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-monitor.h"
56 #include "ofproto-dpif-sflow.h"
57 #include "ofproto-dpif-upcall.h"
58 #include "ofproto-dpif-xlate.h"
59 #include "poll-loop.h"
64 #include "unaligned.h"
66 #include "vlan-bitmap.h"
69 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
71 COVERAGE_DEFINE(ofproto_dpif_expired);
72 COVERAGE_DEFINE(facet_revalidate);
73 COVERAGE_DEFINE(facet_unexpected);
74 COVERAGE_DEFINE(facet_create);
75 COVERAGE_DEFINE(facet_remove);
76 COVERAGE_DEFINE(subfacet_create);
77 COVERAGE_DEFINE(subfacet_destroy);
78 COVERAGE_DEFINE(subfacet_install_fail);
79 COVERAGE_DEFINE(packet_in_overflow);
81 /* Number of implemented OpenFlow tables. */
82 enum { N_TABLES = 255 };
83 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
84 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
94 * - Do include packets and bytes from facets that have been deleted or
95 * whose own statistics have been folded into the rule.
97 * - Do include packets and bytes sent "by hand" that were accounted to
98 * the rule without any facet being involved (this is a rare corner
99 * case in rule_execute()).
101 * - Do not include packet or bytes that can be obtained from any facet's
102 * packet_count or byte_count member or that can be obtained from the
103 * datapath by, e.g., dpif_flow_get() for any subfacet.
105 struct ovs_mutex stats_mutex;
106 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
107 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
110 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
111 static struct rule_dpif *rule_dpif_cast(const struct rule *);
114 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
115 struct ofproto_dpif *ofproto; /* Owning ofproto. */
116 void *aux; /* Key supplied by ofproto's client. */
117 char *name; /* Identifier for log messages. */
120 struct list ports; /* Contains "struct ofport"s. */
121 enum port_vlan_mode vlan_mode; /* VLAN mode */
122 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
123 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
124 * NULL if all VLANs are trunked. */
125 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
126 struct bond *bond; /* Nonnull iff more than one port. */
127 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
130 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
133 static void bundle_remove(struct ofport *);
134 static void bundle_update(struct ofbundle *);
135 static void bundle_destroy(struct ofbundle *);
136 static void bundle_del_port(struct ofport_dpif *);
137 static void bundle_run(struct ofbundle *);
138 static void bundle_wait(struct ofbundle *);
140 static void stp_run(struct ofproto_dpif *ofproto);
141 static void stp_wait(struct ofproto_dpif *ofproto);
142 static int set_stp_port(struct ofport *,
143 const struct ofproto_port_stp_settings *);
145 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
146 enum slow_path_reason,
147 uint64_t *stub, size_t stub_size,
148 const struct nlattr **actionsp,
149 size_t *actions_lenp);
151 /* A subfacet (see "struct subfacet" below) has three possible installation
154 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
155 * case just after the subfacet is created, just before the subfacet is
156 * destroyed, or if the datapath returns an error when we try to install a
159 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
161 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
162 * ofproto_dpif is installed in the datapath.
165 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
166 SF_FAST_PATH, /* Full actions are installed. */
167 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
170 /* A dpif flow and actions associated with a facet.
172 * See also the large comment on struct facet. */
175 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
176 struct list list_node; /* In struct facet's 'facets' list. */
177 struct facet *facet; /* Owning facet. */
178 struct dpif_backer *backer; /* Owning backer. */
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 ofproto_dpif *ofproto;
236 struct list subfacets;
237 long long int used; /* Time last used; time created if not used. */
240 struct flow flow; /* Flow of the creating subfacet. */
241 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
245 * - Do include packets and bytes sent "by hand", e.g. with
248 * - Do include packets and bytes that were obtained from the datapath
249 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
250 * DPIF_FP_ZERO_STATS).
252 * - Do not include packets or bytes that can be obtained from the
253 * datapath for any existing subfacet.
255 uint64_t packet_count; /* Number of packets received. */
256 uint64_t byte_count; /* Number of bytes received. */
258 /* Resubmit statistics. */
259 uint64_t prev_packet_count; /* Number of packets from last stats push. */
260 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
261 long long int prev_used; /* Used time from last stats push. */
264 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
265 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
266 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
268 struct xlate_out xout;
270 /* Storage for a single subfacet, to reduce malloc() time and space
271 * overhead. (A facet always has at least one subfacet and in the common
272 * case has exactly one subfacet. However, 'one_subfacet' may not
273 * always be valid, since it could have been removed after newer
274 * subfacets were pushed onto the 'subfacets' list.) */
275 struct subfacet one_subfacet;
277 long long int learn_rl; /* Rate limiter for facet_learn(). */
280 static struct facet *facet_create(const struct flow_miss *);
281 static void facet_remove(struct facet *);
282 static void facet_free(struct facet *);
284 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
285 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
286 const struct flow *);
287 static bool facet_revalidate(struct facet *);
288 static bool facet_check_consistency(struct facet *);
290 static void facet_flush_stats(struct facet *);
292 static void facet_reset_counters(struct facet *);
293 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
294 struct dpif_flow_stats *, bool may_learn);
295 static void facet_push_stats(struct facet *, bool may_learn);
296 static void facet_learn(struct facet *);
297 static void facet_account(struct facet *);
298 static void push_all_stats(void);
300 static bool facet_is_controller_flow(struct facet *);
303 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
307 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
308 struct list bundle_node; /* In struct ofbundle's "ports" list. */
309 struct cfm *cfm; /* Connectivity Fault Management, if any. */
310 struct bfd *bfd; /* BFD, if any. */
311 bool may_enable; /* May be enabled in bonds. */
312 bool is_tunnel; /* This port is a tunnel. */
313 long long int carrier_seq; /* Carrier status changes. */
314 struct ofport_dpif *peer; /* Peer if patch port. */
317 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
318 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
319 long long int stp_state_entered;
321 /* Queue to DSCP mapping. */
322 struct ofproto_port_queue *qdscp;
325 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
327 * This is deprecated. It is only for compatibility with broken device
328 * drivers in old versions of Linux that do not properly support VLANs when
329 * VLAN devices are not used. When broken device drivers are no longer in
330 * widespread use, we will delete these interfaces. */
331 ofp_port_t realdev_ofp_port;
335 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
337 * This is deprecated. It is only for compatibility with broken device drivers
338 * in old versions of Linux that do not properly support VLANs when VLAN
339 * devices are not used. When broken device drivers are no longer in
340 * widespread use, we will delete these interfaces. */
341 struct vlan_splinter {
342 struct hmap_node realdev_vid_node;
343 struct hmap_node vlandev_node;
344 ofp_port_t realdev_ofp_port;
345 ofp_port_t vlandev_ofp_port;
349 static void vsp_remove(struct ofport_dpif *);
350 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
352 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
355 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
358 static struct ofport_dpif *
359 ofport_dpif_cast(const struct ofport *ofport)
361 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
364 static void port_run(struct ofport_dpif *);
365 static int set_bfd(struct ofport *, const struct smap *);
366 static int set_cfm(struct ofport *, const struct cfm_settings *);
367 static void ofport_update_peer(struct ofport_dpif *);
368 static void run_fast_rl(void);
369 static int run_fast(struct ofproto *);
371 struct dpif_completion {
372 struct list list_node;
373 struct ofoperation *op;
376 /* Reasons that we might need to revalidate every facet, and corresponding
379 * A value of 0 means that there is no need to revalidate.
381 * It would be nice to have some cleaner way to integrate with coverage
382 * counters, but with only a few reasons I guess this is good enough for
384 enum revalidate_reason {
385 REV_RECONFIGURE = 1, /* Switch configuration changed. */
386 REV_STP, /* Spanning tree protocol port status change. */
387 REV_BOND, /* Bonding changed. */
388 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
389 REV_FLOW_TABLE, /* Flow table changed. */
390 REV_MAC_LEARNING, /* Mac learning changed. */
391 REV_INCONSISTENCY /* Facet self-check failed. */
393 COVERAGE_DEFINE(rev_reconfigure);
394 COVERAGE_DEFINE(rev_stp);
395 COVERAGE_DEFINE(rev_bond);
396 COVERAGE_DEFINE(rev_port_toggled);
397 COVERAGE_DEFINE(rev_flow_table);
398 COVERAGE_DEFINE(rev_mac_learning);
399 COVERAGE_DEFINE(rev_inconsistency);
401 struct avg_subfacet_rates {
402 double add_rate; /* Moving average of new flows created per minute. */
403 double del_rate; /* Moving average of flows deleted per minute. */
406 /* All datapaths of a given type share a single dpif backer instance. */
412 struct timer next_expiration;
414 struct ovs_rwlock odp_to_ofport_lock;
415 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
417 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
419 /* Facet revalidation flags applying to facets which use this backer. */
420 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
422 struct hmap drop_keys; /* Set of dropped odp keys. */
423 bool recv_set_enable; /* Enables or disables receiving packets. */
425 struct hmap subfacets;
426 struct governor *governor;
428 /* Subfacet statistics.
430 * These keep track of the total number of subfacets added and deleted and
431 * flow life span. They are useful for computing the flow rates stats
432 * exposed via "ovs-appctl dpif/show". The goal is to learn about
433 * traffic patterns in ways that we can use later to improve Open vSwitch
434 * performance in new situations. */
435 long long int created; /* Time when it is created. */
436 unsigned max_n_subfacet; /* Maximum number of flows */
437 unsigned avg_n_subfacet; /* Average number of flows. */
438 long long int avg_subfacet_life; /* Average life span of subfacets. */
440 /* Number of upcall handling threads. */
441 unsigned int n_handler_threads;
444 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
445 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
447 static void drop_key_clear(struct dpif_backer *);
449 struct ofproto_dpif {
450 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
452 struct dpif_backer *backer;
454 /* Special OpenFlow rules. */
455 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
456 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
457 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
460 struct netflow *netflow;
461 struct dpif_sflow *sflow;
462 struct dpif_ipfix *ipfix;
463 struct hmap bundles; /* Contains "struct ofbundle"s. */
464 struct mac_learning *ml;
465 bool has_bonded_bundles;
466 struct mbridge *mbridge;
469 struct classifier facets; /* Contains 'struct facet's. */
470 long long int consistency_rl;
472 struct ovs_mutex stats_mutex;
473 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
474 * consumed in userspace. */
478 long long int stp_last_tick;
480 /* VLAN splinters. */
481 struct ovs_mutex vsp_mutex;
482 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
483 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
486 struct sset ports; /* Set of standard port names. */
487 struct sset ghost_ports; /* Ports with no datapath port. */
488 struct sset port_poll_set; /* Queued names for port_poll() reply. */
489 int port_poll_errno; /* Last errno for port_poll() reply. */
491 /* Per ofproto's dpif stats. */
496 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
499 /* By default, flows in the datapath are wildcarded (megaflows). They
500 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
501 static bool enable_megaflows = true;
503 /* All existing ofproto_dpif instances, indexed by ->up.name. */
504 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
506 static void ofproto_dpif_unixctl_init(void);
508 static inline struct ofproto_dpif *
509 ofproto_dpif_cast(const struct ofproto *ofproto)
511 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
512 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
515 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
516 ofp_port_t ofp_port);
517 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
518 const struct ofpbuf *packet, struct ds *);
521 static void handle_upcalls(struct dpif_backer *);
523 /* Flow expiration. */
524 static int expire(struct dpif_backer *);
527 static void send_netflow_active_timeouts(struct ofproto_dpif *);
529 /* Global variables. */
530 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
532 /* Initial mappings of port to bridge mappings. */
533 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
535 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
538 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
539 struct ofputil_flow_mod *fm)
541 ofproto_flow_mod(&ofproto->up, fm);
544 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
545 * Takes ownership of 'pin' and pin->packet. */
547 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
548 struct ofputil_packet_in *pin)
550 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
551 COVERAGE_INC(packet_in_overflow);
552 free(CONST_CAST(void *, pin->packet));
557 /* Factory functions. */
560 init(const struct shash *iface_hints)
562 struct shash_node *node;
564 /* Make a local copy, since we don't own 'iface_hints' elements. */
565 SHASH_FOR_EACH(node, iface_hints) {
566 const struct iface_hint *orig_hint = node->data;
567 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
569 new_hint->br_name = xstrdup(orig_hint->br_name);
570 new_hint->br_type = xstrdup(orig_hint->br_type);
571 new_hint->ofp_port = orig_hint->ofp_port;
573 shash_add(&init_ofp_ports, node->name, new_hint);
578 enumerate_types(struct sset *types)
580 dp_enumerate_types(types);
584 enumerate_names(const char *type, struct sset *names)
586 struct ofproto_dpif *ofproto;
589 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
590 if (strcmp(type, ofproto->up.type)) {
593 sset_add(names, ofproto->up.name);
600 del(const char *type, const char *name)
605 error = dpif_open(name, type, &dpif);
607 error = dpif_delete(dpif);
614 port_open_type(const char *datapath_type, const char *port_type)
616 return dpif_port_open_type(datapath_type, port_type);
619 /* Type functions. */
621 static void process_dpif_port_changes(struct dpif_backer *);
622 static void process_dpif_all_ports_changed(struct dpif_backer *);
623 static void process_dpif_port_change(struct dpif_backer *,
624 const char *devname);
625 static void process_dpif_port_error(struct dpif_backer *, int error);
627 static struct ofproto_dpif *
628 lookup_ofproto_dpif_by_port_name(const char *name)
630 struct ofproto_dpif *ofproto;
632 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
633 if (sset_contains(&ofproto->ports, name)) {
642 type_run(const char *type)
644 static long long int push_timer = LLONG_MIN;
645 struct dpif_backer *backer;
647 backer = shash_find_data(&all_dpif_backers, type);
649 /* This is not necessarily a problem, since backers are only
650 * created on demand. */
654 dpif_run(backer->dpif);
656 /* The most natural place to push facet statistics is when they're pulled
657 * from the datapath. However, when there are many flows in the datapath,
658 * this expensive operation can occur so frequently, that it reduces our
659 * ability to quickly set up flows. To reduce the cost, we push statistics
661 if (time_msec() > push_timer) {
662 push_timer = time_msec() + 2000;
666 /* If vswitchd started with other_config:flow_restore_wait set as "true",
667 * and the configuration has now changed to "false", enable receiving
668 * packets from the datapath. */
669 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
672 backer->recv_set_enable = true;
674 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
676 udpif_recv_set(backer->udpif, 0, false);
677 VLOG_ERR("Failed to enable receiving packets in dpif.");
680 udpif_recv_set(backer->udpif, n_handler_threads,
681 backer->recv_set_enable);
682 dpif_flow_flush(backer->dpif);
683 backer->need_revalidate = REV_RECONFIGURE;
686 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
687 * to reset the handler threads. */
688 if (backer->n_handler_threads != n_handler_threads) {
689 udpif_recv_set(backer->udpif, n_handler_threads,
690 backer->recv_set_enable);
691 backer->n_handler_threads = n_handler_threads;
694 if (backer->need_revalidate) {
695 struct ofproto_dpif *ofproto;
696 struct simap_node *node;
697 struct simap tmp_backers;
699 /* Handle tunnel garbage collection. */
700 simap_init(&tmp_backers);
701 simap_swap(&backer->tnl_backers, &tmp_backers);
703 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
704 struct ofport_dpif *iter;
706 if (backer != ofproto->backer) {
710 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
711 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
714 if (!iter->is_tunnel) {
718 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
719 namebuf, sizeof namebuf);
720 node = simap_find(&tmp_backers, dp_port);
722 simap_put(&backer->tnl_backers, dp_port, node->data);
723 simap_delete(&tmp_backers, node);
724 node = simap_find(&backer->tnl_backers, dp_port);
726 node = simap_find(&backer->tnl_backers, dp_port);
728 odp_port_t odp_port = ODPP_NONE;
730 if (!dpif_port_add(backer->dpif, iter->up.netdev,
732 simap_put(&backer->tnl_backers, dp_port,
733 odp_to_u32(odp_port));
734 node = simap_find(&backer->tnl_backers, dp_port);
739 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
740 if (tnl_port_reconfigure(iter, iter->up.netdev,
742 backer->need_revalidate = REV_RECONFIGURE;
747 SIMAP_FOR_EACH (node, &tmp_backers) {
748 dpif_port_del(backer->dpif, u32_to_odp(node->data));
750 simap_destroy(&tmp_backers);
752 switch (backer->need_revalidate) {
753 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
754 case REV_STP: COVERAGE_INC(rev_stp); break;
755 case REV_BOND: COVERAGE_INC(rev_bond); break;
756 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
757 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
758 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
759 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
761 backer->need_revalidate = 0;
763 /* Clear the drop_keys in case we should now be accepting some
764 * formerly dropped flows. */
765 drop_key_clear(backer);
767 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
768 struct facet *facet, *next;
769 struct ofport_dpif *ofport;
770 struct cls_cursor cursor;
771 struct ofbundle *bundle;
773 if (ofproto->backer != backer) {
777 ovs_rwlock_wrlock(&xlate_rwlock);
778 xlate_ofproto_set(ofproto, ofproto->up.name,
779 ofproto->backer->dpif, ofproto->miss_rule,
780 ofproto->no_packet_in_rule, ofproto->ml,
781 ofproto->stp, ofproto->mbridge,
782 ofproto->sflow, ofproto->ipfix,
783 ofproto->up.frag_handling,
784 ofproto->up.forward_bpdu,
785 connmgr_has_in_band(ofproto->up.connmgr),
786 ofproto->netflow != NULL);
788 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
789 xlate_bundle_set(ofproto, bundle, bundle->name,
790 bundle->vlan_mode, bundle->vlan,
791 bundle->trunks, bundle->use_priority_tags,
792 bundle->bond, bundle->lacp,
796 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
797 int stp_port = ofport->stp_port
798 ? stp_port_no(ofport->stp_port)
800 xlate_ofport_set(ofproto, ofport->bundle, ofport,
801 ofport->up.ofp_port, ofport->odp_port,
802 ofport->up.netdev, ofport->cfm,
803 ofport->bfd, ofport->peer, stp_port,
804 ofport->qdscp, ofport->n_qdscp,
805 ofport->up.pp.config, ofport->is_tunnel,
808 ovs_rwlock_unlock(&xlate_rwlock);
810 /* Only ofproto-dpif cares about the facet classifier so we just
811 * lock cls_cursor_init() to appease the thread safety analysis. */
812 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
813 cls_cursor_init(&cursor, &ofproto->facets, NULL);
814 ovs_rwlock_unlock(&ofproto->facets.rwlock);
815 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
816 facet_revalidate(facet);
821 udpif_revalidate(backer->udpif);
824 if (!backer->recv_set_enable) {
825 /* Wake up before a max of 1000ms. */
826 timer_set_duration(&backer->next_expiration, 1000);
827 } else if (timer_expired(&backer->next_expiration)) {
828 int delay = expire(backer);
829 timer_set_duration(&backer->next_expiration, delay);
832 process_dpif_port_changes(backer);
834 if (backer->governor) {
837 governor_run(backer->governor);
839 /* If the governor has shrunk to its minimum size and the number of
840 * subfacets has dwindled, then drop the governor entirely.
842 * For hysteresis, the number of subfacets to drop the governor is
843 * smaller than the number needed to trigger its creation. */
844 n_subfacets = hmap_count(&backer->subfacets);
845 if (n_subfacets * 4 < flow_eviction_threshold
846 && governor_is_idle(backer->governor)) {
847 governor_destroy(backer->governor);
848 backer->governor = NULL;
855 /* Check for and handle port changes in 'backer''s dpif. */
857 process_dpif_port_changes(struct dpif_backer *backer)
863 error = dpif_port_poll(backer->dpif, &devname);
869 process_dpif_all_ports_changed(backer);
873 process_dpif_port_change(backer, devname);
878 process_dpif_port_error(backer, error);
885 process_dpif_all_ports_changed(struct dpif_backer *backer)
887 struct ofproto_dpif *ofproto;
888 struct dpif_port dpif_port;
889 struct dpif_port_dump dump;
890 struct sset devnames;
893 sset_init(&devnames);
894 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
895 if (ofproto->backer == backer) {
896 struct ofport *ofport;
898 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
899 sset_add(&devnames, netdev_get_name(ofport->netdev));
903 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
904 sset_add(&devnames, dpif_port.name);
907 SSET_FOR_EACH (devname, &devnames) {
908 process_dpif_port_change(backer, devname);
910 sset_destroy(&devnames);
914 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
916 struct ofproto_dpif *ofproto;
917 struct dpif_port port;
919 /* Don't report on the datapath's device. */
920 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
924 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
925 &all_ofproto_dpifs) {
926 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
931 ofproto = lookup_ofproto_dpif_by_port_name(devname);
932 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
933 /* The port was removed. If we know the datapath,
934 * report it through poll_set(). If we don't, it may be
935 * notifying us of a removal we initiated, so ignore it.
936 * If there's a pending ENOBUFS, let it stand, since
937 * everything will be reevaluated. */
938 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
939 sset_add(&ofproto->port_poll_set, devname);
940 ofproto->port_poll_errno = 0;
942 } else if (!ofproto) {
943 /* The port was added, but we don't know with which
944 * ofproto we should associate it. Delete it. */
945 dpif_port_del(backer->dpif, port.port_no);
947 struct ofport_dpif *ofport;
949 ofport = ofport_dpif_cast(shash_find_data(
950 &ofproto->up.port_by_name, devname));
952 && ofport->odp_port != port.port_no
953 && !odp_port_to_ofport(backer, port.port_no))
955 /* 'ofport''s datapath port number has changed from
956 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
957 * structures to match. */
958 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
959 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
960 ofport->odp_port = port.port_no;
961 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
962 hash_odp_port(port.port_no));
963 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
964 backer->need_revalidate = REV_RECONFIGURE;
967 dpif_port_destroy(&port);
970 /* Propagate 'error' to all ofprotos based on 'backer'. */
972 process_dpif_port_error(struct dpif_backer *backer, int error)
974 struct ofproto_dpif *ofproto;
976 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
977 if (ofproto->backer == backer) {
978 sset_clear(&ofproto->port_poll_set);
979 ofproto->port_poll_errno = error;
985 dpif_backer_run_fast(struct dpif_backer *backer)
987 handle_upcalls(backer);
993 type_run_fast(const char *type)
995 struct dpif_backer *backer;
997 backer = shash_find_data(&all_dpif_backers, type);
999 /* This is not necessarily a problem, since backers are only
1000 * created on demand. */
1004 return dpif_backer_run_fast(backer);
1010 static long long int port_rl = LLONG_MIN;
1012 if (time_msec() >= port_rl) {
1013 struct ofproto_dpif *ofproto;
1015 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1016 run_fast(&ofproto->up);
1018 port_rl = time_msec() + 200;
1023 type_wait(const char *type)
1025 struct dpif_backer *backer;
1027 backer = shash_find_data(&all_dpif_backers, type);
1029 /* This is not necessarily a problem, since backers are only
1030 * created on demand. */
1034 if (backer->governor) {
1035 governor_wait(backer->governor);
1038 timer_wait(&backer->next_expiration);
1039 dpif_wait(backer->dpif);
1040 udpif_wait(backer->udpif);
1043 /* Basic life-cycle. */
1045 static int add_internal_flows(struct ofproto_dpif *);
1047 static struct ofproto *
1050 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1051 return &ofproto->up;
1055 dealloc(struct ofproto *ofproto_)
1057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1062 close_dpif_backer(struct dpif_backer *backer)
1064 ovs_assert(backer->refcount > 0);
1066 if (--backer->refcount) {
1070 drop_key_clear(backer);
1071 hmap_destroy(&backer->drop_keys);
1073 udpif_destroy(backer->udpif);
1075 simap_destroy(&backer->tnl_backers);
1076 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1077 hmap_destroy(&backer->odp_to_ofport_map);
1078 shash_find_and_delete(&all_dpif_backers, backer->type);
1080 dpif_close(backer->dpif);
1082 ovs_assert(hmap_is_empty(&backer->subfacets));
1083 hmap_destroy(&backer->subfacets);
1084 governor_destroy(backer->governor);
1089 /* Datapath port slated for removal from datapath. */
1090 struct odp_garbage {
1091 struct list list_node;
1092 odp_port_t odp_port;
1096 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1098 struct dpif_backer *backer;
1099 struct dpif_port_dump port_dump;
1100 struct dpif_port port;
1101 struct shash_node *node;
1102 struct list garbage_list;
1103 struct odp_garbage *garbage, *next;
1109 backer = shash_find_data(&all_dpif_backers, type);
1116 backer_name = xasprintf("ovs-%s", type);
1118 /* Remove any existing datapaths, since we assume we're the only
1119 * userspace controlling the datapath. */
1121 dp_enumerate_names(type, &names);
1122 SSET_FOR_EACH(name, &names) {
1123 struct dpif *old_dpif;
1125 /* Don't remove our backer if it exists. */
1126 if (!strcmp(name, backer_name)) {
1130 if (dpif_open(name, type, &old_dpif)) {
1131 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1133 dpif_delete(old_dpif);
1134 dpif_close(old_dpif);
1137 sset_destroy(&names);
1139 backer = xmalloc(sizeof *backer);
1141 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1144 VLOG_ERR("failed to open datapath of type %s: %s", type,
1145 ovs_strerror(error));
1149 backer->udpif = udpif_create(backer, backer->dpif);
1151 backer->type = xstrdup(type);
1152 backer->governor = NULL;
1153 backer->refcount = 1;
1154 hmap_init(&backer->odp_to_ofport_map);
1155 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1156 hmap_init(&backer->drop_keys);
1157 hmap_init(&backer->subfacets);
1158 timer_set_duration(&backer->next_expiration, 1000);
1159 backer->need_revalidate = 0;
1160 simap_init(&backer->tnl_backers);
1161 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1164 if (backer->recv_set_enable) {
1165 dpif_flow_flush(backer->dpif);
1168 /* Loop through the ports already on the datapath and remove any
1169 * that we don't need anymore. */
1170 list_init(&garbage_list);
1171 dpif_port_dump_start(&port_dump, backer->dpif);
1172 while (dpif_port_dump_next(&port_dump, &port)) {
1173 node = shash_find(&init_ofp_ports, port.name);
1174 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1175 garbage = xmalloc(sizeof *garbage);
1176 garbage->odp_port = port.port_no;
1177 list_push_front(&garbage_list, &garbage->list_node);
1180 dpif_port_dump_done(&port_dump);
1182 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1183 dpif_port_del(backer->dpif, garbage->odp_port);
1184 list_remove(&garbage->list_node);
1188 shash_add(&all_dpif_backers, type, backer);
1190 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1192 VLOG_ERR("failed to listen on datapath of type %s: %s",
1193 type, ovs_strerror(error));
1194 close_dpif_backer(backer);
1197 udpif_recv_set(backer->udpif, n_handler_threads,
1198 backer->recv_set_enable);
1199 backer->n_handler_threads = n_handler_threads;
1201 backer->max_n_subfacet = 0;
1202 backer->created = time_msec();
1203 backer->avg_n_subfacet = 0;
1204 backer->avg_subfacet_life = 0;
1210 construct(struct ofproto *ofproto_)
1212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1213 struct shash_node *node, *next;
1217 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1222 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1223 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1225 ofproto->netflow = NULL;
1226 ofproto->sflow = NULL;
1227 ofproto->ipfix = NULL;
1228 ofproto->stp = NULL;
1229 hmap_init(&ofproto->bundles);
1230 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1231 ofproto->mbridge = mbridge_create();
1232 ofproto->has_bonded_bundles = false;
1233 ovs_mutex_init(&ofproto->stats_mutex);
1234 ovs_mutex_init(&ofproto->vsp_mutex);
1236 classifier_init(&ofproto->facets);
1237 ofproto->consistency_rl = LLONG_MIN;
1239 guarded_list_init(&ofproto->pins);
1241 ofproto_dpif_unixctl_init();
1243 hmap_init(&ofproto->vlandev_map);
1244 hmap_init(&ofproto->realdev_vid_map);
1246 sset_init(&ofproto->ports);
1247 sset_init(&ofproto->ghost_ports);
1248 sset_init(&ofproto->port_poll_set);
1249 ofproto->port_poll_errno = 0;
1251 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1252 struct iface_hint *iface_hint = node->data;
1254 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1255 /* Check if the datapath already has this port. */
1256 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1257 sset_add(&ofproto->ports, node->name);
1260 free(iface_hint->br_name);
1261 free(iface_hint->br_type);
1263 shash_delete(&init_ofp_ports, node);
1267 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1268 hash_string(ofproto->up.name, 0));
1269 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1271 ofproto_init_tables(ofproto_, N_TABLES);
1272 error = add_internal_flows(ofproto);
1273 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1276 ofproto->n_missed = 0;
1282 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1283 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1285 struct ofputil_flow_mod fm;
1288 match_init_catchall(&fm.match);
1290 match_set_reg(&fm.match, 0, id);
1291 fm.new_cookie = htonll(0);
1292 fm.cookie = htonll(0);
1293 fm.cookie_mask = htonll(0);
1294 fm.modify_cookie = false;
1295 fm.table_id = TBL_INTERNAL;
1296 fm.command = OFPFC_ADD;
1297 fm.idle_timeout = 0;
1298 fm.hard_timeout = 0;
1302 fm.ofpacts = ofpacts->data;
1303 fm.ofpacts_len = ofpacts->size;
1305 error = ofproto_flow_mod(&ofproto->up, &fm);
1307 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1308 id, ofperr_to_string(error));
1312 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1314 rule_dpif_unref(*rulep);
1323 add_internal_flows(struct ofproto_dpif *ofproto)
1325 struct ofpact_controller *controller;
1326 uint64_t ofpacts_stub[128 / 8];
1327 struct ofpbuf ofpacts;
1331 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1334 controller = ofpact_put_CONTROLLER(&ofpacts);
1335 controller->max_len = UINT16_MAX;
1336 controller->controller_id = 0;
1337 controller->reason = OFPR_NO_MATCH;
1338 ofpact_pad(&ofpacts);
1340 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1345 ofpbuf_clear(&ofpacts);
1346 error = add_internal_flow(ofproto, id++, &ofpacts,
1347 &ofproto->no_packet_in_rule);
1352 error = add_internal_flow(ofproto, id++, &ofpacts,
1353 &ofproto->drop_frags_rule);
1358 destruct(struct ofproto *ofproto_)
1360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1361 struct rule_dpif *rule, *next_rule;
1362 struct ofputil_packet_in *pin, *next_pin;
1363 struct facet *facet, *next_facet;
1364 struct cls_cursor cursor;
1365 struct oftable *table;
1368 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1369 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1370 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1371 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1372 facet_remove(facet);
1375 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1376 ovs_rwlock_wrlock(&xlate_rwlock);
1377 xlate_remove_ofproto(ofproto);
1378 ovs_rwlock_unlock(&xlate_rwlock);
1380 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1381 * use-after-free error. */
1382 udpif_revalidate(ofproto->backer->udpif);
1384 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1386 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1387 struct cls_cursor cursor;
1389 ovs_rwlock_rdlock(&table->cls.rwlock);
1390 cls_cursor_init(&cursor, &table->cls, NULL);
1391 ovs_rwlock_unlock(&table->cls.rwlock);
1392 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1393 ofproto_rule_delete(&ofproto->up, &rule->up);
1397 guarded_list_pop_all(&ofproto->pins, &pins);
1398 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1399 list_remove(&pin->list_node);
1400 free(CONST_CAST(void *, pin->packet));
1403 guarded_list_destroy(&ofproto->pins);
1405 mbridge_unref(ofproto->mbridge);
1407 netflow_destroy(ofproto->netflow);
1408 dpif_sflow_unref(ofproto->sflow);
1409 hmap_destroy(&ofproto->bundles);
1410 mac_learning_unref(ofproto->ml);
1412 classifier_destroy(&ofproto->facets);
1414 hmap_destroy(&ofproto->vlandev_map);
1415 hmap_destroy(&ofproto->realdev_vid_map);
1417 sset_destroy(&ofproto->ports);
1418 sset_destroy(&ofproto->ghost_ports);
1419 sset_destroy(&ofproto->port_poll_set);
1421 ovs_mutex_destroy(&ofproto->stats_mutex);
1422 ovs_mutex_destroy(&ofproto->vsp_mutex);
1424 close_dpif_backer(ofproto->backer);
1428 run_fast(struct ofproto *ofproto_)
1430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1431 struct ofputil_packet_in *pin, *next_pin;
1434 /* Do not perform any periodic activity required by 'ofproto' while
1435 * waiting for flow restore to complete. */
1436 if (ofproto_get_flow_restore_wait()) {
1440 guarded_list_pop_all(&ofproto->pins, &pins);
1441 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1442 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1443 list_remove(&pin->list_node);
1444 free(CONST_CAST(void *, pin->packet));
1448 ofproto_dpif_monitor_run_fast();
1453 run(struct ofproto *ofproto_)
1455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1456 struct ofport_dpif *ofport;
1457 struct ofbundle *bundle;
1460 if (mbridge_need_revalidate(ofproto->mbridge)) {
1461 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1462 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1463 mac_learning_flush(ofproto->ml);
1464 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1467 /* Do not perform any periodic activity below required by 'ofproto' while
1468 * waiting for flow restore to complete. */
1469 if (ofproto_get_flow_restore_wait()) {
1473 error = run_fast(ofproto_);
1478 if (ofproto->netflow) {
1479 if (netflow_run(ofproto->netflow)) {
1480 send_netflow_active_timeouts(ofproto);
1483 if (ofproto->sflow) {
1484 dpif_sflow_run(ofproto->sflow);
1486 if (ofproto->ipfix) {
1487 dpif_ipfix_run(ofproto->ipfix);
1490 ofproto_dpif_monitor_run_fast();
1491 ofproto_dpif_monitor_run();
1493 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1496 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1501 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1502 if (mac_learning_run(ofproto->ml)) {
1503 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1505 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1507 /* Check the consistency of a random facet, to aid debugging. */
1508 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1509 if (time_msec() >= ofproto->consistency_rl
1510 && !classifier_is_empty(&ofproto->facets)
1511 && !ofproto->backer->need_revalidate) {
1512 struct cls_table *table;
1513 struct cls_rule *cr;
1514 struct facet *facet;
1516 ofproto->consistency_rl = time_msec() + 250;
1518 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1519 struct cls_table, hmap_node);
1520 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1522 facet = CONTAINER_OF(cr, struct facet, cr);
1524 if (!facet_check_consistency(facet)) {
1525 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1528 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1534 wait(struct ofproto *ofproto_)
1536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1537 struct ofbundle *bundle;
1539 if (ofproto_get_flow_restore_wait()) {
1543 if (ofproto->sflow) {
1544 dpif_sflow_wait(ofproto->sflow);
1546 if (ofproto->ipfix) {
1547 dpif_ipfix_wait(ofproto->ipfix);
1549 ofproto_dpif_monitor_wait();
1550 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1551 bundle_wait(bundle);
1553 if (ofproto->netflow) {
1554 netflow_wait(ofproto->netflow);
1556 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1557 mac_learning_wait(ofproto->ml);
1558 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1560 if (ofproto->backer->need_revalidate) {
1561 /* Shouldn't happen, but if it does just go around again. */
1562 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1563 poll_immediate_wake();
1568 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1570 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1571 struct cls_cursor cursor;
1572 size_t n_subfacets = 0;
1573 struct facet *facet;
1575 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1576 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1577 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1579 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1580 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1581 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1582 n_subfacets += list_size(&facet->subfacets);
1584 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1585 simap_increase(usage, "subfacets", n_subfacets);
1589 flush(struct ofproto *ofproto_)
1591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1592 struct subfacet *subfacet, *next_subfacet;
1593 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1597 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1598 &ofproto->backer->subfacets) {
1599 if (subfacet->facet->ofproto != ofproto) {
1603 if (subfacet->path != SF_NOT_INSTALLED) {
1604 batch[n_batch++] = subfacet;
1605 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1606 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1610 subfacet_destroy(subfacet);
1615 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1620 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1621 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1623 *arp_match_ip = true;
1624 *actions = (OFPUTIL_A_OUTPUT |
1625 OFPUTIL_A_SET_VLAN_VID |
1626 OFPUTIL_A_SET_VLAN_PCP |
1627 OFPUTIL_A_STRIP_VLAN |
1628 OFPUTIL_A_SET_DL_SRC |
1629 OFPUTIL_A_SET_DL_DST |
1630 OFPUTIL_A_SET_NW_SRC |
1631 OFPUTIL_A_SET_NW_DST |
1632 OFPUTIL_A_SET_NW_TOS |
1633 OFPUTIL_A_SET_TP_SRC |
1634 OFPUTIL_A_SET_TP_DST |
1639 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1641 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1642 struct dpif_dp_stats s;
1643 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1646 strcpy(ots->name, "classifier");
1648 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1649 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1650 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1651 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1653 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1654 ots->lookup_count = htonll(n_lookup);
1655 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1658 static struct ofport *
1661 struct ofport_dpif *port = xmalloc(sizeof *port);
1666 port_dealloc(struct ofport *port_)
1668 struct ofport_dpif *port = ofport_dpif_cast(port_);
1673 port_construct(struct ofport *port_)
1675 struct ofport_dpif *port = ofport_dpif_cast(port_);
1676 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1677 const struct netdev *netdev = port->up.netdev;
1678 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1679 struct dpif_port dpif_port;
1682 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1683 port->bundle = NULL;
1686 port->may_enable = true;
1687 port->stp_port = NULL;
1688 port->stp_state = STP_DISABLED;
1689 port->is_tunnel = false;
1693 port->realdev_ofp_port = 0;
1694 port->vlandev_vid = 0;
1695 port->carrier_seq = netdev_get_carrier_resets(netdev);
1697 if (netdev_vport_is_patch(netdev)) {
1698 /* By bailing out here, we don't submit the port to the sFlow module
1699 * to be considered for counter polling export. This is correct
1700 * because the patch port represents an interface that sFlow considers
1701 * to be "internal" to the switch as a whole, and therefore not an
1702 * candidate for counter polling. */
1703 port->odp_port = ODPP_NONE;
1704 ofport_update_peer(port);
1708 error = dpif_port_query_by_name(ofproto->backer->dpif,
1709 netdev_vport_get_dpif_port(netdev, namebuf,
1716 port->odp_port = dpif_port.port_no;
1718 if (netdev_get_tunnel_config(netdev)) {
1719 tnl_port_add(port, port->up.netdev, port->odp_port);
1720 port->is_tunnel = true;
1722 /* Sanity-check that a mapping doesn't already exist. This
1723 * shouldn't happen for non-tunnel ports. */
1724 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1725 VLOG_ERR("port %s already has an OpenFlow port number",
1727 dpif_port_destroy(&dpif_port);
1731 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1732 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1733 hash_odp_port(port->odp_port));
1734 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1736 dpif_port_destroy(&dpif_port);
1738 if (ofproto->sflow) {
1739 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1746 port_destruct(struct ofport *port_)
1748 struct ofport_dpif *port = ofport_dpif_cast(port_);
1749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1750 const char *devname = netdev_get_name(port->up.netdev);
1751 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1752 const char *dp_port_name;
1754 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1755 ovs_rwlock_wrlock(&xlate_rwlock);
1756 xlate_ofport_remove(port);
1757 ovs_rwlock_unlock(&xlate_rwlock);
1759 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1761 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1762 /* The underlying device is still there, so delete it. This
1763 * happens when the ofproto is being destroyed, since the caller
1764 * assumes that removal of attached ports will happen as part of
1766 if (!port->is_tunnel) {
1767 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1772 port->peer->peer = NULL;
1776 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1777 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1778 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1779 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1783 sset_find_and_delete(&ofproto->ports, devname);
1784 sset_find_and_delete(&ofproto->ghost_ports, devname);
1785 bundle_remove(port_);
1786 set_cfm(port_, NULL);
1787 set_bfd(port_, NULL);
1788 if (ofproto->sflow) {
1789 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1796 port_modified(struct ofport *port_)
1798 struct ofport_dpif *port = ofport_dpif_cast(port_);
1800 if (port->bundle && port->bundle->bond) {
1801 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1805 cfm_set_netdev(port->cfm, port->up.netdev);
1809 bfd_set_netdev(port->bfd, port->up.netdev);
1812 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1813 port->up.pp.hw_addr);
1815 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1817 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1821 ofport_update_peer(port);
1825 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1827 struct ofport_dpif *port = ofport_dpif_cast(port_);
1828 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1829 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1831 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1832 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1833 OFPUTIL_PC_NO_PACKET_IN)) {
1834 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1836 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1837 bundle_update(port->bundle);
1843 set_sflow(struct ofproto *ofproto_,
1844 const struct ofproto_sflow_options *sflow_options)
1846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1847 struct dpif_sflow *ds = ofproto->sflow;
1849 if (sflow_options) {
1851 struct ofport_dpif *ofport;
1853 ds = ofproto->sflow = dpif_sflow_create();
1854 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1855 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1857 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1859 dpif_sflow_set_options(ds, sflow_options);
1862 dpif_sflow_unref(ds);
1863 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1864 ofproto->sflow = NULL;
1872 struct ofproto *ofproto_,
1873 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1874 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1875 size_t n_flow_exporters_options)
1877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1878 struct dpif_ipfix *di = ofproto->ipfix;
1879 bool has_options = bridge_exporter_options || flow_exporters_options;
1881 if (has_options && !di) {
1882 di = ofproto->ipfix = dpif_ipfix_create();
1886 /* Call set_options in any case to cleanly flush the flow
1887 * caches in the last exporters that are to be destroyed. */
1888 dpif_ipfix_set_options(
1889 di, bridge_exporter_options, flow_exporters_options,
1890 n_flow_exporters_options);
1893 dpif_ipfix_unref(di);
1894 ofproto->ipfix = NULL;
1902 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1904 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1909 struct ofproto_dpif *ofproto;
1911 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1912 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1913 ofport->cfm = cfm_create(ofport->up.netdev);
1916 if (cfm_configure(ofport->cfm, s)) {
1923 cfm_unref(ofport->cfm);
1926 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1927 ofport->up.pp.hw_addr);
1932 get_cfm_status(const struct ofport *ofport_,
1933 struct ofproto_cfm_status *status)
1935 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1938 status->faults = cfm_get_fault(ofport->cfm);
1939 status->remote_opstate = cfm_get_opup(ofport->cfm);
1940 status->health = cfm_get_health(ofport->cfm);
1941 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1949 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1952 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1956 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1957 cfg, ofport->up.netdev);
1958 if (ofport->bfd != old) {
1959 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1961 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1962 ofport->up.pp.hw_addr);
1967 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1969 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1972 bfd_get_status(ofport->bfd, smap);
1979 /* Spanning Tree. */
1982 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1984 struct ofproto_dpif *ofproto = ofproto_;
1985 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1986 struct ofport_dpif *ofport;
1988 ofport = stp_port_get_aux(sp);
1990 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1991 ofproto->up.name, port_num);
1993 struct eth_header *eth = pkt->l2;
1995 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1996 if (eth_addr_is_zero(eth->eth_src)) {
1997 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1998 "with unknown MAC", ofproto->up.name, port_num);
2000 ofproto_dpif_send_packet(ofport, pkt);
2006 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2008 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2012 /* Only revalidate flows if the configuration changed. */
2013 if (!s != !ofproto->stp) {
2014 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2018 if (!ofproto->stp) {
2019 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2020 send_bpdu_cb, ofproto);
2021 ofproto->stp_last_tick = time_msec();
2024 stp_set_bridge_id(ofproto->stp, s->system_id);
2025 stp_set_bridge_priority(ofproto->stp, s->priority);
2026 stp_set_hello_time(ofproto->stp, s->hello_time);
2027 stp_set_max_age(ofproto->stp, s->max_age);
2028 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2030 struct ofport *ofport;
2032 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2033 set_stp_port(ofport, NULL);
2036 stp_unref(ofproto->stp);
2037 ofproto->stp = NULL;
2044 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2050 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2051 s->designated_root = stp_get_designated_root(ofproto->stp);
2052 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2061 update_stp_port_state(struct ofport_dpif *ofport)
2063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2064 enum stp_state state;
2066 /* Figure out new state. */
2067 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2071 if (ofport->stp_state != state) {
2072 enum ofputil_port_state of_state;
2075 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2076 netdev_get_name(ofport->up.netdev),
2077 stp_state_name(ofport->stp_state),
2078 stp_state_name(state));
2079 if (stp_learn_in_state(ofport->stp_state)
2080 != stp_learn_in_state(state)) {
2081 /* xxx Learning action flows should also be flushed. */
2082 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2083 mac_learning_flush(ofproto->ml);
2084 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2086 fwd_change = stp_forward_in_state(ofport->stp_state)
2087 != stp_forward_in_state(state);
2089 ofproto->backer->need_revalidate = REV_STP;
2090 ofport->stp_state = state;
2091 ofport->stp_state_entered = time_msec();
2093 if (fwd_change && ofport->bundle) {
2094 bundle_update(ofport->bundle);
2097 /* Update the STP state bits in the OpenFlow port description. */
2098 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2099 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2100 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2101 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2102 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2104 ofproto_port_set_state(&ofport->up, of_state);
2108 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2109 * caller is responsible for assigning STP port numbers and ensuring
2110 * there are no duplicates. */
2112 set_stp_port(struct ofport *ofport_,
2113 const struct ofproto_port_stp_settings *s)
2115 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2116 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2117 struct stp_port *sp = ofport->stp_port;
2119 if (!s || !s->enable) {
2121 ofport->stp_port = NULL;
2122 stp_port_disable(sp);
2123 update_stp_port_state(ofport);
2126 } else if (sp && stp_port_no(sp) != s->port_num
2127 && ofport == stp_port_get_aux(sp)) {
2128 /* The port-id changed, so disable the old one if it's not
2129 * already in use by another port. */
2130 stp_port_disable(sp);
2133 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2134 stp_port_enable(sp);
2136 stp_port_set_aux(sp, ofport);
2137 stp_port_set_priority(sp, s->priority);
2138 stp_port_set_path_cost(sp, s->path_cost);
2140 update_stp_port_state(ofport);
2146 get_stp_port_status(struct ofport *ofport_,
2147 struct ofproto_port_stp_status *s)
2149 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2151 struct stp_port *sp = ofport->stp_port;
2153 if (!ofproto->stp || !sp) {
2159 s->port_id = stp_port_get_id(sp);
2160 s->state = stp_port_get_state(sp);
2161 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2162 s->role = stp_port_get_role(sp);
2163 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2169 stp_run(struct ofproto_dpif *ofproto)
2172 long long int now = time_msec();
2173 long long int elapsed = now - ofproto->stp_last_tick;
2174 struct stp_port *sp;
2177 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2178 ofproto->stp_last_tick = now;
2180 while (stp_get_changed_port(ofproto->stp, &sp)) {
2181 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2184 update_stp_port_state(ofport);
2188 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2189 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2190 mac_learning_flush(ofproto->ml);
2191 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2197 stp_wait(struct ofproto_dpif *ofproto)
2200 poll_timer_wait(1000);
2205 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2208 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2211 if (ofport->n_qdscp != n_qdscp
2212 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2213 n_qdscp * sizeof *qdscp))) {
2214 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2215 free(ofport->qdscp);
2216 ofport->qdscp = n_qdscp
2217 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2219 ofport->n_qdscp = n_qdscp;
2227 /* Expires all MAC learning entries associated with 'bundle' and forces its
2228 * ofproto to revalidate every flow.
2230 * Normally MAC learning entries are removed only from the ofproto associated
2231 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2232 * are removed from every ofproto. When patch ports and SLB bonds are in use
2233 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2234 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2235 * with the host from which it migrated. */
2237 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2239 struct ofproto_dpif *ofproto = bundle->ofproto;
2240 struct mac_learning *ml = ofproto->ml;
2241 struct mac_entry *mac, *next_mac;
2243 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2244 ovs_rwlock_wrlock(&ml->rwlock);
2245 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2246 if (mac->port.p == bundle) {
2248 struct ofproto_dpif *o;
2250 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2252 struct mac_entry *e;
2254 ovs_rwlock_wrlock(&o->ml->rwlock);
2255 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2257 mac_learning_expire(o->ml, e);
2259 ovs_rwlock_unlock(&o->ml->rwlock);
2264 mac_learning_expire(ml, mac);
2267 ovs_rwlock_unlock(&ml->rwlock);
2270 static struct ofbundle *
2271 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2273 struct ofbundle *bundle;
2275 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2276 &ofproto->bundles) {
2277 if (bundle->aux == aux) {
2285 bundle_update(struct ofbundle *bundle)
2287 struct ofport_dpif *port;
2289 bundle->floodable = true;
2290 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2291 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2292 || !stp_forward_in_state(port->stp_state)) {
2293 bundle->floodable = false;
2300 bundle_del_port(struct ofport_dpif *port)
2302 struct ofbundle *bundle = port->bundle;
2304 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2306 list_remove(&port->bundle_node);
2307 port->bundle = NULL;
2310 lacp_slave_unregister(bundle->lacp, port);
2313 bond_slave_unregister(bundle->bond, port);
2316 bundle_update(bundle);
2320 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2321 struct lacp_slave_settings *lacp)
2323 struct ofport_dpif *port;
2325 port = get_ofp_port(bundle->ofproto, ofp_port);
2330 if (port->bundle != bundle) {
2331 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2333 bundle_remove(&port->up);
2336 port->bundle = bundle;
2337 list_push_back(&bundle->ports, &port->bundle_node);
2338 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2339 || !stp_forward_in_state(port->stp_state)) {
2340 bundle->floodable = false;
2344 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2345 lacp_slave_register(bundle->lacp, port, lacp);
2352 bundle_destroy(struct ofbundle *bundle)
2354 struct ofproto_dpif *ofproto;
2355 struct ofport_dpif *port, *next_port;
2361 ofproto = bundle->ofproto;
2362 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2364 ovs_rwlock_wrlock(&xlate_rwlock);
2365 xlate_bundle_remove(bundle);
2366 ovs_rwlock_unlock(&xlate_rwlock);
2368 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2369 bundle_del_port(port);
2372 bundle_flush_macs(bundle, true);
2373 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2375 free(bundle->trunks);
2376 lacp_unref(bundle->lacp);
2377 bond_unref(bundle->bond);
2382 bundle_set(struct ofproto *ofproto_, void *aux,
2383 const struct ofproto_bundle_settings *s)
2385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2386 bool need_flush = false;
2387 struct ofport_dpif *port;
2388 struct ofbundle *bundle;
2389 unsigned long *trunks;
2395 bundle_destroy(bundle_lookup(ofproto, aux));
2399 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2400 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2402 bundle = bundle_lookup(ofproto, aux);
2404 bundle = xmalloc(sizeof *bundle);
2406 bundle->ofproto = ofproto;
2407 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2408 hash_pointer(aux, 0));
2410 bundle->name = NULL;
2412 list_init(&bundle->ports);
2413 bundle->vlan_mode = PORT_VLAN_TRUNK;
2415 bundle->trunks = NULL;
2416 bundle->use_priority_tags = s->use_priority_tags;
2417 bundle->lacp = NULL;
2418 bundle->bond = NULL;
2420 bundle->floodable = true;
2421 mbridge_register_bundle(ofproto->mbridge, bundle);
2424 if (!bundle->name || strcmp(s->name, bundle->name)) {
2426 bundle->name = xstrdup(s->name);
2431 if (!bundle->lacp) {
2432 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2433 bundle->lacp = lacp_create();
2435 lacp_configure(bundle->lacp, s->lacp);
2437 lacp_unref(bundle->lacp);
2438 bundle->lacp = NULL;
2441 /* Update set of ports. */
2443 for (i = 0; i < s->n_slaves; i++) {
2444 if (!bundle_add_port(bundle, s->slaves[i],
2445 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2449 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2450 struct ofport_dpif *next_port;
2452 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2453 for (i = 0; i < s->n_slaves; i++) {
2454 if (s->slaves[i] == port->up.ofp_port) {
2459 bundle_del_port(port);
2463 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2465 if (list_is_empty(&bundle->ports)) {
2466 bundle_destroy(bundle);
2470 /* Set VLAN tagging mode */
2471 if (s->vlan_mode != bundle->vlan_mode
2472 || s->use_priority_tags != bundle->use_priority_tags) {
2473 bundle->vlan_mode = s->vlan_mode;
2474 bundle->use_priority_tags = s->use_priority_tags;
2479 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2480 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2482 if (vlan != bundle->vlan) {
2483 bundle->vlan = vlan;
2487 /* Get trunked VLANs. */
2488 switch (s->vlan_mode) {
2489 case PORT_VLAN_ACCESS:
2493 case PORT_VLAN_TRUNK:
2494 trunks = CONST_CAST(unsigned long *, s->trunks);
2497 case PORT_VLAN_NATIVE_UNTAGGED:
2498 case PORT_VLAN_NATIVE_TAGGED:
2499 if (vlan != 0 && (!s->trunks
2500 || !bitmap_is_set(s->trunks, vlan)
2501 || bitmap_is_set(s->trunks, 0))) {
2502 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2504 trunks = bitmap_clone(s->trunks, 4096);
2506 trunks = bitmap_allocate1(4096);
2508 bitmap_set1(trunks, vlan);
2509 bitmap_set0(trunks, 0);
2511 trunks = CONST_CAST(unsigned long *, s->trunks);
2518 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2519 free(bundle->trunks);
2520 if (trunks == s->trunks) {
2521 bundle->trunks = vlan_bitmap_clone(trunks);
2523 bundle->trunks = trunks;
2528 if (trunks != s->trunks) {
2533 if (!list_is_short(&bundle->ports)) {
2534 bundle->ofproto->has_bonded_bundles = true;
2536 if (bond_reconfigure(bundle->bond, s->bond)) {
2537 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2540 bundle->bond = bond_create(s->bond);
2541 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2544 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2545 bond_slave_register(bundle->bond, port, port->up.netdev);
2548 bond_unref(bundle->bond);
2549 bundle->bond = NULL;
2552 /* If we changed something that would affect MAC learning, un-learn
2553 * everything on this port and force flow revalidation. */
2555 bundle_flush_macs(bundle, false);
2562 bundle_remove(struct ofport *port_)
2564 struct ofport_dpif *port = ofport_dpif_cast(port_);
2565 struct ofbundle *bundle = port->bundle;
2568 bundle_del_port(port);
2569 if (list_is_empty(&bundle->ports)) {
2570 bundle_destroy(bundle);
2571 } else if (list_is_short(&bundle->ports)) {
2572 bond_unref(bundle->bond);
2573 bundle->bond = NULL;
2579 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2581 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2582 struct ofport_dpif *port = port_;
2583 uint8_t ea[ETH_ADDR_LEN];
2586 error = netdev_get_etheraddr(port->up.netdev, ea);
2588 struct ofpbuf packet;
2591 ofpbuf_init(&packet, 0);
2592 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2594 memcpy(packet_pdu, pdu, pdu_size);
2596 ofproto_dpif_send_packet(port, &packet);
2597 ofpbuf_uninit(&packet);
2599 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2600 "%s (%s)", port->bundle->name,
2601 netdev_get_name(port->up.netdev), ovs_strerror(error));
2606 bundle_send_learning_packets(struct ofbundle *bundle)
2608 struct ofproto_dpif *ofproto = bundle->ofproto;
2609 struct ofpbuf *learning_packet;
2610 int error, n_packets, n_errors;
2611 struct mac_entry *e;
2612 struct list packets;
2614 list_init(&packets);
2615 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2616 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2617 if (e->port.p != bundle) {
2620 learning_packet = bond_compose_learning_packet(bundle->bond,
2623 learning_packet->private_p = port_void;
2624 list_push_back(&packets, &learning_packet->list_node);
2627 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2629 error = n_packets = n_errors = 0;
2630 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2633 ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2640 ofpbuf_list_delete(&packets);
2643 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2644 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2645 "packets, last error was: %s",
2646 bundle->name, n_errors, n_packets, ovs_strerror(error));
2648 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2649 bundle->name, n_packets);
2654 bundle_run(struct ofbundle *bundle)
2657 lacp_run(bundle->lacp, send_pdu_cb);
2660 struct ofport_dpif *port;
2662 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2663 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2666 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2667 bundle->ofproto->backer->need_revalidate = REV_BOND;
2670 if (bond_should_send_learning_packets(bundle->bond)) {
2671 bundle_send_learning_packets(bundle);
2677 bundle_wait(struct ofbundle *bundle)
2680 lacp_wait(bundle->lacp);
2683 bond_wait(bundle->bond);
2690 mirror_set__(struct ofproto *ofproto_, void *aux,
2691 const struct ofproto_mirror_settings *s)
2693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2694 struct ofbundle **srcs, **dsts;
2699 mirror_destroy(ofproto->mbridge, aux);
2703 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2704 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2706 for (i = 0; i < s->n_srcs; i++) {
2707 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2710 for (i = 0; i < s->n_dsts; i++) {
2711 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2714 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2715 s->n_dsts, s->src_vlans,
2716 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2723 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2724 uint64_t *packets, uint64_t *bytes)
2727 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2732 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2735 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2736 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2737 mac_learning_flush(ofproto->ml);
2739 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2744 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2747 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2748 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2752 forward_bpdu_changed(struct ofproto *ofproto_)
2754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2755 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2759 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2763 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2764 mac_learning_set_idle_time(ofproto->ml, idle_time);
2765 mac_learning_set_max_entries(ofproto->ml, max_entries);
2766 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2771 static struct ofport_dpif *
2772 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2774 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2775 return ofport ? ofport_dpif_cast(ofport) : NULL;
2778 static struct ofport_dpif *
2779 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2781 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2782 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2786 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2787 struct ofproto_port *ofproto_port,
2788 struct dpif_port *dpif_port)
2790 ofproto_port->name = dpif_port->name;
2791 ofproto_port->type = dpif_port->type;
2792 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2796 ofport_update_peer(struct ofport_dpif *ofport)
2798 const struct ofproto_dpif *ofproto;
2799 struct dpif_backer *backer;
2802 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2806 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2807 backer->need_revalidate = REV_RECONFIGURE;
2810 ofport->peer->peer = NULL;
2811 ofport->peer = NULL;
2814 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2819 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2820 struct ofport *peer_ofport;
2821 struct ofport_dpif *peer;
2824 if (ofproto->backer != backer) {
2828 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2833 peer = ofport_dpif_cast(peer_ofport);
2834 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2835 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2837 ofport->peer = peer;
2838 ofport->peer->peer = ofport;
2848 port_run(struct ofport_dpif *ofport)
2850 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2851 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2852 bool enable = netdev_get_carrier(ofport->up.netdev);
2853 bool cfm_enable = false;
2854 bool bfd_enable = false;
2856 ofport->carrier_seq = carrier_seq;
2859 int cfm_opup = cfm_get_opup(ofport->cfm);
2861 cfm_enable = !cfm_get_fault(ofport->cfm);
2863 if (cfm_opup >= 0) {
2864 cfm_enable = cfm_enable && cfm_opup;
2869 bfd_enable = bfd_forwarding(ofport->bfd);
2872 if (ofport->bfd || ofport->cfm) {
2873 enable = enable && (cfm_enable || bfd_enable);
2876 if (ofport->bundle) {
2877 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2878 if (carrier_changed) {
2879 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2883 if (ofport->may_enable != enable) {
2884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2885 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2888 ofport->may_enable = enable;
2892 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2893 struct ofproto_port *ofproto_port)
2895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2896 struct dpif_port dpif_port;
2899 if (sset_contains(&ofproto->ghost_ports, devname)) {
2900 const char *type = netdev_get_type_from_name(devname);
2902 /* We may be called before ofproto->up.port_by_name is populated with
2903 * the appropriate ofport. For this reason, we must get the name and
2904 * type from the netdev layer directly. */
2906 const struct ofport *ofport;
2908 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2909 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2910 ofproto_port->name = xstrdup(devname);
2911 ofproto_port->type = xstrdup(type);
2917 if (!sset_contains(&ofproto->ports, devname)) {
2920 error = dpif_port_query_by_name(ofproto->backer->dpif,
2921 devname, &dpif_port);
2923 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2929 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2932 const char *devname = netdev_get_name(netdev);
2933 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2934 const char *dp_port_name;
2936 if (netdev_vport_is_patch(netdev)) {
2937 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2941 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2942 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2943 odp_port_t port_no = ODPP_NONE;
2946 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2950 if (netdev_get_tunnel_config(netdev)) {
2951 simap_put(&ofproto->backer->tnl_backers,
2952 dp_port_name, odp_to_u32(port_no));
2956 if (netdev_get_tunnel_config(netdev)) {
2957 sset_add(&ofproto->ghost_ports, devname);
2959 sset_add(&ofproto->ports, devname);
2965 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2968 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2975 sset_find_and_delete(&ofproto->ghost_ports,
2976 netdev_get_name(ofport->up.netdev));
2977 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2978 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2979 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2981 /* The caller is going to close ofport->up.netdev. If this is a
2982 * bonded port, then the bond is using that netdev, so remove it
2983 * from the bond. The client will need to reconfigure everything
2984 * after deleting ports, so then the slave will get re-added. */
2985 bundle_remove(&ofport->up);
2992 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2994 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2999 error = netdev_get_stats(ofport->up.netdev, stats);
3001 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3002 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3004 ovs_mutex_lock(&ofproto->stats_mutex);
3005 /* ofproto->stats.tx_packets represents packets that we created
3006 * internally and sent to some port (e.g. packets sent with
3007 * ofproto_dpif_send_packet()). Account for them as if they had
3008 * come from OFPP_LOCAL and got forwarded. */
3010 if (stats->rx_packets != UINT64_MAX) {
3011 stats->rx_packets += ofproto->stats.tx_packets;
3014 if (stats->rx_bytes != UINT64_MAX) {
3015 stats->rx_bytes += ofproto->stats.tx_bytes;
3018 /* ofproto->stats.rx_packets represents packets that were received on
3019 * some port and we processed internally and dropped (e.g. STP).
3020 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3022 if (stats->tx_packets != UINT64_MAX) {
3023 stats->tx_packets += ofproto->stats.rx_packets;
3026 if (stats->tx_bytes != UINT64_MAX) {
3027 stats->tx_bytes += ofproto->stats.rx_bytes;
3029 ovs_mutex_unlock(&ofproto->stats_mutex);
3035 struct port_dump_state {
3040 struct ofproto_port port;
3045 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3047 *statep = xzalloc(sizeof(struct port_dump_state));
3052 port_dump_next(const struct ofproto *ofproto_, void *state_,
3053 struct ofproto_port *port)
3055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3056 struct port_dump_state *state = state_;
3057 const struct sset *sset;
3058 struct sset_node *node;
3060 if (state->has_port) {
3061 ofproto_port_destroy(&state->port);
3062 state->has_port = false;
3064 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3065 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3068 error = port_query_by_name(ofproto_, node->name, &state->port);
3070 *port = state->port;
3071 state->has_port = true;
3073 } else if (error != ENODEV) {
3078 if (!state->ghost) {
3079 state->ghost = true;
3082 return port_dump_next(ofproto_, state_, port);
3089 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3091 struct port_dump_state *state = state_;
3093 if (state->has_port) {
3094 ofproto_port_destroy(&state->port);
3101 port_poll(const struct ofproto *ofproto_, char **devnamep)
3103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3105 if (ofproto->port_poll_errno) {
3106 int error = ofproto->port_poll_errno;
3107 ofproto->port_poll_errno = 0;
3111 if (sset_is_empty(&ofproto->port_poll_set)) {
3115 *devnamep = sset_pop(&ofproto->port_poll_set);
3120 port_poll_wait(const struct ofproto *ofproto_)
3122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3123 dpif_port_poll_wait(ofproto->backer->dpif);
3127 port_is_lacp_current(const struct ofport *ofport_)
3129 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3130 return (ofport->bundle && ofport->bundle->lacp
3131 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3135 /* Upcall handling. */
3137 struct flow_miss_op {
3138 struct dpif_op dpif_op;
3140 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3141 struct xlate_out xout;
3142 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3144 struct ofpbuf mask; /* Flow mask for "put" ops. */
3145 struct odputil_keybuf maskbuf;
3147 /* If this is a "put" op, then a pointer to the subfacet that should
3148 * be marked as uninstalled if the operation fails. */
3149 struct subfacet *subfacet;
3152 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3153 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3154 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3155 * return value of true). However, for short flows the cost of bookkeeping is
3156 * much higher than the benefits, so when the datapath holds a large number of
3157 * flows we impose some heuristics to decide which flows are likely to be worth
3160 flow_miss_should_make_facet(struct flow_miss *miss)
3162 struct dpif_backer *backer = miss->ofproto->backer;
3165 switch (flow_miss_model) {
3166 case OFPROTO_HANDLE_MISS_AUTO:
3168 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3170 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3174 if (!backer->governor) {
3177 n_subfacets = hmap_count(&backer->subfacets);
3178 if (n_subfacets * 2 <= flow_eviction_threshold) {
3182 backer->governor = governor_create();
3185 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3186 return governor_should_install_flow(backer->governor, hash,
3187 miss->stats.n_packets);
3190 /* Handles 'miss', which matches 'facet'. May add any required datapath
3191 * operations to 'ops', incrementing '*n_ops' for each new op.
3193 * All of the packets in 'miss' are considered to have arrived at time
3194 * 'miss->stats.used'. This is really important only for new facets: if we
3195 * just called time_msec() here, then the new subfacet or its packets could
3196 * look (occasionally) as though it was used some time after the facet was
3197 * used. That can make a one-packet flow look like it has a nonzero duration,
3198 * which looks odd in e.g. NetFlow statistics. */
3200 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3201 struct flow_miss_op *ops, size_t *n_ops)
3203 enum subfacet_path want_path;
3204 struct subfacet *subfacet;
3206 facet->packet_count += miss->stats.n_packets;
3207 facet->prev_packet_count += miss->stats.n_packets;
3208 facet->byte_count += miss->stats.n_bytes;
3209 facet->prev_byte_count += miss->stats.n_bytes;
3211 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3213 /* Don't install the flow if it's the result of the "userspace"
3214 * action for an already installed facet. This can occur when a
3215 * datapath flow with wildcards has a "userspace" action and flows
3216 * sent to userspace result in a different subfacet, which will then
3217 * be rejected as overlapping by the datapath. */
3218 if (miss->upcall_type == DPIF_UC_ACTION
3219 && !list_is_empty(&facet->subfacets)) {
3223 subfacet = subfacet_create(facet, miss);
3224 if (subfacet->path != want_path) {
3225 struct flow_miss_op *op = &ops[(*n_ops)++];
3226 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3228 subfacet->path = want_path;
3230 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3231 if (enable_megaflows) {
3232 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3233 &miss->flow, UINT32_MAX);
3236 op->xout_garbage = false;
3237 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3238 op->subfacet = subfacet;
3239 put->flags = DPIF_FP_CREATE;
3240 put->key = miss->key;
3241 put->key_len = miss->key_len;
3242 put->mask = op->mask.data;
3243 put->mask_len = op->mask.size;
3245 if (want_path == SF_FAST_PATH) {
3246 put->actions = facet->xout.odp_actions.data;
3247 put->actions_len = facet->xout.odp_actions.size;
3249 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3250 op->slow_stub, sizeof op->slow_stub,
3251 &put->actions, &put->actions_len);
3257 /* Handles flow miss 'miss'. May add any required datapath operations
3258 * to 'ops', incrementing '*n_ops' for each new op. */
3260 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3263 struct facet *facet;
3265 miss->ofproto->n_missed += miss->stats.n_packets;
3267 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3269 /* There does not exist a bijection between 'struct flow' and datapath
3270 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3271 * assumption used throughout the facet and subfacet handling code.
3272 * Since we have to handle these misses in userspace anyway, we simply
3273 * skip facet creation, avoiding the problem altogether. */
3274 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3275 || !flow_miss_should_make_facet(miss)) {
3279 facet = facet_create(miss);
3281 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3284 static struct drop_key *
3285 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3288 struct drop_key *drop_key;
3290 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3291 &backer->drop_keys) {
3292 if (drop_key->key_len == key_len
3293 && !memcmp(drop_key->key, key, key_len)) {
3301 drop_key_clear(struct dpif_backer *backer)
3303 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3304 struct drop_key *drop_key, *next;
3306 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3309 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3311 if (error && !VLOG_DROP_WARN(&rl)) {
3312 struct ds ds = DS_EMPTY_INITIALIZER;
3313 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3314 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3315 ovs_strerror(error), ds_cstr(&ds));
3319 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3320 drop_key_destroy(drop_key);
3323 udpif_drop_key_clear(backer->udpif);
3327 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3329 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3330 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3331 struct flow_miss *miss;
3334 /* Process each element in the to-do list, constructing the set of
3335 * operations to batch. */
3337 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3338 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3340 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3342 /* Execute batch. */
3343 for (i = 0; i < n_ops; i++) {
3344 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3346 dpif_operate(backer->dpif, dpif_ops, n_ops);
3348 for (i = 0; i < n_ops; i++) {
3349 if (dpif_ops[i]->error != 0
3350 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3351 && flow_miss_ops[i].subfacet) {
3352 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3354 COVERAGE_INC(subfacet_install_fail);
3356 /* Zero-out subfacet counters when installation failed, but
3357 * datapath reported hits. This should not happen and
3358 * indicates a bug, since if the datapath flow exists, we
3359 * should not be attempting to create a new subfacet. A
3360 * buggy datapath could trigger this, so just zero out the
3361 * counters and log an error. */
3362 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3363 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3364 "datapath reported hits");
3365 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3368 subfacet->path = SF_NOT_INSTALLED;
3374 handle_upcalls(struct dpif_backer *backer)
3376 struct flow_miss_batch *fmb;
3379 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3380 struct drop_key *drop_key = drop_key_next(backer->udpif);
3385 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3386 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3387 hash_bytes(drop_key->key, drop_key->key_len, 0));
3388 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3389 drop_key->key, drop_key->key_len,
3390 NULL, 0, NULL, 0, NULL);
3392 drop_key_destroy(drop_key);
3396 fmb = flow_miss_batch_next(backer->udpif);
3398 handle_flow_misses(backer, fmb);
3399 flow_miss_batch_destroy(fmb);
3403 /* Flow expiration. */
3405 static int subfacet_max_idle(const struct dpif_backer *);
3406 static void update_stats(struct dpif_backer *);
3407 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3408 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3410 /* This function is called periodically by run(). Its job is to collect
3411 * updates for the flows that have been installed into the datapath, most
3412 * importantly when they last were used, and then use that information to
3413 * expire flows that have not been used recently.
3415 * Returns the number of milliseconds after which it should be called again. */
3417 expire(struct dpif_backer *backer)
3419 struct ofproto_dpif *ofproto;
3423 /* Periodically clear out the drop keys in an effort to keep them
3424 * relatively few. */
3425 drop_key_clear(backer);
3427 /* Update stats for each flow in the backer. */
3428 update_stats(backer);
3430 n_subfacets = hmap_count(&backer->subfacets);
3432 struct subfacet *subfacet;
3433 long long int total, now;
3437 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3438 total += now - subfacet->created;
3440 backer->avg_subfacet_life += total / n_subfacets;
3442 backer->avg_subfacet_life /= 2;
3444 backer->avg_n_subfacet += n_subfacets;
3445 backer->avg_n_subfacet /= 2;
3447 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3449 max_idle = subfacet_max_idle(backer);
3450 expire_subfacets(backer, max_idle);
3452 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3453 struct rule *rule, *next_rule;
3455 if (ofproto->backer != backer) {
3459 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3461 ovs_mutex_lock(&ofproto_mutex);
3462 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3463 &ofproto->up.expirable) {
3464 rule_expire(rule_dpif_cast(rule));
3466 ovs_mutex_unlock(&ofproto_mutex);
3468 /* All outstanding data in existing flows has been accounted, so it's a
3469 * good time to do bond rebalancing. */
3470 if (ofproto->has_bonded_bundles) {
3471 struct ofbundle *bundle;
3473 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3475 bond_rebalance(bundle->bond);
3481 return MIN(max_idle, 1000);
3484 /* Updates flow table statistics given that the datapath just reported 'stats'
3485 * as 'subfacet''s statistics. */
3487 update_subfacet_stats(struct subfacet *subfacet,
3488 const struct dpif_flow_stats *stats)
3490 struct facet *facet = subfacet->facet;
3491 struct dpif_flow_stats diff;
3493 diff.tcp_flags = stats->tcp_flags;
3494 diff.used = stats->used;
3496 if (stats->n_packets >= subfacet->dp_packet_count) {
3497 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3499 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3503 if (stats->n_bytes >= subfacet->dp_byte_count) {
3504 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3506 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3510 facet->ofproto->n_hit += diff.n_packets;
3511 subfacet->dp_packet_count = stats->n_packets;
3512 subfacet->dp_byte_count = stats->n_bytes;
3513 subfacet_update_stats(subfacet, &diff);
3515 if (facet->accounted_bytes < facet->byte_count) {
3517 facet_account(facet);
3518 facet->accounted_bytes = facet->byte_count;
3522 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3523 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3525 delete_unexpected_flow(struct dpif_backer *backer,
3526 const struct nlattr *key, size_t key_len)
3528 if (!VLOG_DROP_WARN(&rl)) {
3532 odp_flow_key_format(key, key_len, &s);
3533 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3537 COVERAGE_INC(facet_unexpected);
3538 dpif_flow_del(backer->dpif, key, key_len, NULL);
3541 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3543 * This function also pushes statistics updates to rules which each facet
3544 * resubmits into. Generally these statistics will be accurate. However, if a
3545 * facet changes the rule it resubmits into at some time in between
3546 * update_stats() runs, it is possible that statistics accrued to the
3547 * old rule will be incorrectly attributed to the new rule. This could be
3548 * avoided by calling update_stats() whenever rules are created or
3549 * deleted. However, the performance impact of making so many calls to the
3550 * datapath do not justify the benefit of having perfectly accurate statistics.
3552 * In addition, this function maintains per ofproto flow hit counts. The patch
3553 * port is not treated specially. e.g. A packet ingress from br0 patched into
3554 * br1 will increase the hit count of br0 by 1, however, does not affect
3555 * the hit or miss counts of br1.
3558 update_stats(struct dpif_backer *backer)
3560 const struct dpif_flow_stats *stats;
3561 struct dpif_flow_dump dump;
3562 const struct nlattr *key, *mask;
3563 size_t key_len, mask_len;
3565 dpif_flow_dump_start(&dump, backer->dpif);
3566 while (dpif_flow_dump_next(&dump, &key, &key_len,
3567 &mask, &mask_len, NULL, NULL, &stats)) {
3568 struct subfacet *subfacet;
3571 key_hash = odp_flow_key_hash(key, key_len);
3572 subfacet = subfacet_find(backer, key, key_len, key_hash);
3573 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3575 update_subfacet_stats(subfacet, stats);
3579 /* Stats are updated per-packet. */
3582 case SF_NOT_INSTALLED:
3584 delete_unexpected_flow(backer, key, key_len);
3589 dpif_flow_dump_done(&dump);
3592 /* Calculates and returns the number of milliseconds of idle time after which
3593 * subfacets should expire from the datapath. When a subfacet expires, we fold
3594 * its statistics into its facet, and when a facet's last subfacet expires, we
3595 * fold its statistic into its rule. */
3597 subfacet_max_idle(const struct dpif_backer *backer)
3600 * Idle time histogram.
3602 * Most of the time a switch has a relatively small number of subfacets.
3603 * When this is the case we might as well keep statistics for all of them
3604 * in userspace and to cache them in the kernel datapath for performance as
3607 * As the number of subfacets increases, the memory required to maintain
3608 * statistics about them in userspace and in the kernel becomes
3609 * significant. However, with a large number of subfacets it is likely
3610 * that only a few of them are "heavy hitters" that consume a large amount
3611 * of bandwidth. At this point, only heavy hitters are worth caching in
3612 * the kernel and maintaining in userspaces; other subfacets we can
3615 * The technique used to compute the idle time is to build a histogram with
3616 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3617 * that is installed in the kernel gets dropped in the appropriate bucket.
3618 * After the histogram has been built, we compute the cutoff so that only
3619 * the most-recently-used 1% of subfacets (but at least
3620 * flow_eviction_threshold flows) are kept cached. At least
3621 * the most-recently-used bucket of subfacets is kept, so actually an
3622 * arbitrary number of subfacets can be kept in any given expiration run
3623 * (though the next run will delete most of those unless they receive
3626 * This requires a second pass through the subfacets, in addition to the
3627 * pass made by update_stats(), because the former function never looks at
3628 * uninstallable subfacets.
3630 enum { BUCKET_WIDTH = 100 };
3631 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3632 int buckets[N_BUCKETS] = { 0 };
3633 int total, subtotal, bucket;
3634 struct subfacet *subfacet;
3638 total = hmap_count(&backer->subfacets);
3639 if (total <= flow_eviction_threshold) {
3640 return N_BUCKETS * BUCKET_WIDTH;
3643 /* Build histogram. */
3645 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3646 long long int idle = now - subfacet->used;
3647 int bucket = (idle <= 0 ? 0
3648 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3649 : (unsigned int) idle / BUCKET_WIDTH);
3653 /* Find the first bucket whose flows should be expired. */
3654 subtotal = bucket = 0;
3656 subtotal += buckets[bucket++];
3657 } while (bucket < N_BUCKETS &&
3658 subtotal < MAX(flow_eviction_threshold, total / 100));
3660 if (VLOG_IS_DBG_ENABLED()) {
3664 ds_put_cstr(&s, "keep");
3665 for (i = 0; i < N_BUCKETS; i++) {
3667 ds_put_cstr(&s, ", drop");
3670 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3673 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3677 return bucket * BUCKET_WIDTH;
3681 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3683 /* Cutoff time for most flows. */
3684 long long int normal_cutoff = time_msec() - dp_max_idle;
3686 /* We really want to keep flows for special protocols around, so use a more
3687 * conservative cutoff. */
3688 long long int special_cutoff = time_msec() - 10000;
3690 struct subfacet *subfacet, *next_subfacet;
3691 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3695 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3696 &backer->subfacets) {
3697 long long int cutoff;
3699 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3703 if (subfacet->used < cutoff) {
3704 if (subfacet->path != SF_NOT_INSTALLED) {
3705 batch[n_batch++] = subfacet;
3706 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3707 subfacet_destroy_batch(backer, batch, n_batch);
3711 subfacet_destroy(subfacet);
3717 subfacet_destroy_batch(backer, batch, n_batch);
3721 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3722 * then delete it entirely. */
3724 rule_expire(struct rule_dpif *rule)
3725 OVS_REQUIRES(ofproto_mutex)
3727 uint16_t idle_timeout, hard_timeout;
3728 long long int now = time_msec();
3731 ovs_assert(!rule->up.pending);
3733 /* Has 'rule' expired? */
3734 ovs_mutex_lock(&rule->up.mutex);
3735 hard_timeout = rule->up.hard_timeout;
3736 idle_timeout = rule->up.idle_timeout;
3737 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3738 reason = OFPRR_HARD_TIMEOUT;
3739 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3740 reason = OFPRR_IDLE_TIMEOUT;
3744 ovs_mutex_unlock(&rule->up.mutex);
3747 COVERAGE_INC(ofproto_dpif_expired);
3748 ofproto_rule_expire(&rule->up, reason);
3754 /* Creates and returns a new facet based on 'miss'.
3756 * The caller must already have determined that no facet with an identical
3757 * 'miss->flow' exists in 'miss->ofproto'.
3759 * 'rule' and 'xout' must have been created based on 'miss'.
3761 * 'facet'' statistics are initialized based on 'stats'.
3763 * The facet will initially have no subfacets. The caller should create (at
3764 * least) one subfacet with subfacet_create(). */
3765 static struct facet *
3766 facet_create(const struct flow_miss *miss)
3768 struct ofproto_dpif *ofproto = miss->ofproto;
3769 struct facet *facet;
3772 COVERAGE_INC(facet_create);
3773 facet = xzalloc(sizeof *facet);
3774 facet->ofproto = miss->ofproto;
3775 facet->used = miss->stats.used;
3776 facet->flow = miss->flow;
3777 facet->learn_rl = time_msec() + 500;
3779 list_init(&facet->subfacets);
3780 netflow_flow_init(&facet->nf_flow);
3781 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3783 xlate_out_copy(&facet->xout, &miss->xout);
3785 match_init(&match, &facet->flow, &facet->xout.wc);
3786 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3787 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3788 classifier_insert(&ofproto->facets, &facet->cr);
3789 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3791 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3796 facet_free(struct facet *facet)
3799 xlate_out_uninit(&facet->xout);
3804 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3805 * 'flow' must reflect the data in 'packet'. */
3807 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3808 const struct flow *flow,
3809 struct rule_dpif *rule,
3810 const struct ofpact *ofpacts, size_t ofpacts_len,
3811 struct ofpbuf *packet)
3813 struct odputil_keybuf keybuf;
3814 struct dpif_flow_stats stats;
3815 struct xlate_out xout;
3816 struct xlate_in xin;
3821 ovs_assert((rule != NULL) != (ofpacts != NULL));
3823 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3825 rule_dpif_credit_stats(rule, &stats);
3828 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3829 xin.ofpacts = ofpacts;
3830 xin.ofpacts_len = ofpacts_len;
3831 xin.resubmit_stats = &stats;
3832 xlate_actions(&xin, &xout);
3834 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3835 in_port = flow->in_port.ofp_port;
3836 if (in_port == OFPP_NONE) {
3837 in_port = OFPP_LOCAL;
3839 odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3841 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3842 xout.odp_actions.data, xout.odp_actions.size, packet,
3843 (xout.slow & SLOW_ACTION) != 0);
3844 xlate_out_uninit(&xout);
3849 /* Remove 'facet' from its ofproto and free up the associated memory:
3851 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3852 * rule's statistics, via subfacet_uninstall().
3854 * - Removes 'facet' from its rule and from ofproto->facets.
3857 facet_remove(struct facet *facet)
3859 struct subfacet *subfacet, *next_subfacet;
3861 COVERAGE_INC(facet_remove);
3862 ovs_assert(!list_is_empty(&facet->subfacets));
3864 /* First uninstall all of the subfacets to get final statistics. */
3865 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3866 subfacet_uninstall(subfacet);
3869 /* Flush the final stats to the rule.
3871 * This might require us to have at least one subfacet around so that we
3872 * can use its actions for accounting in facet_account(), which is why we
3873 * have uninstalled but not yet destroyed the subfacets. */
3874 facet_flush_stats(facet);
3876 /* Now we're really all done so destroy everything. */
3877 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3878 &facet->subfacets) {
3879 subfacet_destroy__(subfacet);
3881 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3882 classifier_remove(&facet->ofproto->facets, &facet->cr);
3883 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3884 cls_rule_destroy(&facet->cr);
3888 /* Feed information from 'facet' back into the learning table to keep it in
3889 * sync with what is actually flowing through the datapath. */
3891 facet_learn(struct facet *facet)
3893 long long int now = time_msec();
3895 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3899 facet->learn_rl = now + 500;
3901 if (!facet->xout.has_learn
3902 && !facet->xout.has_normal
3903 && (!facet->xout.has_fin_timeout
3904 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3908 facet_push_stats(facet, true);
3912 facet_account(struct facet *facet)
3914 const struct nlattr *a;
3919 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3922 n_bytes = facet->byte_count - facet->accounted_bytes;
3924 /* This loop feeds byte counters to bond_account() for rebalancing to use
3925 * as a basis. We also need to track the actual VLAN on which the packet
3926 * is going to be sent to ensure that it matches the one passed to
3927 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3930 * We use the actions from an arbitrary subfacet because they should all
3931 * be equally valid for our purpose. */
3932 vlan_tci = facet->flow.vlan_tci;
3933 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3934 facet->xout.odp_actions.size) {
3935 const struct ovs_action_push_vlan *vlan;
3936 struct ofport_dpif *port;
3938 switch (nl_attr_type(a)) {
3939 case OVS_ACTION_ATTR_OUTPUT:
3940 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3941 if (port && port->bundle && port->bundle->bond) {
3942 bond_account(port->bundle->bond, &facet->flow,
3943 vlan_tci_to_vid(vlan_tci), n_bytes);
3947 case OVS_ACTION_ATTR_POP_VLAN:
3948 vlan_tci = htons(0);
3951 case OVS_ACTION_ATTR_PUSH_VLAN:
3952 vlan = nl_attr_get(a);
3953 vlan_tci = vlan->vlan_tci;
3959 /* Returns true if the only action for 'facet' is to send to the controller.
3960 * (We don't report NetFlow expiration messages for such facets because they
3961 * are just part of the control logic for the network, not real traffic). */
3963 facet_is_controller_flow(struct facet *facet)
3966 struct ofproto_dpif *ofproto = facet->ofproto;
3967 const struct ofpact *ofpacts;
3968 struct rule_actions *actions;
3969 struct rule_dpif *rule;
3973 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
3974 actions = rule_dpif_get_actions(rule);
3975 rule_dpif_unref(rule);
3977 ofpacts_len = actions->ofpacts_len;
3978 ofpacts = actions->ofpacts;
3979 is_controller = ofpacts_len > 0
3980 && ofpacts->type == OFPACT_CONTROLLER
3981 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
3982 rule_actions_unref(actions);
3984 return is_controller;
3989 /* Folds all of 'facet''s statistics into its rule. Also updates the
3990 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3991 * 'facet''s statistics in the datapath should have been zeroed and folded into
3992 * its packet and byte counts before this function is called. */
3994 facet_flush_stats(struct facet *facet)
3996 struct ofproto_dpif *ofproto = facet->ofproto;
3997 struct subfacet *subfacet;
3999 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4000 ovs_assert(!subfacet->dp_byte_count);
4001 ovs_assert(!subfacet->dp_packet_count);
4004 facet_push_stats(facet, false);
4005 if (facet->accounted_bytes < facet->byte_count) {
4006 facet_account(facet);
4007 facet->accounted_bytes = facet->byte_count;
4010 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4011 struct ofexpired expired;
4012 expired.flow = facet->flow;
4013 expired.packet_count = facet->packet_count;
4014 expired.byte_count = facet->byte_count;
4015 expired.used = facet->used;
4016 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4019 /* Reset counters to prevent double counting if 'facet' ever gets
4021 facet_reset_counters(facet);
4023 netflow_flow_clear(&facet->nf_flow);
4024 facet->tcp_flags = 0;
4027 /* Searches 'ofproto''s table of facets for one which would be responsible for
4028 * 'flow'. Returns it if found, otherwise a null pointer.
4030 * The returned facet might need revalidation; use facet_lookup_valid()
4031 * instead if that is important. */
4032 static struct facet *
4033 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4035 struct cls_rule *cr;
4037 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4038 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4039 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4040 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4043 /* Searches 'ofproto''s table of facets for one capable that covers
4044 * 'flow'. Returns it if found, otherwise a null pointer.
4046 * The returned facet is guaranteed to be valid. */
4047 static struct facet *
4048 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4050 struct facet *facet;
4052 facet = facet_find(ofproto, flow);
4054 && ofproto->backer->need_revalidate
4055 && !facet_revalidate(facet)) {
4063 facet_check_consistency(struct facet *facet)
4065 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4067 struct xlate_out xout;
4068 struct xlate_in xin;
4071 /* Check the datapath actions for consistency. */
4072 xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4073 xlate_actions(&xin, &xout);
4075 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4076 && facet->xout.slow == xout.slow;
4077 if (!ok && !VLOG_DROP_WARN(&rl)) {
4078 struct ds s = DS_EMPTY_INITIALIZER;
4080 flow_format(&s, &facet->flow);
4081 ds_put_cstr(&s, ": inconsistency in facet");
4083 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4084 ds_put_cstr(&s, " (actions were: ");
4085 format_odp_actions(&s, facet->xout.odp_actions.data,
4086 facet->xout.odp_actions.size);
4087 ds_put_cstr(&s, ") (correct actions: ");
4088 format_odp_actions(&s, xout.odp_actions.data,
4089 xout.odp_actions.size);
4090 ds_put_char(&s, ')');
4093 if (facet->xout.slow != xout.slow) {
4094 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4099 xlate_out_uninit(&xout);
4104 /* Re-searches the classifier for 'facet':
4106 * - If the rule found is different from 'facet''s current rule, moves
4107 * 'facet' to the new rule and recompiles its actions.
4109 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4110 * where it is and recompiles its actions anyway.
4112 * - If any of 'facet''s subfacets correspond to a new flow according to
4113 * xlate_receive(), 'facet' is removed.
4115 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4117 facet_revalidate(struct facet *facet)
4119 struct ofproto_dpif *ofproto = facet->ofproto;
4120 struct rule_dpif *new_rule;
4121 struct subfacet *subfacet;
4122 struct flow_wildcards wc;
4123 struct xlate_out xout;
4124 struct xlate_in xin;
4126 COVERAGE_INC(facet_revalidate);
4128 /* Check that child subfacets still correspond to this facet. Tunnel
4129 * configuration changes could cause a subfacet's OpenFlow in_port to
4131 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4132 struct ofproto_dpif *recv_ofproto;
4133 struct flow recv_flow;
4136 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4137 subfacet->key_len, &recv_flow, NULL,
4138 &recv_ofproto, NULL);
4140 || recv_ofproto != ofproto
4141 || facet != facet_find(ofproto, &recv_flow)) {
4142 facet_remove(facet);
4147 flow_wildcards_init_catchall(&wc);
4148 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4150 /* Calculate new datapath actions.
4152 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4153 * emit a NetFlow expiration and, if so, we need to have the old state
4154 * around to properly compose it. */
4155 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4156 xlate_actions(&xin, &xout);
4157 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4159 /* A facet's slow path reason should only change under dramatic
4160 * circumstances. Rather than try to update everything, it's simpler to
4161 * remove the facet and start over.
4163 * More importantly, if a facet's wildcards change, it will be relatively
4164 * difficult to figure out if its subfacets still belong to it, and if not
4165 * which facet they may belong to. Again, to avoid the complexity, we
4166 * simply give up instead. */
4167 if (facet->xout.slow != xout.slow
4168 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4169 facet_remove(facet);
4170 xlate_out_uninit(&xout);
4171 rule_dpif_unref(new_rule);
4175 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4176 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4177 if (subfacet->path == SF_FAST_PATH) {
4178 struct dpif_flow_stats stats;
4180 subfacet_install(subfacet, &xout.odp_actions, &stats);
4181 subfacet_update_stats(subfacet, &stats);
4185 facet_flush_stats(facet);
4187 ofpbuf_clear(&facet->xout.odp_actions);
4188 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4189 xout.odp_actions.size);
4192 /* Update 'facet' now that we've taken care of all the old state. */
4193 facet->xout.slow = xout.slow;
4194 facet->xout.has_learn = xout.has_learn;
4195 facet->xout.has_normal = xout.has_normal;
4196 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4197 facet->xout.nf_output_iface = xout.nf_output_iface;
4198 facet->xout.mirrors = xout.mirrors;
4199 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4201 ovs_mutex_lock(&new_rule->up.mutex);
4202 facet->used = MAX(facet->used, new_rule->up.created);
4203 ovs_mutex_unlock(&new_rule->up.mutex);
4205 xlate_out_uninit(&xout);
4206 rule_dpif_unref(new_rule);
4211 facet_reset_counters(struct facet *facet)
4213 facet->packet_count = 0;
4214 facet->byte_count = 0;
4215 facet->prev_packet_count = 0;
4216 facet->prev_byte_count = 0;
4217 facet->accounted_bytes = 0;
4221 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4222 struct dpif_flow_stats *stats, bool may_learn)
4224 struct ofport_dpif *in_port;
4225 struct xlate_in xin;
4227 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4228 if (in_port && in_port->is_tunnel) {
4229 netdev_vport_inc_rx(in_port->up.netdev, stats);
4232 xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4233 xin.resubmit_stats = stats;
4234 xin.may_learn = may_learn;
4235 xlate_actions_for_side_effects(&xin);
4239 facet_push_stats(struct facet *facet, bool may_learn)
4241 struct dpif_flow_stats stats;
4243 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4244 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4245 ovs_assert(facet->used >= facet->prev_used);
4247 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4248 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4249 stats.used = facet->used;
4250 stats.tcp_flags = facet->tcp_flags;
4252 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4253 facet->prev_packet_count = facet->packet_count;
4254 facet->prev_byte_count = facet->byte_count;
4255 facet->prev_used = facet->used;
4257 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4259 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4260 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4261 stats.n_packets, stats.n_bytes);
4262 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4267 push_all_stats__(bool run_fast)
4269 static long long int rl = LLONG_MIN;
4270 struct ofproto_dpif *ofproto;
4272 if (time_msec() < rl) {
4276 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4277 struct cls_cursor cursor;
4278 struct facet *facet;
4280 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4281 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4282 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4283 facet_push_stats(facet, false);
4288 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4291 rl = time_msec() + 100;
4295 push_all_stats(void)
4297 push_all_stats__(true);
4301 rule_dpif_credit_stats(struct rule_dpif *rule,
4302 const struct dpif_flow_stats *stats)
4304 ovs_mutex_lock(&rule->stats_mutex);
4305 rule->packet_count += stats->n_packets;
4306 rule->byte_count += stats->n_bytes;
4307 rule->up.used = MAX(rule->up.used, stats->used);
4308 ovs_mutex_unlock(&rule->stats_mutex);
4312 rule_dpif_fail_open(const struct rule_dpif *rule)
4314 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4318 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4319 OVS_REQUIRES(rule->up.mutex)
4321 return rule->up.flow_cookie;
4325 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4326 uint16_t hard_timeout)
4328 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4331 /* Returns 'rule''s actions. The caller owns a reference on the returned
4332 * actions and must eventually release it (with rule_actions_unref()) to avoid
4334 struct rule_actions *
4335 rule_dpif_get_actions(const struct rule_dpif *rule)
4337 return rule_get_actions(&rule->up);
4342 static struct subfacet *
4343 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4344 size_t key_len, uint32_t key_hash)
4346 struct subfacet *subfacet;
4348 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4349 &backer->subfacets) {
4350 if (subfacet->key_len == key_len
4351 && !memcmp(key, subfacet->key, key_len)) {
4359 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4360 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4361 * existing subfacet if there is one, otherwise creates and returns a
4363 static struct subfacet *
4364 subfacet_create(struct facet *facet, struct flow_miss *miss)
4366 struct dpif_backer *backer = miss->ofproto->backer;
4367 const struct nlattr *key = miss->key;
4368 size_t key_len = miss->key_len;
4370 struct subfacet *subfacet;
4372 key_hash = odp_flow_key_hash(key, key_len);
4374 if (list_is_empty(&facet->subfacets)) {
4375 subfacet = &facet->one_subfacet;
4377 subfacet = subfacet_find(backer, key, key_len, key_hash);
4379 if (subfacet->facet == facet) {
4383 /* This shouldn't happen. */
4384 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4385 subfacet_destroy(subfacet);
4388 subfacet = xmalloc(sizeof *subfacet);
4391 COVERAGE_INC(subfacet_create);
4392 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4393 list_push_back(&facet->subfacets, &subfacet->list_node);
4394 subfacet->facet = facet;
4395 subfacet->key = xmemdup(key, key_len);
4396 subfacet->key_len = key_len;
4397 subfacet->used = miss->stats.used;
4398 subfacet->created = subfacet->used;
4399 subfacet->dp_packet_count = 0;
4400 subfacet->dp_byte_count = 0;
4401 subfacet->path = SF_NOT_INSTALLED;
4402 subfacet->backer = backer;
4407 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4408 * its facet within 'ofproto', and frees it. */
4410 subfacet_destroy__(struct subfacet *subfacet)
4412 struct facet *facet = subfacet->facet;
4414 COVERAGE_INC(subfacet_destroy);
4415 subfacet_uninstall(subfacet);
4416 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4417 list_remove(&subfacet->list_node);
4418 free(subfacet->key);
4419 if (subfacet != &facet->one_subfacet) {
4424 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4425 * last remaining subfacet in its facet destroys the facet too. */
4427 subfacet_destroy(struct subfacet *subfacet)
4429 struct facet *facet = subfacet->facet;
4431 if (list_is_singleton(&facet->subfacets)) {
4432 /* facet_remove() needs at least one subfacet (it will remove it). */
4433 facet_remove(facet);
4435 subfacet_destroy__(subfacet);
4440 subfacet_destroy_batch(struct dpif_backer *backer,
4441 struct subfacet **subfacets, int n)
4443 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4444 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4445 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4448 for (i = 0; i < n; i++) {
4449 ops[i].type = DPIF_OP_FLOW_DEL;
4450 ops[i].u.flow_del.key = subfacets[i]->key;
4451 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4452 ops[i].u.flow_del.stats = &stats[i];
4456 dpif_operate(backer->dpif, opsp, n);
4457 for (i = 0; i < n; i++) {
4458 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4459 subfacets[i]->path = SF_NOT_INSTALLED;
4460 subfacet_destroy(subfacets[i]);
4465 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4466 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4467 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4468 * since 'subfacet' was last updated.
4470 * Returns 0 if successful, otherwise a positive errno value. */
4472 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4473 struct dpif_flow_stats *stats)
4475 struct facet *facet = subfacet->facet;
4476 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4477 const struct nlattr *actions = odp_actions->data;
4478 size_t actions_len = odp_actions->size;
4479 struct odputil_keybuf maskbuf;
4482 uint64_t slow_path_stub[128 / 8];
4483 enum dpif_flow_put_flags flags;
4486 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4489 flags |= DPIF_FP_ZERO_STATS;
4492 if (path == SF_SLOW_PATH) {
4493 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4494 slow_path_stub, sizeof slow_path_stub,
4495 &actions, &actions_len);
4498 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4499 if (enable_megaflows) {
4500 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4501 &facet->flow, UINT32_MAX);
4504 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4505 subfacet->key_len, mask.data, mask.size,
4506 actions, actions_len, stats);
4509 subfacet_reset_dp_stats(subfacet, stats);
4513 COVERAGE_INC(subfacet_install_fail);
4515 subfacet->path = path;
4520 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4522 subfacet_uninstall(struct subfacet *subfacet)
4524 if (subfacet->path != SF_NOT_INSTALLED) {
4525 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4526 struct dpif_flow_stats stats;
4529 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4530 subfacet->key_len, &stats);
4531 subfacet_reset_dp_stats(subfacet, &stats);
4533 subfacet_update_stats(subfacet, &stats);
4535 subfacet->path = SF_NOT_INSTALLED;
4537 ovs_assert(subfacet->dp_packet_count == 0);
4538 ovs_assert(subfacet->dp_byte_count == 0);
4542 /* Resets 'subfacet''s datapath statistics counters. This should be called
4543 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4544 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4545 * was reset in the datapath. 'stats' will be modified to include only
4546 * statistics new since 'subfacet' was last updated. */
4548 subfacet_reset_dp_stats(struct subfacet *subfacet,
4549 struct dpif_flow_stats *stats)
4552 && subfacet->dp_packet_count <= stats->n_packets
4553 && subfacet->dp_byte_count <= stats->n_bytes) {
4554 stats->n_packets -= subfacet->dp_packet_count;
4555 stats->n_bytes -= subfacet->dp_byte_count;
4558 subfacet->dp_packet_count = 0;
4559 subfacet->dp_byte_count = 0;
4562 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4564 * Because of the meaning of a subfacet's counters, it only makes sense to do
4565 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4566 * represents a packet that was sent by hand or if it represents statistics
4567 * that have been cleared out of the datapath. */
4569 subfacet_update_stats(struct subfacet *subfacet,
4570 const struct dpif_flow_stats *stats)
4572 if (stats->n_packets || stats->used > subfacet->used) {
4573 struct facet *facet = subfacet->facet;
4575 subfacet->used = MAX(subfacet->used, stats->used);
4576 facet->used = MAX(facet->used, stats->used);
4577 facet->packet_count += stats->n_packets;
4578 facet->byte_count += stats->n_bytes;
4579 facet->tcp_flags |= stats->tcp_flags;
4585 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4586 * the fields that were relevant as part of the lookup. */
4588 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4589 struct flow_wildcards *wc, struct rule_dpif **rule)
4591 struct ofport_dpif *port;
4593 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4596 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4598 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4599 flow->in_port.ofp_port);
4602 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4603 ofproto->no_packet_in_rule, rule);
4607 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4608 const struct flow *flow, struct flow_wildcards *wc,
4609 uint8_t table_id, struct rule_dpif **rule)
4611 const struct cls_rule *cls_rule;
4612 struct classifier *cls;
4616 if (table_id >= N_TABLES) {
4621 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4622 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4625 cls = &ofproto->up.tables[table_id].cls;
4626 ovs_rwlock_rdlock(&cls->rwlock);
4627 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4628 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4629 /* We must pretend that transport ports are unavailable. */
4630 struct flow ofpc_normal_flow = *flow;
4631 ofpc_normal_flow.tp_src = htons(0);
4632 ofpc_normal_flow.tp_dst = htons(0);
4633 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4634 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4635 cls_rule = &ofproto->drop_frags_rule->up.cr;
4637 flow_wildcards_init_exact(wc);
4640 cls_rule = classifier_lookup(cls, flow, wc);
4643 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4644 rule_dpif_ref(*rule);
4645 ovs_rwlock_unlock(&cls->rwlock);
4647 return *rule != NULL;
4650 /* Given a port configuration (specified as zero if there's no port), chooses
4651 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4652 * flow table miss. */
4654 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4655 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4657 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4658 rule_dpif_ref(*rule);
4662 rule_dpif_ref(struct rule_dpif *rule)
4665 ofproto_rule_ref(&rule->up);
4670 rule_dpif_unref(struct rule_dpif *rule)
4673 ofproto_rule_unref(&rule->up);
4678 complete_operation(struct rule_dpif *rule)
4679 OVS_REQUIRES(ofproto_mutex)
4681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4683 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4684 ofoperation_complete(rule->up.pending, 0);
4687 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4689 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4692 static struct rule *
4695 struct rule_dpif *rule = xmalloc(sizeof *rule);
4700 rule_dealloc(struct rule *rule_)
4702 struct rule_dpif *rule = rule_dpif_cast(rule_);
4707 rule_construct(struct rule *rule_)
4709 struct rule_dpif *rule = rule_dpif_cast(rule_);
4710 ovs_mutex_init(&rule->stats_mutex);
4711 ovs_mutex_lock(&rule->stats_mutex);
4712 rule->packet_count = 0;
4713 rule->byte_count = 0;
4714 ovs_mutex_unlock(&rule->stats_mutex);
4719 rule_insert(struct rule *rule_)
4720 OVS_REQUIRES(ofproto_mutex)
4722 struct rule_dpif *rule = rule_dpif_cast(rule_);
4723 complete_operation(rule);
4727 rule_delete(struct rule *rule_)
4728 OVS_REQUIRES(ofproto_mutex)
4730 struct rule_dpif *rule = rule_dpif_cast(rule_);
4731 complete_operation(rule);
4735 rule_destruct(struct rule *rule_)
4737 struct rule_dpif *rule = rule_dpif_cast(rule_);
4738 ovs_mutex_destroy(&rule->stats_mutex);
4742 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4744 struct rule_dpif *rule = rule_dpif_cast(rule_);
4746 /* push_all_stats() can handle flow misses which, when using the learn
4747 * action, can cause rules to be added and deleted. This can corrupt our
4748 * caller's datastructures which assume that rule_get_stats() doesn't have
4749 * an impact on the flow table. To be safe, we disable miss handling. */
4750 push_all_stats__(false);
4752 /* Start from historical data for 'rule' itself that are no longer tracked
4753 * in facets. This counts, for example, facets that have expired. */
4754 ovs_mutex_lock(&rule->stats_mutex);
4755 *packets = rule->packet_count;
4756 *bytes = rule->byte_count;
4757 ovs_mutex_unlock(&rule->stats_mutex);
4761 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4762 struct ofpbuf *packet)
4764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4766 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4770 rule_execute(struct rule *rule, const struct flow *flow,
4771 struct ofpbuf *packet)
4773 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4774 ofpbuf_delete(packet);
4779 rule_modify_actions(struct rule *rule_, bool reset_counters)
4780 OVS_REQUIRES(ofproto_mutex)
4782 struct rule_dpif *rule = rule_dpif_cast(rule_);
4784 if (reset_counters) {
4785 ovs_mutex_lock(&rule->stats_mutex);
4786 rule->packet_count = 0;
4787 rule->byte_count = 0;
4788 ovs_mutex_unlock(&rule->stats_mutex);
4791 complete_operation(rule);
4794 /* Sends 'packet' out 'ofport'.
4795 * May modify 'packet'.
4796 * Returns 0 if successful, otherwise a positive errno value. */
4798 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4803 error = xlate_send_packet(ofport, packet);
4805 ovs_mutex_lock(&ofproto->stats_mutex);
4806 ofproto->stats.tx_packets++;
4807 ofproto->stats.tx_bytes += packet->size;
4808 ovs_mutex_unlock(&ofproto->stats_mutex);
4812 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4813 * The action will state 'slow' as the reason that the action is in the slow
4814 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4815 * dump-flows" output to see why a flow is in the slow path.)
4817 * The 'stub_size' bytes in 'stub' will be used to store the action.
4818 * 'stub_size' must be large enough for the action.
4820 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4823 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4824 enum slow_path_reason slow,
4825 uint64_t *stub, size_t stub_size,
4826 const struct nlattr **actionsp, size_t *actions_lenp)
4828 union user_action_cookie cookie;
4831 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4832 cookie.slow_path.unused = 0;
4833 cookie.slow_path.reason = slow;
4835 ofpbuf_use_stack(&buf, stub, stub_size);
4836 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4837 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4839 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4841 odp_port_t odp_port;
4844 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4845 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4846 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4848 *actionsp = buf.data;
4849 *actions_lenp = buf.size;
4853 set_frag_handling(struct ofproto *ofproto_,
4854 enum ofp_config_flags frag_handling)
4856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4857 if (frag_handling != OFPC_FRAG_REASM) {
4858 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4866 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4867 const struct flow *flow,
4868 const struct ofpact *ofpacts, size_t ofpacts_len)
4870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4872 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4873 ofpacts_len, packet);
4880 set_netflow(struct ofproto *ofproto_,
4881 const struct netflow_options *netflow_options)
4883 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4885 if (netflow_options) {
4886 if (!ofproto->netflow) {
4887 ofproto->netflow = netflow_create();
4888 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4890 return netflow_set_options(ofproto->netflow, netflow_options);
4891 } else if (ofproto->netflow) {
4892 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4893 netflow_destroy(ofproto->netflow);
4894 ofproto->netflow = NULL;
4901 get_netflow_ids(const struct ofproto *ofproto_,
4902 uint8_t *engine_type, uint8_t *engine_id)
4904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4906 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4910 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
4912 if (!facet_is_controller_flow(facet) &&
4913 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4914 struct subfacet *subfacet;
4915 struct ofexpired expired;
4917 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4918 if (subfacet->path == SF_FAST_PATH) {
4919 struct dpif_flow_stats stats;
4921 subfacet_install(subfacet, &facet->xout.odp_actions,
4923 subfacet_update_stats(subfacet, &stats);
4927 expired.flow = facet->flow;
4928 expired.packet_count = facet->packet_count;
4929 expired.byte_count = facet->byte_count;
4930 expired.used = facet->used;
4931 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4936 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
4938 struct cls_cursor cursor;
4939 struct facet *facet;
4941 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4942 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4943 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4944 send_active_timeout(ofproto, facet);
4946 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4949 static struct ofproto_dpif *
4950 ofproto_dpif_lookup(const char *name)
4952 struct ofproto_dpif *ofproto;
4954 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4955 hash_string(name, 0), &all_ofproto_dpifs) {
4956 if (!strcmp(ofproto->up.name, name)) {
4964 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4965 const char *argv[], void *aux OVS_UNUSED)
4967 struct ofproto_dpif *ofproto;
4970 ofproto = ofproto_dpif_lookup(argv[1]);
4972 unixctl_command_reply_error(conn, "no such bridge");
4975 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4976 mac_learning_flush(ofproto->ml);
4977 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4979 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4980 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4981 mac_learning_flush(ofproto->ml);
4982 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4986 unixctl_command_reply(conn, "table successfully flushed");
4989 static struct ofport_dpif *
4990 ofbundle_get_a_port(const struct ofbundle *bundle)
4992 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4997 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4998 const char *argv[], void *aux OVS_UNUSED)
5000 struct ds ds = DS_EMPTY_INITIALIZER;
5001 const struct ofproto_dpif *ofproto;
5002 const struct mac_entry *e;
5004 ofproto = ofproto_dpif_lookup(argv[1]);
5006 unixctl_command_reply_error(conn, "no such bridge");
5010 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5011 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5012 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5013 struct ofbundle *bundle = e->port.p;
5014 char name[OFP_MAX_PORT_NAME_LEN];
5016 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5018 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5019 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5020 mac_entry_age(ofproto->ml, e));
5022 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5023 unixctl_command_reply(conn, ds_cstr(&ds));
5028 struct xlate_out xout;
5029 struct xlate_in xin;
5035 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5037 struct rule_actions *actions;
5040 ds_put_char_multiple(result, '\t', level);
5042 ds_put_cstr(result, "No match\n");
5046 ovs_mutex_lock(&rule->up.mutex);
5047 cookie = rule->up.flow_cookie;
5048 ovs_mutex_unlock(&rule->up.mutex);
5050 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5051 rule ? rule->up.table_id : 0, ntohll(cookie));
5052 cls_rule_format(&rule->up.cr, result);
5053 ds_put_char(result, '\n');
5055 actions = rule_dpif_get_actions(rule);
5057 ds_put_char_multiple(result, '\t', level);
5058 ds_put_cstr(result, "OpenFlow ");
5059 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5060 ds_put_char(result, '\n');
5062 rule_actions_unref(actions);
5066 trace_format_flow(struct ds *result, int level, const char *title,
5067 struct trace_ctx *trace)
5069 ds_put_char_multiple(result, '\t', level);
5070 ds_put_format(result, "%s: ", title);
5071 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5072 ds_put_cstr(result, "unchanged");
5074 flow_format(result, &trace->xin.flow);
5075 trace->flow = trace->xin.flow;
5077 ds_put_char(result, '\n');
5081 trace_format_regs(struct ds *result, int level, const char *title,
5082 struct trace_ctx *trace)
5086 ds_put_char_multiple(result, '\t', level);
5087 ds_put_format(result, "%s:", title);
5088 for (i = 0; i < FLOW_N_REGS; i++) {
5089 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5091 ds_put_char(result, '\n');
5095 trace_format_odp(struct ds *result, int level, const char *title,
5096 struct trace_ctx *trace)
5098 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5100 ds_put_char_multiple(result, '\t', level);
5101 ds_put_format(result, "%s: ", title);
5102 format_odp_actions(result, odp_actions->data, odp_actions->size);
5103 ds_put_char(result, '\n');
5107 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5109 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5110 struct ds *result = trace->result;
5112 ds_put_char(result, '\n');
5113 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5114 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5115 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5116 trace_format_rule(result, recurse + 1, rule);
5120 trace_report(struct xlate_in *xin, const char *s, int recurse)
5122 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5123 struct ds *result = trace->result;
5125 ds_put_char_multiple(result, '\t', recurse);
5126 ds_put_cstr(result, s);
5127 ds_put_char(result, '\n');
5131 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5132 void *aux OVS_UNUSED)
5134 const struct dpif_backer *backer = NULL;
5135 struct ofproto_dpif *ofproto;
5136 struct ofpbuf odp_key, odp_mask;
5137 struct ofpbuf *packet;
5140 struct simap port_names;
5146 ofpbuf_init(&odp_key, 0);
5147 ofpbuf_init(&odp_mask, 0);
5148 simap_init(&port_names);
5150 /* Handle "-generate" or a hex string as the last argument. */
5151 if (!strcmp(argv[argc - 1], "-generate")) {
5152 packet = ofpbuf_new(0);
5155 const char *error = eth_from_hex(argv[argc - 1], &packet);
5158 } else if (argc == 4) {
5159 /* The 3-argument form must end in "-generate' or a hex string. */
5160 unixctl_command_reply_error(conn, error);
5165 /* odp_flow can have its in_port specified as a name instead of port no.
5166 * We do not yet know whether a given flow is a odp_flow or a br_flow.
5167 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5168 * we need to create a simap of name to port no. */
5170 const char *dp_type;
5171 if (!strncmp(argv[1], "ovs-", 4)) {
5172 dp_type = argv[1] + 4;
5176 backer = shash_find_data(&all_dpif_backers, dp_type);
5178 struct shash_node *node;
5179 if (shash_count(&all_dpif_backers) == 1) {
5180 node = shash_first(&all_dpif_backers);
5181 backer = node->data;
5184 if (backer && backer->dpif) {
5185 struct dpif_port dpif_port;
5186 struct dpif_port_dump port_dump;
5187 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5188 simap_put(&port_names, dpif_port.name,
5189 odp_to_u32(dpif_port.port_no));
5193 /* Parse the flow and determine whether a datapath or
5194 * bridge is specified. If function odp_flow_key_from_string()
5195 * returns 0, the flow is a odp_flow. If function
5196 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5197 if (!odp_flow_from_string(argv[argc - 1], &port_names, &odp_key, &odp_mask)) {
5199 unixctl_command_reply_error(conn, "Cannot find the datapath");
5203 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5204 NULL, &ofproto, NULL)) {
5205 unixctl_command_reply_error(conn, "Invalid datapath flow");
5208 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5209 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5211 unixctl_command_reply_error(conn, "Must specify bridge name");
5215 ofproto = ofproto_dpif_lookup(argv[1]);
5217 unixctl_command_reply_error(conn, "Unknown bridge name");
5221 unixctl_command_reply_error(conn, "Bad flow syntax");
5225 /* Generate a packet, if requested. */
5227 if (!packet->size) {
5228 flow_compose(packet, &flow);
5230 union flow_in_port in_port_;
5232 in_port_ = flow.in_port;
5233 ds_put_cstr(&result, "Packet: ");
5234 s = ofp_packet_to_string(packet->data, packet->size);
5235 ds_put_cstr(&result, s);
5238 /* Use the metadata from the flow and the packet argument
5239 * to reconstruct the flow. */
5240 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5245 ofproto_trace(ofproto, &flow, packet, &result);
5246 unixctl_command_reply(conn, ds_cstr(&result));
5249 ds_destroy(&result);
5250 ofpbuf_delete(packet);
5251 ofpbuf_uninit(&odp_key);
5252 ofpbuf_uninit(&odp_mask);
5253 simap_destroy(&port_names);
5257 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5258 const struct ofpbuf *packet, struct ds *ds)
5260 struct rule_dpif *rule;
5261 struct flow_wildcards wc;
5263 ds_put_cstr(ds, "Flow: ");
5264 flow_format(ds, flow);
5265 ds_put_char(ds, '\n');
5267 flow_wildcards_init_catchall(&wc);
5268 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5270 trace_format_rule(ds, 0, rule);
5271 if (rule == ofproto->miss_rule) {
5272 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5273 } else if (rule == ofproto->no_packet_in_rule) {
5274 ds_put_cstr(ds, "\nNo match, packets dropped because "
5275 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5276 } else if (rule == ofproto->drop_frags_rule) {
5277 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5278 "and the fragment handling mode is \"drop\".\n");
5282 uint64_t odp_actions_stub[1024 / 8];
5283 struct ofpbuf odp_actions;
5284 struct trace_ctx trace;
5288 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5291 ofpbuf_use_stub(&odp_actions,
5292 odp_actions_stub, sizeof odp_actions_stub);
5293 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5294 trace.xin.resubmit_hook = trace_resubmit;
5295 trace.xin.report_hook = trace_report;
5297 xlate_actions(&trace.xin, &trace.xout);
5298 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5300 ds_put_char(ds, '\n');
5301 trace_format_flow(ds, 0, "Final flow", &trace);
5303 match_init(&match, flow, &trace.xout.wc);
5304 ds_put_cstr(ds, "Relevant fields: ");
5305 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5306 ds_put_char(ds, '\n');
5308 ds_put_cstr(ds, "Datapath actions: ");
5309 format_odp_actions(ds, trace.xout.odp_actions.data,
5310 trace.xout.odp_actions.size);
5312 if (trace.xout.slow) {
5313 enum slow_path_reason slow;
5315 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5316 "slow path because it:");
5318 slow = trace.xout.slow;
5320 enum slow_path_reason bit = rightmost_1bit(slow);
5322 ds_put_format(ds, "\n\t- %s.",
5323 slow_path_reason_to_explanation(bit));
5329 xlate_out_uninit(&trace.xout);
5332 rule_dpif_unref(rule);
5335 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5336 * 'reply' describing the results. */
5338 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5340 struct cls_cursor cursor;
5341 struct facet *facet;
5345 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5346 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5347 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5348 if (!facet_check_consistency(facet)) {
5352 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5354 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5358 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5359 ofproto->up.name, errors);
5361 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5366 ofproto_dpif_self_check(struct unixctl_conn *conn,
5367 int argc, const char *argv[], void *aux OVS_UNUSED)
5369 struct ds reply = DS_EMPTY_INITIALIZER;
5370 struct ofproto_dpif *ofproto;
5373 ofproto = ofproto_dpif_lookup(argv[1]);
5375 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5376 "ofproto/list for help)");
5379 ofproto_dpif_self_check__(ofproto, &reply);
5381 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5382 ofproto_dpif_self_check__(ofproto, &reply);
5386 unixctl_command_reply(conn, ds_cstr(&reply));
5390 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5391 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5392 * to destroy 'ofproto_shash' and free the returned value. */
5393 static const struct shash_node **
5394 get_ofprotos(struct shash *ofproto_shash)
5396 const struct ofproto_dpif *ofproto;
5398 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5399 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5400 shash_add_nocopy(ofproto_shash, name, ofproto);
5403 return shash_sort(ofproto_shash);
5407 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5408 const char *argv[] OVS_UNUSED,
5409 void *aux OVS_UNUSED)
5411 struct ds ds = DS_EMPTY_INITIALIZER;
5412 struct shash ofproto_shash;
5413 const struct shash_node **sorted_ofprotos;
5416 shash_init(&ofproto_shash);
5417 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5418 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5419 const struct shash_node *node = sorted_ofprotos[i];
5420 ds_put_format(&ds, "%s\n", node->name);
5423 shash_destroy(&ofproto_shash);
5424 free(sorted_ofprotos);
5426 unixctl_command_reply(conn, ds_cstr(&ds));
5431 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5433 const struct shash_node **ofprotos;
5434 struct ofproto_dpif *ofproto;
5435 struct shash ofproto_shash;
5436 uint64_t n_hit, n_missed;
5439 n_hit = n_missed = 0;
5440 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5441 if (ofproto->backer == backer) {
5442 n_missed += ofproto->n_missed;
5443 n_hit += ofproto->n_hit;
5447 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5448 dpif_name(backer->dpif), n_hit, n_missed);
5449 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5450 " life span: %lldms\n", hmap_count(&backer->subfacets),
5451 backer->avg_n_subfacet, backer->max_n_subfacet,
5452 backer->avg_subfacet_life);
5454 shash_init(&ofproto_shash);
5455 ofprotos = get_ofprotos(&ofproto_shash);
5456 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5457 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5458 const struct shash_node **ports;
5461 if (ofproto->backer != backer) {
5465 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5466 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5468 ports = shash_sort(&ofproto->up.port_by_name);
5469 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5470 const struct shash_node *node = ports[j];
5471 struct ofport *ofport = node->data;
5473 odp_port_t odp_port;
5475 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5478 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5479 if (odp_port != ODPP_NONE) {
5480 ds_put_format(ds, "%"PRIu32":", odp_port);
5482 ds_put_cstr(ds, "none:");
5485 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5488 if (!netdev_get_config(ofport->netdev, &config)) {
5489 const struct smap_node **nodes;
5492 nodes = smap_sort(&config);
5493 for (i = 0; i < smap_count(&config); i++) {
5494 const struct smap_node *node = nodes[i];
5495 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5496 node->key, node->value);
5500 smap_destroy(&config);
5502 ds_put_char(ds, ')');
5503 ds_put_char(ds, '\n');
5507 shash_destroy(&ofproto_shash);
5512 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5513 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5515 struct ds ds = DS_EMPTY_INITIALIZER;
5516 const struct shash_node **backers;
5519 backers = shash_sort(&all_dpif_backers);
5520 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5521 dpif_show_backer(backers[i]->data, &ds);
5525 unixctl_command_reply(conn, ds_cstr(&ds));
5529 /* Dump the megaflow (facet) cache. This is useful to check the
5530 * correctness of flow wildcarding, since the same mechanism is used for
5531 * both xlate caching and kernel wildcarding.
5533 * It's important to note that in the output the flow description uses
5534 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5536 * This command is only needed for advanced debugging, so it's not
5537 * documented in the man page. */
5539 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5540 int argc OVS_UNUSED, const char *argv[],
5541 void *aux OVS_UNUSED)
5543 struct ds ds = DS_EMPTY_INITIALIZER;
5544 const struct ofproto_dpif *ofproto;
5545 long long int now = time_msec();
5546 struct cls_cursor cursor;
5547 struct facet *facet;
5549 ofproto = ofproto_dpif_lookup(argv[1]);
5551 unixctl_command_reply_error(conn, "no such bridge");
5555 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5556 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5557 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5558 cls_rule_format(&facet->cr, &ds);
5559 ds_put_cstr(&ds, ", ");
5560 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5561 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5562 ds_put_cstr(&ds, "Datapath actions: ");
5563 if (facet->xout.slow) {
5564 uint64_t slow_path_stub[128 / 8];
5565 const struct nlattr *actions;
5568 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5569 slow_path_stub, sizeof slow_path_stub,
5570 &actions, &actions_len);
5571 format_odp_actions(&ds, actions, actions_len);
5573 format_odp_actions(&ds, facet->xout.odp_actions.data,
5574 facet->xout.odp_actions.size);
5576 ds_put_cstr(&ds, "\n");
5578 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5580 ds_chomp(&ds, '\n');
5581 unixctl_command_reply(conn, ds_cstr(&ds));
5585 /* Disable using the megaflows.
5587 * This command is only needed for advanced debugging, so it's not
5588 * documented in the man page. */
5590 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5591 int argc OVS_UNUSED,
5592 const char *argv[] OVS_UNUSED,
5593 void *aux OVS_UNUSED)
5595 struct ofproto_dpif *ofproto;
5597 enable_megaflows = false;
5599 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5600 flush(&ofproto->up);
5603 unixctl_command_reply(conn, "megaflows disabled");
5606 /* Re-enable using megaflows.
5608 * This command is only needed for advanced debugging, so it's not
5609 * documented in the man page. */
5611 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5612 int argc OVS_UNUSED,
5613 const char *argv[] OVS_UNUSED,
5614 void *aux OVS_UNUSED)
5616 struct ofproto_dpif *ofproto;
5618 enable_megaflows = true;
5620 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5621 flush(&ofproto->up);
5624 unixctl_command_reply(conn, "megaflows enabled");
5628 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5629 int argc OVS_UNUSED, const char *argv[],
5630 void *aux OVS_UNUSED)
5632 struct ds ds = DS_EMPTY_INITIALIZER;
5633 const struct ofproto_dpif *ofproto;
5634 struct subfacet *subfacet;
5636 ofproto = ofproto_dpif_lookup(argv[1]);
5638 unixctl_command_reply_error(conn, "no such bridge");
5642 update_stats(ofproto->backer);
5644 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5645 struct facet *facet = subfacet->facet;
5646 struct odputil_keybuf maskbuf;
5649 if (facet->ofproto != ofproto) {
5653 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5654 if (enable_megaflows) {
5655 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5656 &facet->flow, UINT32_MAX);
5659 odp_flow_format(subfacet->key, subfacet->key_len,
5660 mask.data, mask.size, NULL, &ds, false);
5662 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5663 subfacet->dp_packet_count, subfacet->dp_byte_count);
5664 if (subfacet->used) {
5665 ds_put_format(&ds, "%.3fs",
5666 (time_msec() - subfacet->used) / 1000.0);
5668 ds_put_format(&ds, "never");
5670 if (subfacet->facet->tcp_flags) {
5671 ds_put_cstr(&ds, ", flags:");
5672 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5675 ds_put_cstr(&ds, ", actions:");
5676 if (facet->xout.slow) {
5677 uint64_t slow_path_stub[128 / 8];
5678 const struct nlattr *actions;
5681 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5682 slow_path_stub, sizeof slow_path_stub,
5683 &actions, &actions_len);
5684 format_odp_actions(&ds, actions, actions_len);
5686 format_odp_actions(&ds, facet->xout.odp_actions.data,
5687 facet->xout.odp_actions.size);
5689 ds_put_char(&ds, '\n');
5692 unixctl_command_reply(conn, ds_cstr(&ds));
5697 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5698 int argc OVS_UNUSED, const char *argv[],
5699 void *aux OVS_UNUSED)
5701 struct ds ds = DS_EMPTY_INITIALIZER;
5702 struct ofproto_dpif *ofproto;
5704 ofproto = ofproto_dpif_lookup(argv[1]);
5706 unixctl_command_reply_error(conn, "no such bridge");
5710 flush(&ofproto->up);
5712 unixctl_command_reply(conn, ds_cstr(&ds));
5717 ofproto_dpif_unixctl_init(void)
5719 static bool registered;
5725 unixctl_command_register(
5727 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5728 1, 3, ofproto_unixctl_trace, NULL);
5729 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5730 ofproto_unixctl_fdb_flush, NULL);
5731 unixctl_command_register("fdb/show", "bridge", 1, 1,
5732 ofproto_unixctl_fdb_show, NULL);
5733 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5734 ofproto_dpif_self_check, NULL);
5735 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5736 ofproto_unixctl_dpif_dump_dps, NULL);
5737 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5739 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5740 ofproto_unixctl_dpif_dump_flows, NULL);
5741 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5742 ofproto_unixctl_dpif_del_flows, NULL);
5743 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5744 ofproto_unixctl_dpif_dump_megaflows, NULL);
5745 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5746 ofproto_unixctl_dpif_disable_megaflows, NULL);
5747 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5748 ofproto_unixctl_dpif_enable_megaflows, NULL);
5751 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5753 * This is deprecated. It is only for compatibility with broken device drivers
5754 * in old versions of Linux that do not properly support VLANs when VLAN
5755 * devices are not used. When broken device drivers are no longer in
5756 * widespread use, we will delete these interfaces. */
5759 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5762 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5764 if (realdev_ofp_port == ofport->realdev_ofp_port
5765 && vid == ofport->vlandev_vid) {
5769 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5771 if (ofport->realdev_ofp_port) {
5774 if (realdev_ofp_port && ofport->bundle) {
5775 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5776 * themselves be part of a bundle. */
5777 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5780 ofport->realdev_ofp_port = realdev_ofp_port;
5781 ofport->vlandev_vid = vid;
5783 if (realdev_ofp_port) {
5784 vsp_add(ofport, realdev_ofp_port, vid);
5791 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5793 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5797 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5798 OVS_EXCLUDED(ofproto->vsp_mutex)
5802 ovs_mutex_lock(&ofproto->vsp_mutex);
5803 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
5804 ovs_mutex_unlock(&ofproto->vsp_mutex);
5809 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5810 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5811 OVS_REQUIRES(ofproto->vsp_mutex)
5813 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5814 int vid = vlan_tci_to_vid(vlan_tci);
5815 const struct vlan_splinter *vsp;
5817 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5818 hash_realdev_vid(realdev_ofp_port, vid),
5819 &ofproto->realdev_vid_map) {
5820 if (vsp->realdev_ofp_port == realdev_ofp_port
5821 && vsp->vid == vid) {
5822 return vsp->vlandev_ofp_port;
5826 return realdev_ofp_port;
5829 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5830 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5831 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5832 * 'vlan_tci' 9, it would return the port number of eth0.9.
5834 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5835 * function just returns its 'realdev_ofp_port' argument. */
5837 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5838 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5839 OVS_EXCLUDED(ofproto->vsp_mutex)
5843 ovs_mutex_lock(&ofproto->vsp_mutex);
5844 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5845 ovs_mutex_unlock(&ofproto->vsp_mutex);
5849 static struct vlan_splinter *
5850 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5852 struct vlan_splinter *vsp;
5854 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5855 hash_ofp_port(vlandev_ofp_port),
5856 &ofproto->vlandev_map) {
5857 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5865 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5866 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5867 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5868 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5869 * eth0 and store 9 in '*vid'.
5871 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5872 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5875 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5876 ofp_port_t vlandev_ofp_port, int *vid)
5877 OVS_REQUIRES(ofproto->vsp_mutex)
5879 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5880 const struct vlan_splinter *vsp;
5882 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5887 return vsp->realdev_ofp_port;
5893 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5894 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5895 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5896 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
5897 * always the case unless VLAN splinters are enabled), returns false without
5898 * making any changes. */
5900 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
5901 OVS_EXCLUDED(ofproto->vsp_mutex)
5906 ovs_mutex_lock(&ofproto->vsp_mutex);
5907 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5908 ovs_mutex_unlock(&ofproto->vsp_mutex);
5913 /* Cause the flow to be processed as if it came in on the real device with
5914 * the VLAN device's VLAN ID. */
5915 flow->in_port.ofp_port = realdev;
5916 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5921 vsp_remove(struct ofport_dpif *port)
5923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5924 struct vlan_splinter *vsp;
5926 ovs_mutex_lock(&ofproto->vsp_mutex);
5927 vsp = vlandev_find(ofproto, port->up.ofp_port);
5929 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5930 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5933 port->realdev_ofp_port = 0;
5935 VLOG_ERR("missing vlan device record");
5937 ovs_mutex_unlock(&ofproto->vsp_mutex);
5941 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5943 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5945 ovs_mutex_lock(&ofproto->vsp_mutex);
5946 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5947 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5948 == realdev_ofp_port)) {
5949 struct vlan_splinter *vsp;
5951 vsp = xmalloc(sizeof *vsp);
5952 vsp->realdev_ofp_port = realdev_ofp_port;
5953 vsp->vlandev_ofp_port = port->up.ofp_port;
5956 port->realdev_ofp_port = realdev_ofp_port;
5958 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5959 hash_ofp_port(port->up.ofp_port));
5960 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5961 hash_realdev_vid(realdev_ofp_port, vid));
5963 VLOG_ERR("duplicate vlan device record");
5965 ovs_mutex_unlock(&ofproto->vsp_mutex);
5969 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5971 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
5972 return ofport ? ofport->odp_port : ODPP_NONE;
5975 struct ofport_dpif *
5976 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5978 struct ofport_dpif *port;
5980 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5981 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5982 &backer->odp_to_ofport_map) {
5983 if (port->odp_port == odp_port) {
5984 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5989 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5994 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5996 struct ofport_dpif *port;
5998 port = odp_port_to_ofport(ofproto->backer, odp_port);
5999 if (port && &ofproto->up == port->up.ofproto) {
6000 return port->up.ofp_port;
6006 const struct ofproto_class ofproto_dpif_class = {
6041 port_is_lacp_current,
6042 NULL, /* rule_choose_table */
6051 rule_modify_actions,
6065 get_stp_port_status,
6072 is_mirror_output_bundle,
6073 forward_bpdu_changed,
6074 set_mac_table_config,
6076 NULL, /* meter_get_features */
6077 NULL, /* meter_set */
6078 NULL, /* meter_get */
6079 NULL, /* meter_del */
6080 NULL, /* group_alloc */
6081 NULL, /* group_construct */
6082 NULL, /* group_destruct */
6083 NULL, /* group_dealloc */
6084 NULL, /* group_modify */
6085 NULL, /* group_get_stats */