we
4021 * have uninstalled but not yet destroyed the subfacets. */
4022 facet_flush_stats(facet);
4024 /* Now we're really all done so destroy everything. */
4025 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4026 &facet->subfacets) {
4027 subfacet_destroy__(subfacet);
4029 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4030 classifier_remove(&facet->ofproto->facets, &facet->cr);
4031 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4032 cls_rule_destroy(&facet->cr);
4036 /* Feed information from 'facet' back into the learning table to keep it in
4037 * sync with what is actually flowing through the datapath. */
4039 facet_learn(struct facet *facet)
4041 long long int now = time_msec();
4043 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4047 facet->learn_rl = now + 500;
4049 if (!facet->xout.has_learn
4050 && !facet->xout.has_normal
4051 && (!facet->xout.has_fin_timeout
4052 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4056 facet_push_stats(facet, true);
4060 facet_account(struct facet *facet)
4062 const struct nlattr *a;
4067 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4070 n_bytes = facet->byte_count - facet->accounted_bytes;
4072 /* This loop feeds byte counters to bond_account() for rebalancing to use
4073 * as a basis. We also need to track the actual VLAN on which the packet
4074 * is going to be sent to ensure that it matches the one passed to
4075 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4078 * We use the actions from an arbitrary subfacet because they should all
4079 * be equally valid for our purpose. */
4080 vlan_tci = facet->flow.vlan_tci;
4081 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4082 facet->xout.odp_actions.size) {
4083 const struct ovs_action_push_vlan *vlan;
4084 struct ofport_dpif *port;
4086 switch (nl_attr_type(a)) {
4087 case OVS_ACTION_ATTR_OUTPUT:
4088 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4089 if (port && port->bundle && port->bundle->bond) {
4090 bond_account(port->bundle->bond, &facet->flow,
4091 vlan_tci_to_vid(vlan_tci), n_bytes);
4095 case OVS_ACTION_ATTR_POP_VLAN:
4096 vlan_tci = htons(0);
4099 case OVS_ACTION_ATTR_PUSH_VLAN:
4100 vlan = nl_attr_get(a);
4101 vlan_tci = vlan->vlan_tci;
4107 /* Returns true if the only action for 'facet' is to send to the controller.
4108 * (We don't report NetFlow expiration messages for such facets because they
4109 * are just part of the control logic for the network, not real traffic). */
4111 facet_is_controller_flow(struct facet *facet)
4114 struct ofproto_dpif *ofproto = facet->ofproto;
4115 const struct ofpact *ofpacts;
4116 struct rule_actions *actions;
4117 struct rule_dpif *rule;
4121 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4122 actions = rule_dpif_get_actions(rule);
4123 rule_dpif_unref(rule);
4125 ofpacts_len = actions->ofpacts_len;
4126 ofpacts = actions->ofpacts;
4127 is_controller = ofpacts_len > 0
4128 && ofpacts->type == OFPACT_CONTROLLER
4129 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4130 rule_actions_unref(actions);
4132 return is_controller;
4137 /* Folds all of 'facet''s statistics into its rule. Also updates the
4138 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4139 * 'facet''s statistics in the datapath should have been zeroed and folded into
4140 * its packet and byte counts before this function is called. */
4142 facet_flush_stats(struct facet *facet)
4144 struct ofproto_dpif *ofproto = facet->ofproto;
4145 struct subfacet *subfacet;
4147 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4148 ovs_assert(!subfacet->dp_byte_count);
4149 ovs_assert(!subfacet->dp_packet_count);
4152 facet_push_stats(facet, false);
4153 if (facet->accounted_bytes < facet->byte_count) {
4154 facet_account(facet);
4155 facet->accounted_bytes = facet->byte_count;
4158 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4159 struct ofexpired expired;
4160 expired.flow = facet->flow;
4161 expired.packet_count = facet->packet_count;
4162 expired.byte_count = facet->byte_count;
4163 expired.used = facet->used;
4164 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4167 /* Reset counters to prevent double counting if 'facet' ever gets
4169 facet_reset_counters(facet);
4171 netflow_flow_clear(&facet->nf_flow);
4172 facet->tcp_flags = 0;
4175 /* Searches 'ofproto''s table of facets for one which would be responsible for
4176 * 'flow'. Returns it if found, otherwise a null pointer.
4178 * The returned facet might need revalidation; use facet_lookup_valid()
4179 * instead if that is important. */
4180 static struct facet *
4181 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4183 struct cls_rule *cr;
4185 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4186 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4187 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4188 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4191 /* Searches 'ofproto''s table of facets for one capable that covers
4192 * 'flow'. Returns it if found, otherwise a null pointer.
4194 * The returned facet is guaranteed to be valid. */
4195 static struct facet *
4196 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4198 struct facet *facet;
4200 facet = facet_find(ofproto, flow);
4202 && ofproto->backer->need_revalidate
4203 && !facet_revalidate(facet)) {
4211 facet_check_consistency(struct facet *facet)
4213 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4215 struct xlate_out xout;
4216 struct xlate_in xin;
4219 /* Check the datapath actions for consistency. */
4220 xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4221 xlate_actions(&xin, &xout);
4223 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4224 && facet->xout.slow == xout.slow;
4225 if (!ok && !VLOG_DROP_WARN(&rl)) {
4226 struct ds s = DS_EMPTY_INITIALIZER;
4228 flow_format(&s, &facet->flow);
4229 ds_put_cstr(&s, ": inconsistency in facet");
4231 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4232 ds_put_cstr(&s, " (actions were: ");
4233 format_odp_actions(&s, facet->xout.odp_actions.data,
4234 facet->xout.odp_actions.size);
4235 ds_put_cstr(&s, ") (correct actions: ");
4236 format_odp_actions(&s, xout.odp_actions.data,
4237 xout.odp_actions.size);
4238 ds_put_char(&s, ')');
4241 if (facet->xout.slow != xout.slow) {
4242 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4247 xlate_out_uninit(&xout);
4252 /* Re-searches the classifier for 'facet':
4254 * - If the rule found is different from 'facet''s current rule, moves
4255 * 'facet' to the new rule and recompiles its actions.
4257 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4258 * where it is and recompiles its actions anyway.
4260 * - If any of 'facet''s subfacets correspond to a new flow according to
4261 * xlate_receive(), 'facet' is removed.
4263 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4265 facet_revalidate(struct facet *facet)
4267 struct ofproto_dpif *ofproto = facet->ofproto;
4268 struct rule_dpif *new_rule;
4269 struct subfacet *subfacet;
4270 struct flow_wildcards wc;
4271 struct xlate_out xout;
4272 struct xlate_in xin;
4274 COVERAGE_INC(facet_revalidate);
4276 /* Check that child subfacets still correspond to this facet. Tunnel
4277 * configuration changes could cause a subfacet's OpenFlow in_port to
4279 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4280 struct ofproto_dpif *recv_ofproto;
4281 struct flow recv_flow;
4284 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4285 subfacet->key_len, &recv_flow, NULL,
4286 &recv_ofproto, NULL);
4288 || recv_ofproto != ofproto
4289 || facet != facet_find(ofproto, &recv_flow)) {
4290 facet_remove(facet);
4295 flow_wildcards_init_catchall(&wc);
4296 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4298 /* Calculate new datapath actions.
4300 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4301 * emit a NetFlow expiration and, if so, we need to have the old state
4302 * around to properly compose it. */
4303 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4304 xlate_actions(&xin, &xout);
4305 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4306 /* Make sure non -packet fields are not masked. If not cleared,
4307 * the memcmp() below may fail, causing an otherwise valid facet
4309 flow_wildcards_clear_non_packet_fields(&xout.wc);
4311 /* A facet's slow path reason should only change under dramatic
4312 * circumstances. Rather than try to update everything, it's simpler to
4313 * remove the facet and start over.
4315 * More importantly, if a facet's wildcards change, it will be relatively
4316 * difficult to figure out if its subfacets still belong to it, and if not
4317 * which facet they may belong to. Again, to avoid the complexity, we
4318 * simply give up instead. */
4319 if (facet->xout.slow != xout.slow
4320 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4321 facet_remove(facet);
4322 xlate_out_uninit(&xout);
4323 rule_dpif_unref(new_rule);
4327 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4328 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4329 if (subfacet->path == SF_FAST_PATH) {
4330 struct dpif_flow_stats stats;
4332 subfacet_install(subfacet, &xout.odp_actions, &stats);
4333 subfacet_update_stats(subfacet, &stats);
4337 facet_flush_stats(facet);
4339 ofpbuf_clear(&facet->xout.odp_actions);
4340 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4341 xout.odp_actions.size);
4344 /* Update 'facet' now that we've taken care of all the old state. */
4345 facet->xout.slow = xout.slow;
4346 facet->xout.has_learn = xout.has_learn;
4347 facet->xout.has_normal = xout.has_normal;
4348 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4349 facet->xout.nf_output_iface = xout.nf_output_iface;
4350 facet->xout.mirrors = xout.mirrors;
4351 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4353 ovs_mutex_lock(&new_rule->up.mutex);
4354 facet->used = MAX(facet->used, new_rule->up.created);
4355 ovs_mutex_unlock(&new_rule->up.mutex);
4357 xlate_out_uninit(&xout);
4358 rule_dpif_unref(new_rule);
4363 facet_reset_counters(struct facet *facet)
4365 facet->packet_count = 0;
4366 facet->byte_count = 0;
4367 facet->prev_packet_count = 0;
4368 facet->prev_byte_count = 0;
4369 facet->accounted_bytes = 0;
4373 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4374 struct dpif_flow_stats *stats, bool may_learn)
4376 struct ofport_dpif *in_port;
4377 struct xlate_in xin;
4379 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4380 if (in_port && in_port->is_tunnel) {
4381 netdev_vport_inc_rx(in_port->up.netdev, stats);
4384 xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4385 xin.resubmit_stats = stats;
4386 xin.may_learn = may_learn;
4387 xlate_actions_for_side_effects(&xin);
4391 facet_push_stats(struct facet *facet, bool may_learn)
4393 struct dpif_flow_stats stats;
4395 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4396 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4397 ovs_assert(facet->used >= facet->prev_used);
4399 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4400 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4401 stats.used = facet->used;
4402 stats.tcp_flags = facet->tcp_flags;
4404 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4405 facet->prev_packet_count = facet->packet_count;
4406 facet->prev_byte_count = facet->byte_count;
4407 facet->prev_used = facet->used;
4409 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4411 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4412 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4413 stats.n_packets, stats.n_bytes);
4414 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4419 push_all_stats__(bool run_fast)
4421 static long long int rl = LLONG_MIN;
4422 struct ofproto_dpif *ofproto;
4424 if (time_msec() < rl) {
4428 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4429 struct cls_cursor cursor;
4430 struct facet *facet;
4432 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4433 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4434 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4435 facet_push_stats(facet, false);
4440 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4443 rl = time_msec() + 100;
4447 push_all_stats(void)
4449 push_all_stats__(true);
4453 rule_dpif_credit_stats(struct rule_dpif *rule,
4454 const struct dpif_flow_stats *stats)
4456 ovs_mutex_lock(&rule->stats_mutex);
4457 rule->packet_count += stats->n_packets;
4458 rule->byte_count += stats->n_bytes;
4459 rule->up.used = MAX(rule->up.used, stats->used);
4460 ovs_mutex_unlock(&rule->stats_mutex);
4464 rule_dpif_fail_open(const struct rule_dpif *rule)
4466 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4470 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4471 OVS_REQUIRES(rule->up.mutex)
4473 return rule->up.flow_cookie;
4477 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4478 uint16_t hard_timeout)
4480 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4483 /* Returns 'rule''s actions. The caller owns a reference on the returned
4484 * actions and must eventually release it (with rule_actions_unref()) to avoid
4486 struct rule_actions *
4487 rule_dpif_get_actions(const struct rule_dpif *rule)
4489 return rule_get_actions(&rule->up);
4494 static struct subfacet *
4495 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4496 size_t key_len, uint32_t key_hash)
4498 struct subfacet *subfacet;
4500 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4501 &backer->subfacets) {
4502 if (subfacet->key_len == key_len
4503 && !memcmp(key, subfacet->key, key_len)) {
4511 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4512 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4513 * existing subfacet if there is one, otherwise creates and returns a
4515 static struct subfacet *
4516 subfacet_create(struct facet *facet, struct flow_miss *miss)
4518 struct dpif_backer *backer = miss->ofproto->backer;
4519 enum odp_key_fitness key_fitness = miss->key_fitness;
4520 const struct nlattr *key = miss->key;
4521 size_t key_len = miss->key_len;
4523 struct subfacet *subfacet;
4525 key_hash = odp_flow_key_hash(key, key_len);
4527 if (list_is_empty(&facet->subfacets)) {
4528 subfacet = &facet->one_subfacet;
4530 subfacet = subfacet_find(backer, key, key_len, key_hash);
4532 if (subfacet->facet == facet) {
4536 /* This shouldn't happen. */
4537 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4538 subfacet_destroy(subfacet);
4541 subfacet = xmalloc(sizeof *subfacet);
4544 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4545 list_push_back(&facet->subfacets, &subfacet->list_node);
4546 subfacet->facet = facet;
4547 subfacet->key_fitness = key_fitness;
4548 subfacet->key = xmemdup(key, key_len);
4549 subfacet->key_len = key_len;
4550 subfacet->used = miss->stats.used;
4551 subfacet->created = subfacet->used;
4552 subfacet->dp_packet_count = 0;
4553 subfacet->dp_byte_count = 0;
4554 subfacet->path = SF_NOT_INSTALLED;
4555 subfacet->backer = backer;
4557 backer->subfacet_add_count++;
4561 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4562 * its facet within 'ofproto', and frees it. */
4564 subfacet_destroy__(struct subfacet *subfacet)
4566 struct facet *facet = subfacet->facet;
4567 struct ofproto_dpif *ofproto = facet->ofproto;
4569 /* Update ofproto stats before uninstall the subfacet. */
4570 ofproto->backer->subfacet_del_count++;
4572 subfacet_uninstall(subfacet);
4573 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4574 list_remove(&subfacet->list_node);
4575 free(subfacet->key);
4576 if (subfacet != &facet->one_subfacet) {
4581 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4582 * last remaining subfacet in its facet destroys the facet too. */
4584 subfacet_destroy(struct subfacet *subfacet)
4586 struct facet *facet = subfacet->facet;
4588 if (list_is_singleton(&facet->subfacets)) {
4589 /* facet_remove() needs at least one subfacet (it will remove it). */
4590 facet_remove(facet);
4592 subfacet_destroy__(subfacet);
4597 subfacet_destroy_batch(struct dpif_backer *backer,
4598 struct subfacet **subfacets, int n)
4600 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4601 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4602 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4605 for (i = 0; i < n; i++) {
4606 ops[i].type = DPIF_OP_FLOW_DEL;
4607 ops[i].u.flow_del.key = subfacets[i]->key;
4608 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4609 ops[i].u.flow_del.stats = &stats[i];
4613 dpif_operate(backer->dpif, opsp, n);
4614 for (i = 0; i < n; i++) {
4615 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4616 subfacets[i]->path = SF_NOT_INSTALLED;
4617 subfacet_destroy(subfacets[i]);
4622 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4623 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4624 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4625 * since 'subfacet' was last updated.
4627 * Returns 0 if successful, otherwise a positive errno value. */
4629 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4630 struct dpif_flow_stats *stats)
4632 struct facet *facet = subfacet->facet;
4633 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4634 const struct nlattr *actions = odp_actions->data;
4635 size_t actions_len = odp_actions->size;
4636 struct odputil_keybuf maskbuf;
4639 uint64_t slow_path_stub[128 / 8];
4640 enum dpif_flow_put_flags flags;
4643 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4646 flags |= DPIF_FP_ZERO_STATS;
4649 if (path == SF_SLOW_PATH) {
4650 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4651 slow_path_stub, sizeof slow_path_stub,
4652 &actions, &actions_len);
4655 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4656 if (enable_megaflows) {
4657 mask_to_netlink_attr(&mask, facet->ofproto,
4658 &facet->xout.wc.masks, &facet->flow);
4661 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4662 subfacet->key_len, mask.data, mask.size,
4663 actions, actions_len, stats);
4666 subfacet_reset_dp_stats(subfacet, stats);
4670 COVERAGE_INC(subfacet_install_fail);
4672 subfacet->path = path;
4677 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4679 subfacet_uninstall(struct subfacet *subfacet)
4681 if (subfacet->path != SF_NOT_INSTALLED) {
4682 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4683 struct dpif_flow_stats stats;
4686 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4687 subfacet->key_len, &stats);
4688 subfacet_reset_dp_stats(subfacet, &stats);
4690 subfacet_update_stats(subfacet, &stats);
4692 subfacet->path = SF_NOT_INSTALLED;
4694 ovs_assert(subfacet->dp_packet_count == 0);
4695 ovs_assert(subfacet->dp_byte_count == 0);
4699 /* Resets 'subfacet''s datapath statistics counters. This should be called
4700 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4701 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4702 * was reset in the datapath. 'stats' will be modified to include only
4703 * statistics new since 'subfacet' was last updated. */
4705 subfacet_reset_dp_stats(struct subfacet *subfacet,
4706 struct dpif_flow_stats *stats)
4709 && subfacet->dp_packet_count <= stats->n_packets
4710 && subfacet->dp_byte_count <= stats->n_bytes) {
4711 stats->n_packets -= subfacet->dp_packet_count;
4712 stats->n_bytes -= subfacet->dp_byte_count;
4715 subfacet->dp_packet_count = 0;
4716 subfacet->dp_byte_count = 0;
4719 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4721 * Because of the meaning of a subfacet's counters, it only makes sense to do
4722 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4723 * represents a packet that was sent by hand or if it represents statistics
4724 * that have been cleared out of the datapath. */
4726 subfacet_update_stats(struct subfacet *subfacet,
4727 const struct dpif_flow_stats *stats)
4729 if (stats->n_packets || stats->used > subfacet->used) {
4730 struct facet *facet = subfacet->facet;
4732 subfacet->used = MAX(subfacet->used, stats->used);
4733 facet->used = MAX(facet->used, stats->used);
4734 facet->packet_count += stats->n_packets;
4735 facet->byte_count += stats->n_bytes;
4736 facet->tcp_flags |= stats->tcp_flags;
4742 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4743 * the fields that were relevant as part of the lookup. */
4745 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4746 struct flow_wildcards *wc, struct rule_dpif **rule)
4748 struct ofport_dpif *port;
4750 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4753 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4755 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4756 flow->in_port.ofp_port);
4759 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4760 ofproto->no_packet_in_rule, rule);
4764 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4765 const struct flow *flow, struct flow_wildcards *wc,
4766 uint8_t table_id, struct rule_dpif **rule)
4768 const struct cls_rule *cls_rule;
4769 struct classifier *cls;
4773 if (table_id >= N_TABLES) {
4778 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4779 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4782 cls = &ofproto->up.tables[table_id].cls;
4783 ovs_rwlock_rdlock(&cls->rwlock);
4784 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4785 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4786 /* We must pretend that transport ports are unavailable. */
4787 struct flow ofpc_normal_flow = *flow;
4788 ofpc_normal_flow.tp_src = htons(0);
4789 ofpc_normal_flow.tp_dst = htons(0);
4790 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4791 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4792 cls_rule = &ofproto->drop_frags_rule->up.cr;
4794 flow_wildcards_init_exact(wc);
4797 cls_rule = classifier_lookup(cls, flow, wc);
4800 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4801 rule_dpif_ref(*rule);
4802 ovs_rwlock_unlock(&cls->rwlock);
4804 return *rule != NULL;
4807 /* Given a port configuration (specified as zero if there's no port), chooses
4808 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4809 * flow table miss. */
4811 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4812 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4814 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4815 rule_dpif_ref(*rule);
4819 rule_dpif_ref(struct rule_dpif *rule)
4822 ofproto_rule_ref(&rule->up);
4827 rule_dpif_unref(struct rule_dpif *rule)
4830 ofproto_rule_unref(&rule->up);
4835 complete_operation(struct rule_dpif *rule)
4836 OVS_REQUIRES(ofproto_mutex)
4838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4840 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4841 ofoperation_complete(rule->up.pending, 0);
4844 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4846 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4849 static struct rule *
4852 struct rule_dpif *rule = xmalloc(sizeof *rule);
4857 rule_dealloc(struct rule *rule_)
4859 struct rule_dpif *rule = rule_dpif_cast(rule_);
4864 rule_construct(struct rule *rule_)
4866 struct rule_dpif *rule = rule_dpif_cast(rule_);
4867 ovs_mutex_init(&rule->stats_mutex);
4868 ovs_mutex_lock(&rule->stats_mutex);
4869 rule->packet_count = 0;
4870 rule->byte_count = 0;
4871 ovs_mutex_unlock(&rule->stats_mutex);
4876 rule_insert(struct rule *rule_)
4877 OVS_REQUIRES(ofproto_mutex)
4879 struct rule_dpif *rule = rule_dpif_cast(rule_);
4880 complete_operation(rule);
4884 rule_delete(struct rule *rule_)
4885 OVS_REQUIRES(ofproto_mutex)
4887 struct rule_dpif *rule = rule_dpif_cast(rule_);
4888 complete_operation(rule);
4892 rule_destruct(struct rule *rule_)
4894 struct rule_dpif *rule = rule_dpif_cast(rule_);
4895 ovs_mutex_destroy(&rule->stats_mutex);
4899 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4902 struct rule_dpif *rule = rule_dpif_cast(rule_);
4904 /* push_all_stats() can handle flow misses which, when using the learn
4905 * action, can cause rules to be added and deleted. This can corrupt our
4906 * caller's datastructures which assume that rule_get_stats() doesn't have
4907 * an impact on the flow table. To be safe, we disable miss handling. */
4909 push_all_stats__(false);
4912 /* Start from historical data for 'rule' itself that are no longer tracked
4913 * in facets. This counts, for example, facets that have expired. */
4914 ovs_mutex_lock(&rule->stats_mutex);
4915 *packets = rule->packet_count;
4916 *bytes = rule->byte_count;
4917 ovs_mutex_unlock(&rule->stats_mutex);
4921 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4922 struct ofpbuf *packet)
4924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4925 struct dpif_flow_stats stats;
4926 struct xlate_out xout;
4927 struct xlate_in xin;
4929 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4930 rule_dpif_credit_stats(rule, &stats);
4932 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4933 xin.resubmit_stats = &stats;
4934 xlate_actions(&xin, &xout);
4936 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4937 xout.odp_actions.size, packet);
4939 xlate_out_uninit(&xout);
4943 rule_execute(struct rule *rule, const struct flow *flow,
4944 struct ofpbuf *packet)
4946 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4947 ofpbuf_delete(packet);
4952 rule_modify_actions(struct rule *rule_, bool reset_counters)
4953 OVS_REQUIRES(ofproto_mutex)
4955 struct rule_dpif *rule = rule_dpif_cast(rule_);
4957 if (reset_counters) {
4958 ovs_mutex_lock(&rule->stats_mutex);
4959 rule->packet_count = 0;
4960 rule->byte_count = 0;
4961 ovs_mutex_unlock(&rule->stats_mutex);
4964 complete_operation(rule);
4967 /* Sends 'packet' out 'ofport'.
4968 * May modify 'packet'.
4969 * Returns 0 if successful, otherwise a positive errno value. */
4971 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4974 uint64_t odp_actions_stub[1024 / 8];
4975 struct ofpbuf key, odp_actions;
4976 struct dpif_flow_stats stats;
4977 struct odputil_keybuf keybuf;
4978 struct ofpact_output output;
4979 struct xlate_out xout;
4980 struct xlate_in xin;
4982 union flow_in_port in_port_;
4985 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4986 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4988 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4989 in_port_.ofp_port = OFPP_NONE;
4990 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4991 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4993 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4995 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4996 output.port = ofport->up.ofp_port;
4999 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5000 xin.ofpacts_len = sizeof output;
5001 xin.ofpacts = &output.ofpact;
5002 xin.resubmit_stats = &stats;
5003 xlate_actions(&xin, &xout);
5005 error = dpif_execute(ofproto->backer->dpif,
5007 xout.odp_actions.data, xout.odp_actions.size,
5009 xlate_out_uninit(&xout);
5012 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5013 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5014 ovs_strerror(error));
5017 ofproto->stats.tx_packets++;
5018 ofproto->stats.tx_bytes += packet->size;
5022 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5023 * The action will state 'slow' as the reason that the action is in the slow
5024 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5025 * dump-flows" output to see why a flow is in the slow path.)
5027 * The 'stub_size' bytes in 'stub' will be used to store the action.
5028 * 'stub_size' must be large enough for the action.
5030 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5033 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5034 enum slow_path_reason slow,
5035 uint64_t *stub, size_t stub_size,
5036 const struct nlattr **actionsp, size_t *actions_lenp)
5038 union user_action_cookie cookie;
5041 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5042 cookie.slow_path.unused = 0;
5043 cookie.slow_path.reason = slow;
5045 ofpbuf_use_stack(&buf, stub, stub_size);
5046 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5047 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5049 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5051 odp_port_t odp_port;
5054 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5055 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5056 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5058 *actionsp = buf.data;
5059 *actions_lenp = buf.size;
5063 set_frag_handling(struct ofproto *ofproto_,
5064 enum ofp_config_flags frag_handling)
5066 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5067 if (frag_handling != OFPC_FRAG_REASM) {
5068 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5076 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5077 const struct flow *flow,
5078 const struct ofpact *ofpacts, size_t ofpacts_len)
5080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5081 struct odputil_keybuf keybuf;
5082 struct dpif_flow_stats stats;
5083 struct xlate_out xout;
5084 struct xlate_in xin;
5088 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5089 odp_flow_key_from_flow(&key, flow,
5090 ofp_port_to_odp_port(ofproto,
5091 flow->in_port.ofp_port));
5093 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5095 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5096 xin.resubmit_stats = &stats;
5097 xin.ofpacts_len = ofpacts_len;
5098 xin.ofpacts = ofpacts;
5100 xlate_actions(&xin, &xout);
5101 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5102 xout.odp_actions.data, xout.odp_actions.size, packet);
5103 xlate_out_uninit(&xout);
5111 set_netflow(struct ofproto *ofproto_,
5112 const struct netflow_options *netflow_options)
5114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5116 if (netflow_options) {
5117 if (!ofproto->netflow) {
5118 ofproto->netflow = netflow_create();
5119 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5121 return netflow_set_options(ofproto->netflow, netflow_options);
5122 } else if (ofproto->netflow) {
5123 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5124 netflow_destroy(ofproto->netflow);
5125 ofproto->netflow = NULL;
5132 get_netflow_ids(const struct ofproto *ofproto_,
5133 uint8_t *engine_type, uint8_t *engine_id)
5135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5137 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5141 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5143 if (!facet_is_controller_flow(facet) &&
5144 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5145 struct subfacet *subfacet;
5146 struct ofexpired expired;
5148 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5149 if (subfacet->path == SF_FAST_PATH) {
5150 struct dpif_flow_stats stats;
5152 subfacet_install(subfacet, &facet->xout.odp_actions,
5154 subfacet_update_stats(subfacet, &stats);
5158 expired.flow = facet->flow;
5159 expired.packet_count = facet->packet_count;
5160 expired.byte_count = facet->byte_count;
5161 expired.used = facet->used;
5162 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5167 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5169 struct cls_cursor cursor;
5170 struct facet *facet;
5172 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5173 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5174 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5175 send_active_timeout(ofproto, facet);
5177 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5180 static struct ofproto_dpif *
5181 ofproto_dpif_lookup(const char *name)
5183 struct ofproto_dpif *ofproto;
5185 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5186 hash_string(name, 0), &all_ofproto_dpifs) {
5187 if (!strcmp(ofproto->up.name, name)) {
5195 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5196 const char *argv[], void *aux OVS_UNUSED)
5198 struct ofproto_dpif *ofproto;
5201 ofproto = ofproto_dpif_lookup(argv[1]);
5203 unixctl_command_reply_error(conn, "no such bridge");
5206 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5207 mac_learning_flush(ofproto->ml);
5208 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5210 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5211 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5212 mac_learning_flush(ofproto->ml);
5213 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5217 unixctl_command_reply(conn, "table successfully flushed");
5220 static struct ofport_dpif *
5221 ofbundle_get_a_port(const struct ofbundle *bundle)
5223 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5228 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5229 const char *argv[], void *aux OVS_UNUSED)
5231 struct ds ds = DS_EMPTY_INITIALIZER;
5232 const struct ofproto_dpif *ofproto;
5233 const struct mac_entry *e;
5235 ofproto = ofproto_dpif_lookup(argv[1]);
5237 unixctl_command_reply_error(conn, "no such bridge");
5241 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5242 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5243 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5244 struct ofbundle *bundle = e->port.p;
5245 char name[OFP_MAX_PORT_NAME_LEN];
5247 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5249 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5250 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5251 mac_entry_age(ofproto->ml, e));
5253 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5254 unixctl_command_reply(conn, ds_cstr(&ds));
5259 struct xlate_out xout;
5260 struct xlate_in xin;
5266 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5268 struct rule_actions *actions;
5271 ds_put_char_multiple(result, '\t', level);
5273 ds_put_cstr(result, "No match\n");
5277 ovs_mutex_lock(&rule->up.mutex);
5278 cookie = rule->up.flow_cookie;
5279 ovs_mutex_unlock(&rule->up.mutex);
5281 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5282 rule ? rule->up.table_id : 0, ntohll(cookie));
5283 cls_rule_format(&rule->up.cr, result);
5284 ds_put_char(result, '\n');
5286 actions = rule_dpif_get_actions(rule);
5288 ds_put_char_multiple(result, '\t', level);
5289 ds_put_cstr(result, "OpenFlow ");
5290 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5291 ds_put_char(result, '\n');
5293 rule_actions_unref(actions);
5297 trace_format_flow(struct ds *result, int level, const char *title,
5298 struct trace_ctx *trace)
5300 ds_put_char_multiple(result, '\t', level);
5301 ds_put_format(result, "%s: ", title);
5302 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5303 ds_put_cstr(result, "unchanged");
5305 flow_format(result, &trace->xin.flow);
5306 trace->flow = trace->xin.flow;
5308 ds_put_char(result, '\n');
5312 trace_format_regs(struct ds *result, int level, const char *title,
5313 struct trace_ctx *trace)
5317 ds_put_char_multiple(result, '\t', level);
5318 ds_put_format(result, "%s:", title);
5319 for (i = 0; i < FLOW_N_REGS; i++) {
5320 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5322 ds_put_char(result, '\n');
5326 trace_format_odp(struct ds *result, int level, const char *title,
5327 struct trace_ctx *trace)
5329 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5331 ds_put_char_multiple(result, '\t', level);
5332 ds_put_format(result, "%s: ", title);
5333 format_odp_actions(result, odp_actions->data, odp_actions->size);
5334 ds_put_char(result, '\n');
5338 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5340 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5341 struct ds *result = trace->result;
5343 ds_put_char(result, '\n');
5344 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5345 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5346 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5347 trace_format_rule(result, recurse + 1, rule);
5351 trace_report(struct xlate_in *xin, const char *s, int recurse)
5353 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5354 struct ds *result = trace->result;
5356 ds_put_char_multiple(result, '\t', recurse);
5357 ds_put_cstr(result, s);
5358 ds_put_char(result, '\n');
5362 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5363 void *aux OVS_UNUSED)
5365 const struct dpif_backer *backer;
5366 struct ofproto_dpif *ofproto;
5367 struct ofpbuf odp_key, odp_mask;
5368 struct ofpbuf *packet;
5376 ofpbuf_init(&odp_key, 0);
5377 ofpbuf_init(&odp_mask, 0);
5379 /* Handle "-generate" or a hex string as the last argument. */
5380 if (!strcmp(argv[argc - 1], "-generate")) {
5381 packet = ofpbuf_new(0);
5384 const char *error = eth_from_hex(argv[argc - 1], &packet);
5387 } else if (argc == 4) {
5388 /* The 3-argument form must end in "-generate' or a hex string. */
5389 unixctl_command_reply_error(conn, error);
5394 /* Parse the flow and determine whether a datapath or
5395 * bridge is specified. If function odp_flow_key_from_string()
5396 * returns 0, the flow is a odp_flow. If function
5397 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5398 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5399 /* If the odp_flow is the second argument,
5400 * the datapath name is the first argument. */
5402 const char *dp_type;
5403 if (!strncmp(argv[1], "ovs-", 4)) {
5404 dp_type = argv[1] + 4;
5408 backer = shash_find_data(&all_dpif_backers, dp_type);
5410 unixctl_command_reply_error(conn, "Cannot find datapath "
5415 /* No datapath name specified, so there should be only one
5417 struct shash_node *node;
5418 if (shash_count(&all_dpif_backers) != 1) {
5419 unixctl_command_reply_error(conn, "Must specify datapath "
5420 "name, there is more than one type of datapath");
5423 node = shash_first(&all_dpif_backers);
5424 backer = node->data;
5427 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5428 NULL, &ofproto, NULL)) {
5429 unixctl_command_reply_error(conn, "Invalid datapath flow");
5432 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5433 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5435 unixctl_command_reply_error(conn, "Must specify bridge name");
5439 ofproto = ofproto_dpif_lookup(argv[1]);
5441 unixctl_command_reply_error(conn, "Unknown bridge name");
5445 unixctl_command_reply_error(conn, "Bad flow syntax");
5449 /* Generate a packet, if requested. */
5451 if (!packet->size) {
5452 flow_compose(packet, &flow);
5454 union flow_in_port in_port_;
5456 in_port_ = flow.in_port;
5457 ds_put_cstr(&result, "Packet: ");
5458 s = ofp_packet_to_string(packet->data, packet->size);
5459 ds_put_cstr(&result, s);
5462 /* Use the metadata from the flow and the packet argument
5463 * to reconstruct the flow. */
5464 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5469 ofproto_trace(ofproto, &flow, packet, &result);
5470 unixctl_command_reply(conn, ds_cstr(&result));
5473 ds_destroy(&result);
5474 ofpbuf_delete(packet);
5475 ofpbuf_uninit(&odp_key);
5476 ofpbuf_uninit(&odp_mask);
5480 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5481 const struct ofpbuf *packet, struct ds *ds)
5483 struct rule_dpif *rule;
5484 struct flow_wildcards wc;
5486 ds_put_cstr(ds, "Flow: ");
5487 flow_format(ds, flow);
5488 ds_put_char(ds, '\n');
5490 flow_wildcards_init_catchall(&wc);
5491 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5493 trace_format_rule(ds, 0, rule);
5494 if (rule == ofproto->miss_rule) {
5495 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5496 } else if (rule == ofproto->no_packet_in_rule) {
5497 ds_put_cstr(ds, "\nNo match, packets dropped because "
5498 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5499 } else if (rule == ofproto->drop_frags_rule) {
5500 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5501 "and the fragment handling mode is \"drop\".\n");
5505 uint64_t odp_actions_stub[1024 / 8];
5506 struct ofpbuf odp_actions;
5507 struct trace_ctx trace;
5511 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5514 ofpbuf_use_stub(&odp_actions,
5515 odp_actions_stub, sizeof odp_actions_stub);
5516 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5517 trace.xin.resubmit_hook = trace_resubmit;
5518 trace.xin.report_hook = trace_report;
5520 xlate_actions(&trace.xin, &trace.xout);
5521 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5523 ds_put_char(ds, '\n');
5524 trace_format_flow(ds, 0, "Final flow", &trace);
5526 match_init(&match, flow, &trace.xout.wc);
5527 ds_put_cstr(ds, "Relevant fields: ");
5528 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5529 ds_put_char(ds, '\n');
5531 ds_put_cstr(ds, "Datapath actions: ");
5532 format_odp_actions(ds, trace.xout.odp_actions.data,
5533 trace.xout.odp_actions.size);
5535 if (trace.xout.slow) {
5536 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5537 "slow path because it:");
5538 switch (trace.xout.slow) {
5540 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5543 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5546 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5549 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5551 case SLOW_CONTROLLER:
5552 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5553 "to the OpenFlow controller.");
5560 xlate_out_uninit(&trace.xout);
5563 rule_dpif_unref(rule);
5566 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5567 * 'reply' describing the results. */
5569 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5571 struct cls_cursor cursor;
5572 struct facet *facet;
5576 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5577 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5578 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5579 if (!facet_check_consistency(facet)) {
5583 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5585 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5589 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5590 ofproto->up.name, errors);
5592 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5597 ofproto_dpif_self_check(struct unixctl_conn *conn,
5598 int argc, const char *argv[], void *aux OVS_UNUSED)
5600 struct ds reply = DS_EMPTY_INITIALIZER;
5601 struct ofproto_dpif *ofproto;
5604 ofproto = ofproto_dpif_lookup(argv[1]);
5606 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5607 "ofproto/list for help)");
5610 ofproto_dpif_self_check__(ofproto, &reply);
5612 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5613 ofproto_dpif_self_check__(ofproto, &reply);
5617 unixctl_command_reply(conn, ds_cstr(&reply));
5621 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5622 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5623 * to destroy 'ofproto_shash' and free the returned value. */
5624 static const struct shash_node **
5625 get_ofprotos(struct shash *ofproto_shash)
5627 const struct ofproto_dpif *ofproto;
5629 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5630 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5631 shash_add_nocopy(ofproto_shash, name, ofproto);
5634 return shash_sort(ofproto_shash);
5638 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5639 const char *argv[] OVS_UNUSED,
5640 void *aux OVS_UNUSED)
5642 struct ds ds = DS_EMPTY_INITIALIZER;
5643 struct shash ofproto_shash;
5644 const struct shash_node **sorted_ofprotos;
5647 shash_init(&ofproto_shash);
5648 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5649 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5650 const struct shash_node *node = sorted_ofprotos[i];
5651 ds_put_format(&ds, "%s\n", node->name);
5654 shash_destroy(&ofproto_shash);
5655 free(sorted_ofprotos);
5657 unixctl_command_reply(conn, ds_cstr(&ds));
5662 show_dp_rates(struct ds *ds, const char *heading,
5663 const struct avg_subfacet_rates *rates)
5665 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5666 heading, rates->add_rate, rates->del_rate);
5670 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5672 const struct shash_node **ofprotos;
5673 struct ofproto_dpif *ofproto;
5674 struct shash ofproto_shash;
5675 uint64_t n_hit, n_missed;
5676 long long int minutes;
5679 n_hit = n_missed = 0;
5680 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5681 if (ofproto->backer == backer) {
5682 n_missed += ofproto->n_missed;
5683 n_hit += ofproto->n_hit;
5687 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5688 dpif_name(backer->dpif), n_hit, n_missed);
5689 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5690 " life span: %lldms\n", hmap_count(&backer->subfacets),
5691 backer->avg_n_subfacet, backer->max_n_subfacet,
5692 backer->avg_subfacet_life);
5694 minutes = (time_msec() - backer->created) / (1000 * 60);
5695 if (minutes >= 60) {
5696 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5698 if (minutes >= 60 * 24) {
5699 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5701 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5703 shash_init(&ofproto_shash);
5704 ofprotos = get_ofprotos(&ofproto_shash);
5705 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5706 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5707 const struct shash_node **ports;
5710 if (ofproto->backer != backer) {
5714 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5715 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5717 ports = shash_sort(&ofproto->up.port_by_name);
5718 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5719 const struct shash_node *node = ports[j];
5720 struct ofport *ofport = node->data;
5722 odp_port_t odp_port;
5724 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5727 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5728 if (odp_port != ODPP_NONE) {
5729 ds_put_format(ds, "%"PRIu32":", odp_port);
5731 ds_put_cstr(ds, "none:");
5734 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5737 if (!netdev_get_config(ofport->netdev, &config)) {
5738 const struct smap_node **nodes;
5741 nodes = smap_sort(&config);
5742 for (i = 0; i < smap_count(&config); i++) {
5743 const struct smap_node *node = nodes[i];
5744 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5745 node->key, node->value);
5749 smap_destroy(&config);
5751 ds_put_char(ds, ')');
5752 ds_put_char(ds, '\n');
5756 shash_destroy(&ofproto_shash);
5761 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5762 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5764 struct ds ds = DS_EMPTY_INITIALIZER;
5765 const struct shash_node **backers;
5768 backers = shash_sort(&all_dpif_backers);
5769 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5770 dpif_show_backer(backers[i]->data, &ds);
5774 unixctl_command_reply(conn, ds_cstr(&ds));
5778 /* Dump the megaflow (facet) cache. This is useful to check the
5779 * correctness of flow wildcarding, since the same mechanism is used for
5780 * both xlate caching and kernel wildcarding.
5782 * It's important to note that in the output the flow description uses
5783 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5785 * This command is only needed for advanced debugging, so it's not
5786 * documented in the man page. */
5788 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5789 int argc OVS_UNUSED, const char *argv[],
5790 void *aux OVS_UNUSED)
5792 struct ds ds = DS_EMPTY_INITIALIZER;
5793 const struct ofproto_dpif *ofproto;
5794 long long int now = time_msec();
5795 struct cls_cursor cursor;
5796 struct facet *facet;
5798 ofproto = ofproto_dpif_lookup(argv[1]);
5800 unixctl_command_reply_error(conn, "no such bridge");
5804 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5805 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5806 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5807 cls_rule_format(&facet->cr, &ds);
5808 ds_put_cstr(&ds, ", ");
5809 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5810 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5811 ds_put_cstr(&ds, "Datapath actions: ");
5812 if (facet->xout.slow) {
5813 uint64_t slow_path_stub[128 / 8];
5814 const struct nlattr *actions;
5817 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5818 slow_path_stub, sizeof slow_path_stub,
5819 &actions, &actions_len);
5820 format_odp_actions(&ds, actions, actions_len);
5822 format_odp_actions(&ds, facet->xout.odp_actions.data,
5823 facet->xout.odp_actions.size);
5825 ds_put_cstr(&ds, "\n");
5827 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5829 ds_chomp(&ds, '\n');
5830 unixctl_command_reply(conn, ds_cstr(&ds));
5834 /* Disable using the megaflows.
5836 * This command is only needed for advanced debugging, so it's not
5837 * documented in the man page. */
5839 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5840 int argc OVS_UNUSED,
5841 const char *argv[] OVS_UNUSED,
5842 void *aux OVS_UNUSED)
5844 struct ofproto_dpif *ofproto;
5846 enable_megaflows = false;
5848 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5849 flush(&ofproto->up);
5852 unixctl_command_reply(conn, "megaflows disabled");
5855 /* Re-enable using megaflows.
5857 * This command is only needed for advanced debugging, so it's not
5858 * documented in the man page. */
5860 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5861 int argc OVS_UNUSED,
5862 const char *argv[] OVS_UNUSED,
5863 void *aux OVS_UNUSED)
5865 struct ofproto_dpif *ofproto;
5867 enable_megaflows = true;
5869 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5870 flush(&ofproto->up);
5873 unixctl_command_reply(conn, "megaflows enabled");
5877 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5878 int argc OVS_UNUSED, const char *argv[],
5879 void *aux OVS_UNUSED)
5881 struct ds ds = DS_EMPTY_INITIALIZER;
5882 const struct ofproto_dpif *ofproto;
5883 struct subfacet *subfacet;
5885 ofproto = ofproto_dpif_lookup(argv[1]);
5887 unixctl_command_reply_error(conn, "no such bridge");
5891 update_stats(ofproto->backer);
5893 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5894 struct facet *facet = subfacet->facet;
5895 struct odputil_keybuf maskbuf;
5898 if (facet->ofproto != ofproto) {
5902 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5903 if (enable_megaflows) {
5904 mask_to_netlink_attr(&mask, facet->ofproto,
5905 &facet->xout.wc.masks, &facet->flow);
5908 odp_flow_format(subfacet->key, subfacet->key_len,
5909 mask.data, mask.size, &ds, false);
5911 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5912 subfacet->dp_packet_count, subfacet->dp_byte_count);
5913 if (subfacet->used) {
5914 ds_put_format(&ds, "%.3fs",
5915 (time_msec() - subfacet->used) / 1000.0);
5917 ds_put_format(&ds, "never");
5919 if (subfacet->facet->tcp_flags) {
5920 ds_put_cstr(&ds, ", flags:");
5921 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5924 ds_put_cstr(&ds, ", actions:");
5925 if (facet->xout.slow) {
5926 uint64_t slow_path_stub[128 / 8];
5927 const struct nlattr *actions;
5930 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5931 slow_path_stub, sizeof slow_path_stub,
5932 &actions, &actions_len);
5933 format_odp_actions(&ds, actions, actions_len);
5935 format_odp_actions(&ds, facet->xout.odp_actions.data,
5936 facet->xout.odp_actions.size);
5938 ds_put_char(&ds, '\n');
5941 unixctl_command_reply(conn, ds_cstr(&ds));
5946 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5947 int argc OVS_UNUSED, const char *argv[],
5948 void *aux OVS_UNUSED)
5950 struct ds ds = DS_EMPTY_INITIALIZER;
5951 struct ofproto_dpif *ofproto;
5953 ofproto = ofproto_dpif_lookup(argv[1]);
5955 unixctl_command_reply_error(conn, "no such bridge");
5959 flush(&ofproto->up);
5961 unixctl_command_reply(conn, ds_cstr(&ds));
5966 ofproto_dpif_unixctl_init(void)
5968 static bool registered;
5974 unixctl_command_register(
5976 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5977 1, 3, ofproto_unixctl_trace, NULL);
5978 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5979 ofproto_unixctl_fdb_flush, NULL);
5980 unixctl_command_register("fdb/show", "bridge", 1, 1,
5981 ofproto_unixctl_fdb_show, NULL);
5982 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5983 ofproto_dpif_self_check, NULL);
5984 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5985 ofproto_unixctl_dpif_dump_dps, NULL);
5986 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5988 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5989 ofproto_unixctl_dpif_dump_flows, NULL);
5990 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5991 ofproto_unixctl_dpif_del_flows, NULL);
5992 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5993 ofproto_unixctl_dpif_dump_megaflows, NULL);
5994 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5995 ofproto_unixctl_dpif_disable_megaflows, NULL);
5996 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5997 ofproto_unixctl_dpif_enable_megaflows, NULL);
6000 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6002 * This is deprecated. It is only for compatibility with broken device drivers
6003 * in old versions of Linux that do not properly support VLANs when VLAN
6004 * devices are not used. When broken device drivers are no longer in
6005 * widespread use, we will delete these interfaces. */
6008 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6011 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6013 if (realdev_ofp_port == ofport->realdev_ofp_port
6014 && vid == ofport->vlandev_vid) {
6018 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6020 if (ofport->realdev_ofp_port) {
6023 if (realdev_ofp_port && ofport->bundle) {
6024 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6025 * themselves be part of a bundle. */
6026 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6029 ofport->realdev_ofp_port = realdev_ofp_port;
6030 ofport->vlandev_vid = vid;
6032 if (realdev_ofp_port) {
6033 vsp_add(ofport, realdev_ofp_port, vid);
6040 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6042 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6046 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6047 OVS_EXCLUDED(ofproto->vsp_mutex)
6051 ovs_mutex_lock(&ofproto->vsp_mutex);
6052 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6053 ovs_mutex_unlock(&ofproto->vsp_mutex);
6058 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6059 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6060 OVS_REQUIRES(ofproto->vsp_mutex)
6062 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6063 int vid = vlan_tci_to_vid(vlan_tci);
6064 const struct vlan_splinter *vsp;
6066 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6067 hash_realdev_vid(realdev_ofp_port, vid),
6068 &ofproto->realdev_vid_map) {
6069 if (vsp->realdev_ofp_port == realdev_ofp_port
6070 && vsp->vid == vid) {
6071 return vsp->vlandev_ofp_port;
6075 return realdev_ofp_port;
6078 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6079 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6080 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6081 * 'vlan_tci' 9, it would return the port number of eth0.9.
6083 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6084 * function just returns its 'realdev_ofp_port' argument. */
6086 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6087 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6088 OVS_EXCLUDED(ofproto->vsp_mutex)
6092 ovs_mutex_lock(&ofproto->vsp_mutex);
6093 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6094 ovs_mutex_unlock(&ofproto->vsp_mutex);
6098 static struct vlan_splinter *
6099 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6101 struct vlan_splinter *vsp;
6103 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6104 hash_ofp_port(vlandev_ofp_port),
6105 &ofproto->vlandev_map) {
6106 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6114 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6115 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6116 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6117 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6118 * eth0 and store 9 in '*vid'.
6120 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6121 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6124 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6125 ofp_port_t vlandev_ofp_port, int *vid)
6126 OVS_REQUIRES(ofproto->vsp_mutex)
6128 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6129 const struct vlan_splinter *vsp;
6131 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6136 return vsp->realdev_ofp_port;
6142 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6143 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6144 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6145 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6146 * always the case unless VLAN splinters are enabled), returns false without
6147 * making any changes. */
6149 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6150 OVS_EXCLUDED(ofproto->vsp_mutex)
6155 ovs_mutex_lock(&ofproto->vsp_mutex);
6156 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6157 ovs_mutex_unlock(&ofproto->vsp_mutex);
6162 /* Cause the flow to be processed as if it came in on the real device with
6163 * the VLAN device's VLAN ID. */
6164 flow->in_port.ofp_port = realdev;
6165 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6170 vsp_remove(struct ofport_dpif *port)
6172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6173 struct vlan_splinter *vsp;
6175 ovs_mutex_lock(&ofproto->vsp_mutex);
6176 vsp = vlandev_find(ofproto, port->up.ofp_port);
6178 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6179 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6182 port->realdev_ofp_port = 0;
6184 VLOG_ERR("missing vlan device record");
6186 ovs_mutex_unlock(&ofproto->vsp_mutex);
6190 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6194 ovs_mutex_lock(&ofproto->vsp_mutex);
6195 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6196 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6197 == realdev_ofp_port)) {
6198 struct vlan_splinter *vsp;
6200 vsp = xmalloc(sizeof *vsp);
6201 vsp->realdev_ofp_port = realdev_ofp_port;
6202 vsp->vlandev_ofp_port = port->up.ofp_port;
6205 port->realdev_ofp_port = realdev_ofp_port;
6207 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6208 hash_ofp_port(port->up.ofp_port));
6209 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6210 hash_realdev_vid(realdev_ofp_port, vid));
6212 VLOG_ERR("duplicate vlan device record");
6214 ovs_mutex_unlock(&ofproto->vsp_mutex);
6218 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6220 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6221 return ofport ? ofport->odp_port : ODPP_NONE;
6224 struct ofport_dpif *
6225 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6227 struct ofport_dpif *port;
6229 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6230 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6231 &backer->odp_to_ofport_map) {
6232 if (port->odp_port == odp_port) {
6233 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6238 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6243 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6245 struct ofport_dpif *port;
6247 port = odp_port_to_ofport(ofproto->backer, odp_port);
6248 if (port && &ofproto->up == port->up.ofproto) {
6249 return port->up.ofp_port;
6255 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6256 * most heavily weighted element. 'base' designates the rate of decay: after
6257 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6260 exp_mavg(double *avg, int base, double new)
6262 *avg = (*avg * (base - 1) + new) / base;
6266 update_moving_averages(struct dpif_backer *backer)
6268 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6269 long long int minutes = (time_msec() - backer->created) / min_ms;
6272 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6274 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6277 backer->lifetime.add_rate = 0.0;
6278 backer->lifetime.del_rate = 0.0;
6281 /* Update hourly averages on the minute boundaries. */
6282 if (time_msec() - backer->last_minute >= min_ms) {
6283 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6284 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6286 /* Update daily averages on the hour boundaries. */
6287 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6288 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6289 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6292 backer->total_subfacet_add_count += backer->subfacet_add_count;
6293 backer->total_subfacet_del_count += backer->subfacet_del_count;
6294 backer->subfacet_add_count = 0;
6295 backer->subfacet_del_count = 0;
6296 backer->last_minute += min_ms;
6300 /* Appends a representation of 'mask' as OVS_KEY_ATTR_* attributes to 'buf'.
6301 * Ommit vlan mask if the flow's vlan was added by VLAN splinter. */
6303 mask_to_netlink_attr(struct ofpbuf *buf, const struct ofproto_dpif *ofproto,
6304 const struct flow *mask, const struct flow *flow)
6306 ofp_port_t vlandev_port;
6307 struct flow orig_flow;
6309 vlandev_port = vsp_realdev_to_vlandev(ofproto, flow->in_port.ofp_port,
6312 if (vlandev_port != flow->in_port.ofp_port) {
6313 /* This flow was received from the VLAN port, do not
6314 * generate vlan masks. */
6316 orig_flow.vlan_tci = 0;
6320 odp_flow_key_from_mask(buf, mask, flow, UINT32_MAX);
6323 const struct ofproto_class ofproto_dpif_class = {
6358 port_is_lacp_current,
6359 NULL, /* rule_choose_table */
6368 rule_modify_actions,
6382 get_stp_port_status,
6389 is_mirror_output_bundle,
6390 forward_bpdu_changed,
6391 set_mac_table_config,
6393 NULL, /* meter_get_features */
6394 NULL, /* meter_set */
6395 NULL, /* meter_get */
6396 NULL, /* meter_del */