2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
34 #include "guarded-list.h"
38 #include "mac-learning.h"
39 #include "meta-flow.h"
40 #include "multipath.h"
41 #include "netdev-vport.h"
46 #include "odp-execute.h"
49 #include "ofp-actions.h"
50 #include "ofp-parse.h"
51 #include "ofp-print.h"
52 #include "ofproto-dpif-governor.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-sflow.h"
56 #include "ofproto-dpif-upcall.h"
57 #include "ofproto-dpif-xlate.h"
58 #include "poll-loop.h"
63 #include "unaligned.h"
65 #include "vlan-bitmap.h"
68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
70 COVERAGE_DEFINE(ofproto_dpif_expired);
71 COVERAGE_DEFINE(facet_revalidate);
72 COVERAGE_DEFINE(facet_unexpected);
73 COVERAGE_DEFINE(facet_create);
74 COVERAGE_DEFINE(facet_remove);
75 COVERAGE_DEFINE(subfacet_create);
76 COVERAGE_DEFINE(subfacet_destroy);
77 COVERAGE_DEFINE(subfacet_install_fail);
78 COVERAGE_DEFINE(packet_in_overflow);
80 /* Number of implemented OpenFlow tables. */
81 enum { N_TABLES = 255 };
82 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
83 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
93 * - Do include packets and bytes from facets that have been deleted or
94 * whose own statistics have been folded into the rule.
96 * - Do include packets and bytes sent "by hand" that were accounted to
97 * the rule without any facet being involved (this is a rare corner
98 * case in rule_execute()).
100 * - Do not include packet or bytes that can be obtained from any facet's
101 * packet_count or byte_count member or that can be obtained from the
102 * datapath by, e.g., dpif_flow_get() for any subfacet.
104 struct ovs_mutex stats_mutex;
105 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
106 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
109 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
110 static struct rule_dpif *rule_dpif_cast(const struct rule *);
113 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 void *aux; /* Key supplied by ofproto's client. */
116 char *name; /* Identifier for log messages. */
119 struct list ports; /* Contains "struct ofport"s. */
120 enum port_vlan_mode vlan_mode; /* VLAN mode */
121 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
122 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
123 * NULL if all VLANs are trunked. */
124 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
125 struct bond *bond; /* Nonnull iff more than one port. */
126 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
129 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
132 static void bundle_remove(struct ofport *);
133 static void bundle_update(struct ofbundle *);
134 static void bundle_destroy(struct ofbundle *);
135 static void bundle_del_port(struct ofport_dpif *);
136 static void bundle_run(struct ofbundle *);
137 static void bundle_wait(struct ofbundle *);
139 static void stp_run(struct ofproto_dpif *ofproto);
140 static void stp_wait(struct ofproto_dpif *ofproto);
141 static int set_stp_port(struct ofport *,
142 const struct ofproto_port_stp_settings *);
144 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
145 enum slow_path_reason,
146 uint64_t *stub, size_t stub_size,
147 const struct nlattr **actionsp,
148 size_t *actions_lenp);
150 /* A subfacet (see "struct subfacet" below) has three possible installation
153 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
154 * case just after the subfacet is created, just before the subfacet is
155 * destroyed, or if the datapath returns an error when we try to install a
158 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
160 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
161 * ofproto_dpif is installed in the datapath.
164 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
165 SF_FAST_PATH, /* Full actions are installed. */
166 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
169 /* A dpif flow and actions associated with a facet.
171 * See also the large comment on struct facet. */
174 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
175 struct list list_node; /* In struct facet's 'facets' list. */
176 struct facet *facet; /* Owning facet. */
177 struct dpif_backer *backer; /* Owning backer. */
182 long long int used; /* Time last used; time created if not used. */
183 long long int created; /* Time created. */
185 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
186 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
188 enum subfacet_path path; /* Installed in datapath? */
191 #define SUBFACET_DESTROY_MAX_BATCH 50
193 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
194 static struct subfacet *subfacet_find(struct dpif_backer *,
195 const struct nlattr *key, size_t key_len,
197 static void subfacet_destroy(struct subfacet *);
198 static void subfacet_destroy__(struct subfacet *);
199 static void subfacet_destroy_batch(struct dpif_backer *,
200 struct subfacet **, int n);
201 static void subfacet_reset_dp_stats(struct subfacet *,
202 struct dpif_flow_stats *);
203 static void subfacet_update_stats(struct subfacet *,
204 const struct dpif_flow_stats *);
205 static int subfacet_install(struct subfacet *,
206 const struct ofpbuf *odp_actions,
207 struct dpif_flow_stats *);
208 static void subfacet_uninstall(struct subfacet *);
210 /* A unique, non-overlapping instantiation of an OpenFlow flow.
212 * A facet associates a "struct flow", which represents the Open vSwitch
213 * userspace idea of an exact-match flow, with one or more subfacets.
214 * While the facet is created based on an exact-match flow, it is stored
215 * within the ofproto based on the wildcards that could be expressed
216 * based on the flow table and other configuration. (See the 'wc'
217 * description in "struct xlate_out" for more details.)
219 * Each subfacet tracks the datapath's idea of the flow equivalent to
220 * the facet. When the kernel module (or other dpif implementation) and
221 * Open vSwitch userspace agree on the definition of a flow key, there
222 * is exactly one subfacet per facet. If the dpif implementation
223 * supports more-specific flow matching than userspace, however, a facet
224 * can have more than one subfacet. Examples include the dpif
225 * implementation not supporting the same wildcards as userspace or some
226 * distinction in flow that userspace simply doesn't understand.
228 * Flow expiration works in terms of subfacets, so a facet must have at
229 * least one subfacet or it will never expire, leaking memory. */
232 struct ofproto_dpif *ofproto;
235 struct list subfacets;
236 long long int used; /* Time last used; time created if not used. */
239 struct flow flow; /* Flow of the creating subfacet. */
240 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
244 * - Do include packets and bytes sent "by hand", e.g. with
247 * - Do include packets and bytes that were obtained from the datapath
248 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
249 * DPIF_FP_ZERO_STATS).
251 * - Do not include packets or bytes that can be obtained from the
252 * datapath for any existing subfacet.
254 uint64_t packet_count; /* Number of packets received. */
255 uint64_t byte_count; /* Number of bytes received. */
257 /* Resubmit statistics. */
258 uint64_t prev_packet_count; /* Number of packets from last stats push. */
259 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
260 long long int prev_used; /* Used time from last stats push. */
263 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
264 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
265 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
267 struct xlate_out xout;
269 /* Storage for a single subfacet, to reduce malloc() time and space
270 * overhead. (A facet always has at least one subfacet and in the common
271 * case has exactly one subfacet. However, 'one_subfacet' may not
272 * always be valid, since it could have been removed after newer
273 * subfacets were pushed onto the 'subfacets' list.) */
274 struct subfacet one_subfacet;
276 long long int learn_rl; /* Rate limiter for facet_learn(). */
279 static struct facet *facet_create(const struct flow_miss *);
280 static void facet_remove(struct facet *);
281 static void facet_free(struct facet *);
283 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
284 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
285 const struct flow *);
286 static bool facet_revalidate(struct facet *);
287 static bool facet_check_consistency(struct facet *);
289 static void facet_flush_stats(struct facet *);
291 static void facet_reset_counters(struct facet *);
292 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
293 struct dpif_flow_stats *, bool may_learn);
294 static void facet_push_stats(struct facet *, bool may_learn);
295 static void facet_learn(struct facet *);
296 static void facet_account(struct facet *);
297 static void push_all_stats(void);
299 static bool facet_is_controller_flow(struct facet *);
302 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
306 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
307 struct list bundle_node; /* In struct ofbundle's "ports" list. */
308 struct cfm *cfm; /* Connectivity Fault Management, if any. */
309 struct bfd *bfd; /* BFD, if any. */
310 bool may_enable; /* May be enabled in bonds. */
311 bool is_tunnel; /* This port is a tunnel. */
312 long long int carrier_seq; /* Carrier status changes. */
313 struct ofport_dpif *peer; /* Peer if patch port. */
316 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
317 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
318 long long int stp_state_entered;
320 /* Queue to DSCP mapping. */
321 struct ofproto_port_queue *qdscp;
324 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
326 * This is deprecated. It is only for compatibility with broken device
327 * drivers in old versions of Linux that do not properly support VLANs when
328 * VLAN devices are not used. When broken device drivers are no longer in
329 * widespread use, we will delete these interfaces. */
330 ofp_port_t realdev_ofp_port;
334 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
336 * This is deprecated. It is only for compatibility with broken device drivers
337 * in old versions of Linux that do not properly support VLANs when VLAN
338 * devices are not used. When broken device drivers are no longer in
339 * widespread use, we will delete these interfaces. */
340 struct vlan_splinter {
341 struct hmap_node realdev_vid_node;
342 struct hmap_node vlandev_node;
343 ofp_port_t realdev_ofp_port;
344 ofp_port_t vlandev_ofp_port;
348 static void vsp_remove(struct ofport_dpif *);
349 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
351 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
354 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
357 static struct ofport_dpif *
358 ofport_dpif_cast(const struct ofport *ofport)
360 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
363 static void port_run(struct ofport_dpif *);
364 static void port_run_fast(struct ofport_dpif *);
365 static void port_wait(struct ofport_dpif *);
366 static int set_bfd(struct ofport *, const struct smap *);
367 static int set_cfm(struct ofport *, const struct cfm_settings *);
368 static void ofport_update_peer(struct ofport_dpif *);
369 static void run_fast_rl(void);
370 static int run_fast(struct ofproto *);
372 struct dpif_completion {
373 struct list list_node;
374 struct ofoperation *op;
377 /* Reasons that we might need to revalidate every facet, and corresponding
380 * A value of 0 means that there is no need to revalidate.
382 * It would be nice to have some cleaner way to integrate with coverage
383 * counters, but with only a few reasons I guess this is good enough for
385 enum revalidate_reason {
386 REV_RECONFIGURE = 1, /* Switch configuration changed. */
387 REV_STP, /* Spanning tree protocol port status change. */
388 REV_BOND, /* Bonding changed. */
389 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
390 REV_FLOW_TABLE, /* Flow table changed. */
391 REV_MAC_LEARNING, /* Mac learning changed. */
392 REV_INCONSISTENCY /* Facet self-check failed. */
394 COVERAGE_DEFINE(rev_reconfigure);
395 COVERAGE_DEFINE(rev_stp);
396 COVERAGE_DEFINE(rev_bond);
397 COVERAGE_DEFINE(rev_port_toggled);
398 COVERAGE_DEFINE(rev_flow_table);
399 COVERAGE_DEFINE(rev_mac_learning);
400 COVERAGE_DEFINE(rev_inconsistency);
402 struct avg_subfacet_rates {
403 double add_rate; /* Moving average of new flows created per minute. */
404 double del_rate; /* Moving average of flows deleted per minute. */
407 /* All datapaths of a given type share a single dpif backer instance. */
413 struct timer next_expiration;
415 struct ovs_rwlock odp_to_ofport_lock;
416 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
418 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
420 /* Facet revalidation flags applying to facets which use this backer. */
421 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
423 struct hmap drop_keys; /* Set of dropped odp keys. */
424 bool recv_set_enable; /* Enables or disables receiving packets. */
426 struct hmap subfacets;
427 struct governor *governor;
429 /* Subfacet statistics.
431 * These keep track of the total number of subfacets added and deleted and
432 * flow life span. They are useful for computing the flow rates stats
433 * exposed via "ovs-appctl dpif/show". The goal is to learn about
434 * traffic patterns in ways that we can use later to improve Open vSwitch
435 * performance in new situations. */
436 long long int created; /* Time when it is created. */
437 unsigned max_n_subfacet; /* Maximum number of flows */
438 unsigned avg_n_subfacet; /* Average number of flows. */
439 long long int avg_subfacet_life; /* Average life span of subfacets. */
441 /* Number of upcall handling threads. */
442 unsigned int n_handler_threads;
445 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
446 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
448 static void drop_key_clear(struct dpif_backer *);
450 struct ofproto_dpif {
451 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
453 struct dpif_backer *backer;
455 /* Special OpenFlow rules. */
456 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
457 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
458 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
461 struct netflow *netflow;
462 struct dpif_sflow *sflow;
463 struct dpif_ipfix *ipfix;
464 struct hmap bundles; /* Contains "struct ofbundle"s. */
465 struct mac_learning *ml;
466 bool has_bonded_bundles;
467 struct mbridge *mbridge;
470 struct classifier facets; /* Contains 'struct facet's. */
471 long long int consistency_rl;
473 struct netdev_stats stats; /* To account packets generated and consumed in
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->vsp_mutex);
1235 classifier_init(&ofproto->facets);
1236 ofproto->consistency_rl = LLONG_MIN;
1238 guarded_list_init(&ofproto->pins);
1240 ofproto_dpif_unixctl_init();
1242 hmap_init(&ofproto->vlandev_map);
1243 hmap_init(&ofproto->realdev_vid_map);
1245 sset_init(&ofproto->ports);
1246 sset_init(&ofproto->ghost_ports);
1247 sset_init(&ofproto->port_poll_set);
1248 ofproto->port_poll_errno = 0;
1250 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1251 struct iface_hint *iface_hint = node->data;
1253 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1254 /* Check if the datapath already has this port. */
1255 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1256 sset_add(&ofproto->ports, node->name);
1259 free(iface_hint->br_name);
1260 free(iface_hint->br_type);
1262 shash_delete(&init_ofp_ports, node);
1266 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1267 hash_string(ofproto->up.name, 0));
1268 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1270 ofproto_init_tables(ofproto_, N_TABLES);
1271 error = add_internal_flows(ofproto);
1272 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1275 ofproto->n_missed = 0;
1281 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1282 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1284 struct ofputil_flow_mod fm;
1287 match_init_catchall(&fm.match);
1289 match_set_reg(&fm.match, 0, id);
1290 fm.new_cookie = htonll(0);
1291 fm.cookie = htonll(0);
1292 fm.cookie_mask = htonll(0);
1293 fm.modify_cookie = false;
1294 fm.table_id = TBL_INTERNAL;
1295 fm.command = OFPFC_ADD;
1296 fm.idle_timeout = 0;
1297 fm.hard_timeout = 0;
1301 fm.ofpacts = ofpacts->data;
1302 fm.ofpacts_len = ofpacts->size;
1304 error = ofproto_flow_mod(&ofproto->up, &fm);
1306 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1307 id, ofperr_to_string(error));
1311 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1313 rule_dpif_unref(*rulep);
1322 add_internal_flows(struct ofproto_dpif *ofproto)
1324 struct ofpact_controller *controller;
1325 uint64_t ofpacts_stub[128 / 8];
1326 struct ofpbuf ofpacts;
1330 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1333 controller = ofpact_put_CONTROLLER(&ofpacts);
1334 controller->max_len = UINT16_MAX;
1335 controller->controller_id = 0;
1336 controller->reason = OFPR_NO_MATCH;
1337 ofpact_pad(&ofpacts);
1339 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1344 ofpbuf_clear(&ofpacts);
1345 error = add_internal_flow(ofproto, id++, &ofpacts,
1346 &ofproto->no_packet_in_rule);
1351 error = add_internal_flow(ofproto, id++, &ofpacts,
1352 &ofproto->drop_frags_rule);
1357 destruct(struct ofproto *ofproto_)
1359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1360 struct rule_dpif *rule, *next_rule;
1361 struct ofputil_packet_in *pin, *next_pin;
1362 struct facet *facet, *next_facet;
1363 struct cls_cursor cursor;
1364 struct oftable *table;
1367 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1368 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1369 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1370 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1371 facet_remove(facet);
1374 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1375 ovs_rwlock_wrlock(&xlate_rwlock);
1376 xlate_remove_ofproto(ofproto);
1377 ovs_rwlock_unlock(&xlate_rwlock);
1379 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1380 * use-after-free error. */
1381 udpif_revalidate(ofproto->backer->udpif);
1383 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1385 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1386 struct cls_cursor cursor;
1388 ovs_rwlock_rdlock(&table->cls.rwlock);
1389 cls_cursor_init(&cursor, &table->cls, NULL);
1390 ovs_rwlock_unlock(&table->cls.rwlock);
1391 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1392 ofproto_rule_delete(&ofproto->up, &rule->up);
1396 guarded_list_pop_all(&ofproto->pins, &pins);
1397 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1398 list_remove(&pin->list_node);
1399 free(CONST_CAST(void *, pin->packet));
1402 guarded_list_destroy(&ofproto->pins);
1404 mbridge_unref(ofproto->mbridge);
1406 netflow_destroy(ofproto->netflow);
1407 dpif_sflow_unref(ofproto->sflow);
1408 hmap_destroy(&ofproto->bundles);
1409 mac_learning_unref(ofproto->ml);
1411 classifier_destroy(&ofproto->facets);
1413 hmap_destroy(&ofproto->vlandev_map);
1414 hmap_destroy(&ofproto->realdev_vid_map);
1416 sset_destroy(&ofproto->ports);
1417 sset_destroy(&ofproto->ghost_ports);
1418 sset_destroy(&ofproto->port_poll_set);
1420 ovs_mutex_destroy(&ofproto->vsp_mutex);
1422 close_dpif_backer(ofproto->backer);
1426 run_fast(struct ofproto *ofproto_)
1428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1429 struct ofputil_packet_in *pin, *next_pin;
1430 struct ofport_dpif *ofport;
1433 /* Do not perform any periodic activity required by 'ofproto' while
1434 * waiting for flow restore to complete. */
1435 if (ofproto_get_flow_restore_wait()) {
1439 guarded_list_pop_all(&ofproto->pins, &pins);
1440 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1441 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1442 list_remove(&pin->list_node);
1443 free(CONST_CAST(void *, pin->packet));
1447 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1448 port_run_fast(ofport);
1455 run(struct ofproto *ofproto_)
1457 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1458 struct ofport_dpif *ofport;
1459 struct ofbundle *bundle;
1462 if (mbridge_need_revalidate(ofproto->mbridge)) {
1463 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1464 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1465 mac_learning_flush(ofproto->ml);
1466 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1469 /* Do not perform any periodic activity below required by 'ofproto' while
1470 * waiting for flow restore to complete. */
1471 if (ofproto_get_flow_restore_wait()) {
1475 error = run_fast(ofproto_);
1480 if (ofproto->netflow) {
1481 if (netflow_run(ofproto->netflow)) {
1482 send_netflow_active_timeouts(ofproto);
1485 if (ofproto->sflow) {
1486 dpif_sflow_run(ofproto->sflow);
1488 if (ofproto->ipfix) {
1489 dpif_ipfix_run(ofproto->ipfix);
1492 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1495 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1500 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1501 if (mac_learning_run(ofproto->ml)) {
1502 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1504 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1506 /* Check the consistency of a random facet, to aid debugging. */
1507 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1508 if (time_msec() >= ofproto->consistency_rl
1509 && !classifier_is_empty(&ofproto->facets)
1510 && !ofproto->backer->need_revalidate) {
1511 struct cls_table *table;
1512 struct cls_rule *cr;
1513 struct facet *facet;
1515 ofproto->consistency_rl = time_msec() + 250;
1517 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1518 struct cls_table, hmap_node);
1519 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1521 facet = CONTAINER_OF(cr, struct facet, cr);
1523 if (!facet_check_consistency(facet)) {
1524 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1527 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1533 wait(struct ofproto *ofproto_)
1535 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1536 struct ofport_dpif *ofport;
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 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1552 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1553 bundle_wait(bundle);
1555 if (ofproto->netflow) {
1556 netflow_wait(ofproto->netflow);
1558 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1559 mac_learning_wait(ofproto->ml);
1560 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1562 if (ofproto->backer->need_revalidate) {
1563 /* Shouldn't happen, but if it does just go around again. */
1564 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1565 poll_immediate_wake();
1570 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1572 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1573 struct cls_cursor cursor;
1574 size_t n_subfacets = 0;
1575 struct facet *facet;
1577 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1578 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1579 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1581 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1582 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1583 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1584 n_subfacets += list_size(&facet->subfacets);
1586 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1587 simap_increase(usage, "subfacets", n_subfacets);
1591 flush(struct ofproto *ofproto_)
1593 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1594 struct subfacet *subfacet, *next_subfacet;
1595 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1599 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1600 &ofproto->backer->subfacets) {
1601 if (subfacet->facet->ofproto != ofproto) {
1605 if (subfacet->path != SF_NOT_INSTALLED) {
1606 batch[n_batch++] = subfacet;
1607 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1608 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1612 subfacet_destroy(subfacet);
1617 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1622 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1623 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1625 *arp_match_ip = true;
1626 *actions = (OFPUTIL_A_OUTPUT |
1627 OFPUTIL_A_SET_VLAN_VID |
1628 OFPUTIL_A_SET_VLAN_PCP |
1629 OFPUTIL_A_STRIP_VLAN |
1630 OFPUTIL_A_SET_DL_SRC |
1631 OFPUTIL_A_SET_DL_DST |
1632 OFPUTIL_A_SET_NW_SRC |
1633 OFPUTIL_A_SET_NW_DST |
1634 OFPUTIL_A_SET_NW_TOS |
1635 OFPUTIL_A_SET_TP_SRC |
1636 OFPUTIL_A_SET_TP_DST |
1641 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1644 struct dpif_dp_stats s;
1645 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1648 strcpy(ots->name, "classifier");
1650 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1651 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1652 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1653 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1655 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1656 ots->lookup_count = htonll(n_lookup);
1657 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1660 static struct ofport *
1663 struct ofport_dpif *port = xmalloc(sizeof *port);
1668 port_dealloc(struct ofport *port_)
1670 struct ofport_dpif *port = ofport_dpif_cast(port_);
1675 port_construct(struct ofport *port_)
1677 struct ofport_dpif *port = ofport_dpif_cast(port_);
1678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1679 const struct netdev *netdev = port->up.netdev;
1680 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1681 struct dpif_port dpif_port;
1684 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1685 port->bundle = NULL;
1688 port->may_enable = true;
1689 port->stp_port = NULL;
1690 port->stp_state = STP_DISABLED;
1691 port->is_tunnel = false;
1695 port->realdev_ofp_port = 0;
1696 port->vlandev_vid = 0;
1697 port->carrier_seq = netdev_get_carrier_resets(netdev);
1699 if (netdev_vport_is_patch(netdev)) {
1700 /* By bailing out here, we don't submit the port to the sFlow module
1701 * to be considered for counter polling export. This is correct
1702 * because the patch port represents an interface that sFlow considers
1703 * to be "internal" to the switch as a whole, and therefore not an
1704 * candidate for counter polling. */
1705 port->odp_port = ODPP_NONE;
1706 ofport_update_peer(port);
1710 error = dpif_port_query_by_name(ofproto->backer->dpif,
1711 netdev_vport_get_dpif_port(netdev, namebuf,
1718 port->odp_port = dpif_port.port_no;
1720 if (netdev_get_tunnel_config(netdev)) {
1721 tnl_port_add(port, port->up.netdev, port->odp_port);
1722 port->is_tunnel = true;
1724 /* Sanity-check that a mapping doesn't already exist. This
1725 * shouldn't happen for non-tunnel ports. */
1726 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1727 VLOG_ERR("port %s already has an OpenFlow port number",
1729 dpif_port_destroy(&dpif_port);
1733 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1734 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1735 hash_odp_port(port->odp_port));
1736 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1738 dpif_port_destroy(&dpif_port);
1740 if (ofproto->sflow) {
1741 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1748 port_destruct(struct ofport *port_)
1750 struct ofport_dpif *port = ofport_dpif_cast(port_);
1751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1752 const char *devname = netdev_get_name(port->up.netdev);
1753 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1754 const char *dp_port_name;
1756 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1757 ovs_rwlock_wrlock(&xlate_rwlock);
1758 xlate_ofport_remove(port);
1759 ovs_rwlock_unlock(&xlate_rwlock);
1761 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1763 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1764 /* The underlying device is still there, so delete it. This
1765 * happens when the ofproto is being destroyed, since the caller
1766 * assumes that removal of attached ports will happen as part of
1768 if (!port->is_tunnel) {
1769 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1774 port->peer->peer = NULL;
1778 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1779 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1780 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1781 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1785 sset_find_and_delete(&ofproto->ports, devname);
1786 sset_find_and_delete(&ofproto->ghost_ports, devname);
1787 bundle_remove(port_);
1788 set_cfm(port_, NULL);
1789 set_bfd(port_, NULL);
1790 if (ofproto->sflow) {
1791 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1798 port_modified(struct ofport *port_)
1800 struct ofport_dpif *port = ofport_dpif_cast(port_);
1802 if (port->bundle && port->bundle->bond) {
1803 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1807 cfm_set_netdev(port->cfm, port->up.netdev);
1811 bfd_set_netdev(port->bfd, port->up.netdev);
1814 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1816 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1820 ofport_update_peer(port);
1824 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1826 struct ofport_dpif *port = ofport_dpif_cast(port_);
1827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1828 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1830 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1831 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1832 OFPUTIL_PC_NO_PACKET_IN)) {
1833 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1835 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1836 bundle_update(port->bundle);
1842 set_sflow(struct ofproto *ofproto_,
1843 const struct ofproto_sflow_options *sflow_options)
1845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1846 struct dpif_sflow *ds = ofproto->sflow;
1848 if (sflow_options) {
1850 struct ofport_dpif *ofport;
1852 ds = ofproto->sflow = dpif_sflow_create();
1853 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1854 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1856 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1858 dpif_sflow_set_options(ds, sflow_options);
1861 dpif_sflow_unref(ds);
1862 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1863 ofproto->sflow = NULL;
1871 struct ofproto *ofproto_,
1872 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1873 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1874 size_t n_flow_exporters_options)
1876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1877 struct dpif_ipfix *di = ofproto->ipfix;
1878 bool has_options = bridge_exporter_options || flow_exporters_options;
1880 if (has_options && !di) {
1881 di = ofproto->ipfix = dpif_ipfix_create();
1885 /* Call set_options in any case to cleanly flush the flow
1886 * caches in the last exporters that are to be destroyed. */
1887 dpif_ipfix_set_options(
1888 di, bridge_exporter_options, flow_exporters_options,
1889 n_flow_exporters_options);
1892 dpif_ipfix_unref(di);
1893 ofproto->ipfix = NULL;
1901 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1903 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1910 struct ofproto_dpif *ofproto;
1912 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1914 ofport->cfm = cfm_create(ofport->up.netdev);
1917 if (cfm_configure(ofport->cfm, s)) {
1923 cfm_unref(ofport->cfm);
1929 get_cfm_status(const struct ofport *ofport_,
1930 struct ofproto_cfm_status *status)
1932 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1935 status->faults = cfm_get_fault(ofport->cfm);
1936 status->remote_opstate = cfm_get_opup(ofport->cfm);
1937 status->health = cfm_get_health(ofport->cfm);
1938 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1946 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1949 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1953 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1954 cfg, ofport->up.netdev);
1955 if (ofport->bfd != old) {
1956 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1963 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1965 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1968 bfd_get_status(ofport->bfd, smap);
1975 /* Spanning Tree. */
1978 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1980 struct ofproto_dpif *ofproto = ofproto_;
1981 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1982 struct ofport_dpif *ofport;
1984 ofport = stp_port_get_aux(sp);
1986 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1987 ofproto->up.name, port_num);
1989 struct eth_header *eth = pkt->l2;
1991 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1992 if (eth_addr_is_zero(eth->eth_src)) {
1993 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1994 "with unknown MAC", ofproto->up.name, port_num);
1996 ofproto_dpif_send_packet(ofport, pkt);
2002 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2004 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2008 /* Only revalidate flows if the configuration changed. */
2009 if (!s != !ofproto->stp) {
2010 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2014 if (!ofproto->stp) {
2015 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2016 send_bpdu_cb, ofproto);
2017 ofproto->stp_last_tick = time_msec();
2020 stp_set_bridge_id(ofproto->stp, s->system_id);
2021 stp_set_bridge_priority(ofproto->stp, s->priority);
2022 stp_set_hello_time(ofproto->stp, s->hello_time);
2023 stp_set_max_age(ofproto->stp, s->max_age);
2024 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2026 struct ofport *ofport;
2028 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2029 set_stp_port(ofport, NULL);
2032 stp_unref(ofproto->stp);
2033 ofproto->stp = NULL;
2040 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2042 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2046 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2047 s->designated_root = stp_get_designated_root(ofproto->stp);
2048 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2057 update_stp_port_state(struct ofport_dpif *ofport)
2059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2060 enum stp_state state;
2062 /* Figure out new state. */
2063 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2067 if (ofport->stp_state != state) {
2068 enum ofputil_port_state of_state;
2071 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2072 netdev_get_name(ofport->up.netdev),
2073 stp_state_name(ofport->stp_state),
2074 stp_state_name(state));
2075 if (stp_learn_in_state(ofport->stp_state)
2076 != stp_learn_in_state(state)) {
2077 /* xxx Learning action flows should also be flushed. */
2078 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2079 mac_learning_flush(ofproto->ml);
2080 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2082 fwd_change = stp_forward_in_state(ofport->stp_state)
2083 != stp_forward_in_state(state);
2085 ofproto->backer->need_revalidate = REV_STP;
2086 ofport->stp_state = state;
2087 ofport->stp_state_entered = time_msec();
2089 if (fwd_change && ofport->bundle) {
2090 bundle_update(ofport->bundle);
2093 /* Update the STP state bits in the OpenFlow port description. */
2094 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2095 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2096 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2097 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2098 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2100 ofproto_port_set_state(&ofport->up, of_state);
2104 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2105 * caller is responsible for assigning STP port numbers and ensuring
2106 * there are no duplicates. */
2108 set_stp_port(struct ofport *ofport_,
2109 const struct ofproto_port_stp_settings *s)
2111 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2113 struct stp_port *sp = ofport->stp_port;
2115 if (!s || !s->enable) {
2117 ofport->stp_port = NULL;
2118 stp_port_disable(sp);
2119 update_stp_port_state(ofport);
2122 } else if (sp && stp_port_no(sp) != s->port_num
2123 && ofport == stp_port_get_aux(sp)) {
2124 /* The port-id changed, so disable the old one if it's not
2125 * already in use by another port. */
2126 stp_port_disable(sp);
2129 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2130 stp_port_enable(sp);
2132 stp_port_set_aux(sp, ofport);
2133 stp_port_set_priority(sp, s->priority);
2134 stp_port_set_path_cost(sp, s->path_cost);
2136 update_stp_port_state(ofport);
2142 get_stp_port_status(struct ofport *ofport_,
2143 struct ofproto_port_stp_status *s)
2145 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2147 struct stp_port *sp = ofport->stp_port;
2149 if (!ofproto->stp || !sp) {
2155 s->port_id = stp_port_get_id(sp);
2156 s->state = stp_port_get_state(sp);
2157 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2158 s->role = stp_port_get_role(sp);
2159 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2165 stp_run(struct ofproto_dpif *ofproto)
2168 long long int now = time_msec();
2169 long long int elapsed = now - ofproto->stp_last_tick;
2170 struct stp_port *sp;
2173 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2174 ofproto->stp_last_tick = now;
2176 while (stp_get_changed_port(ofproto->stp, &sp)) {
2177 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2180 update_stp_port_state(ofport);
2184 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2185 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2186 mac_learning_flush(ofproto->ml);
2187 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2193 stp_wait(struct ofproto_dpif *ofproto)
2196 poll_timer_wait(1000);
2201 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2204 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2207 if (ofport->n_qdscp != n_qdscp
2208 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2209 n_qdscp * sizeof *qdscp))) {
2210 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2211 free(ofport->qdscp);
2212 ofport->qdscp = n_qdscp
2213 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2215 ofport->n_qdscp = n_qdscp;
2223 /* Expires all MAC learning entries associated with 'bundle' and forces its
2224 * ofproto to revalidate every flow.
2226 * Normally MAC learning entries are removed only from the ofproto associated
2227 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2228 * are removed from every ofproto. When patch ports and SLB bonds are in use
2229 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2230 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2231 * with the host from which it migrated. */
2233 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2235 struct ofproto_dpif *ofproto = bundle->ofproto;
2236 struct mac_learning *ml = ofproto->ml;
2237 struct mac_entry *mac, *next_mac;
2239 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2240 ovs_rwlock_wrlock(&ml->rwlock);
2241 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2242 if (mac->port.p == bundle) {
2244 struct ofproto_dpif *o;
2246 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2248 struct mac_entry *e;
2250 ovs_rwlock_wrlock(&o->ml->rwlock);
2251 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2253 mac_learning_expire(o->ml, e);
2255 ovs_rwlock_unlock(&o->ml->rwlock);
2260 mac_learning_expire(ml, mac);
2263 ovs_rwlock_unlock(&ml->rwlock);
2266 static struct ofbundle *
2267 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2269 struct ofbundle *bundle;
2271 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2272 &ofproto->bundles) {
2273 if (bundle->aux == aux) {
2281 bundle_update(struct ofbundle *bundle)
2283 struct ofport_dpif *port;
2285 bundle->floodable = true;
2286 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2287 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2288 || !stp_forward_in_state(port->stp_state)) {
2289 bundle->floodable = false;
2296 bundle_del_port(struct ofport_dpif *port)
2298 struct ofbundle *bundle = port->bundle;
2300 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2302 list_remove(&port->bundle_node);
2303 port->bundle = NULL;
2306 lacp_slave_unregister(bundle->lacp, port);
2309 bond_slave_unregister(bundle->bond, port);
2312 bundle_update(bundle);
2316 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2317 struct lacp_slave_settings *lacp)
2319 struct ofport_dpif *port;
2321 port = get_ofp_port(bundle->ofproto, ofp_port);
2326 if (port->bundle != bundle) {
2327 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2329 bundle_remove(&port->up);
2332 port->bundle = bundle;
2333 list_push_back(&bundle->ports, &port->bundle_node);
2334 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2335 || !stp_forward_in_state(port->stp_state)) {
2336 bundle->floodable = false;
2340 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2341 lacp_slave_register(bundle->lacp, port, lacp);
2348 bundle_destroy(struct ofbundle *bundle)
2350 struct ofproto_dpif *ofproto;
2351 struct ofport_dpif *port, *next_port;
2357 ofproto = bundle->ofproto;
2358 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2360 ovs_rwlock_wrlock(&xlate_rwlock);
2361 xlate_bundle_remove(bundle);
2362 ovs_rwlock_unlock(&xlate_rwlock);
2364 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2365 bundle_del_port(port);
2368 bundle_flush_macs(bundle, true);
2369 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2371 free(bundle->trunks);
2372 lacp_unref(bundle->lacp);
2373 bond_unref(bundle->bond);
2378 bundle_set(struct ofproto *ofproto_, void *aux,
2379 const struct ofproto_bundle_settings *s)
2381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2382 bool need_flush = false;
2383 struct ofport_dpif *port;
2384 struct ofbundle *bundle;
2385 unsigned long *trunks;
2391 bundle_destroy(bundle_lookup(ofproto, aux));
2395 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2396 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2398 bundle = bundle_lookup(ofproto, aux);
2400 bundle = xmalloc(sizeof *bundle);
2402 bundle->ofproto = ofproto;
2403 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2404 hash_pointer(aux, 0));
2406 bundle->name = NULL;
2408 list_init(&bundle->ports);
2409 bundle->vlan_mode = PORT_VLAN_TRUNK;
2411 bundle->trunks = NULL;
2412 bundle->use_priority_tags = s->use_priority_tags;
2413 bundle->lacp = NULL;
2414 bundle->bond = NULL;
2416 bundle->floodable = true;
2417 mbridge_register_bundle(ofproto->mbridge, bundle);
2420 if (!bundle->name || strcmp(s->name, bundle->name)) {
2422 bundle->name = xstrdup(s->name);
2427 if (!bundle->lacp) {
2428 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2429 bundle->lacp = lacp_create();
2431 lacp_configure(bundle->lacp, s->lacp);
2433 lacp_unref(bundle->lacp);
2434 bundle->lacp = NULL;
2437 /* Update set of ports. */
2439 for (i = 0; i < s->n_slaves; i++) {
2440 if (!bundle_add_port(bundle, s->slaves[i],
2441 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2445 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2446 struct ofport_dpif *next_port;
2448 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2449 for (i = 0; i < s->n_slaves; i++) {
2450 if (s->slaves[i] == port->up.ofp_port) {
2455 bundle_del_port(port);
2459 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2461 if (list_is_empty(&bundle->ports)) {
2462 bundle_destroy(bundle);
2466 /* Set VLAN tagging mode */
2467 if (s->vlan_mode != bundle->vlan_mode
2468 || s->use_priority_tags != bundle->use_priority_tags) {
2469 bundle->vlan_mode = s->vlan_mode;
2470 bundle->use_priority_tags = s->use_priority_tags;
2475 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2476 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2478 if (vlan != bundle->vlan) {
2479 bundle->vlan = vlan;
2483 /* Get trunked VLANs. */
2484 switch (s->vlan_mode) {
2485 case PORT_VLAN_ACCESS:
2489 case PORT_VLAN_TRUNK:
2490 trunks = CONST_CAST(unsigned long *, s->trunks);
2493 case PORT_VLAN_NATIVE_UNTAGGED:
2494 case PORT_VLAN_NATIVE_TAGGED:
2495 if (vlan != 0 && (!s->trunks
2496 || !bitmap_is_set(s->trunks, vlan)
2497 || bitmap_is_set(s->trunks, 0))) {
2498 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2500 trunks = bitmap_clone(s->trunks, 4096);
2502 trunks = bitmap_allocate1(4096);
2504 bitmap_set1(trunks, vlan);
2505 bitmap_set0(trunks, 0);
2507 trunks = CONST_CAST(unsigned long *, s->trunks);
2514 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2515 free(bundle->trunks);
2516 if (trunks == s->trunks) {
2517 bundle->trunks = vlan_bitmap_clone(trunks);
2519 bundle->trunks = trunks;
2524 if (trunks != s->trunks) {
2529 if (!list_is_short(&bundle->ports)) {
2530 bundle->ofproto->has_bonded_bundles = true;
2532 if (bond_reconfigure(bundle->bond, s->bond)) {
2533 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2536 bundle->bond = bond_create(s->bond);
2537 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2540 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2541 bond_slave_register(bundle->bond, port, port->up.netdev);
2544 bond_unref(bundle->bond);
2545 bundle->bond = NULL;
2548 /* If we changed something that would affect MAC learning, un-learn
2549 * everything on this port and force flow revalidation. */
2551 bundle_flush_macs(bundle, false);
2558 bundle_remove(struct ofport *port_)
2560 struct ofport_dpif *port = ofport_dpif_cast(port_);
2561 struct ofbundle *bundle = port->bundle;
2564 bundle_del_port(port);
2565 if (list_is_empty(&bundle->ports)) {
2566 bundle_destroy(bundle);
2567 } else if (list_is_short(&bundle->ports)) {
2568 bond_unref(bundle->bond);
2569 bundle->bond = NULL;
2575 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2577 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2578 struct ofport_dpif *port = port_;
2579 uint8_t ea[ETH_ADDR_LEN];
2582 error = netdev_get_etheraddr(port->up.netdev, ea);
2584 struct ofpbuf packet;
2587 ofpbuf_init(&packet, 0);
2588 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2590 memcpy(packet_pdu, pdu, pdu_size);
2592 ofproto_dpif_send_packet(port, &packet);
2593 ofpbuf_uninit(&packet);
2595 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2596 "%s (%s)", port->bundle->name,
2597 netdev_get_name(port->up.netdev), ovs_strerror(error));
2602 bundle_send_learning_packets(struct ofbundle *bundle)
2604 struct ofproto_dpif *ofproto = bundle->ofproto;
2605 struct ofpbuf *learning_packet;
2606 int error, n_packets, n_errors;
2607 struct mac_entry *e;
2608 struct list packets;
2610 list_init(&packets);
2611 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2612 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2613 if (e->port.p != bundle) {
2616 learning_packet = bond_compose_learning_packet(bundle->bond,
2619 learning_packet->private_p = port_void;
2620 list_push_back(&packets, &learning_packet->list_node);
2623 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2625 error = n_packets = n_errors = 0;
2626 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2629 ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2636 ofpbuf_list_delete(&packets);
2639 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2640 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2641 "packets, last error was: %s",
2642 bundle->name, n_errors, n_packets, ovs_strerror(error));
2644 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2645 bundle->name, n_packets);
2650 bundle_run(struct ofbundle *bundle)
2653 lacp_run(bundle->lacp, send_pdu_cb);
2656 struct ofport_dpif *port;
2658 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2659 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2662 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2663 bundle->ofproto->backer->need_revalidate = REV_BOND;
2666 if (bond_should_send_learning_packets(bundle->bond)) {
2667 bundle_send_learning_packets(bundle);
2673 bundle_wait(struct ofbundle *bundle)
2676 lacp_wait(bundle->lacp);
2679 bond_wait(bundle->bond);
2686 mirror_set__(struct ofproto *ofproto_, void *aux,
2687 const struct ofproto_mirror_settings *s)
2689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2690 struct ofbundle **srcs, **dsts;
2695 mirror_destroy(ofproto->mbridge, aux);
2699 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2700 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2702 for (i = 0; i < s->n_srcs; i++) {
2703 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2706 for (i = 0; i < s->n_dsts; i++) {
2707 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2710 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2711 s->n_dsts, s->src_vlans,
2712 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2719 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2720 uint64_t *packets, uint64_t *bytes)
2723 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2728 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2731 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2732 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2733 mac_learning_flush(ofproto->ml);
2735 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2740 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2743 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2744 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2748 forward_bpdu_changed(struct ofproto *ofproto_)
2750 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2751 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2755 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2759 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2760 mac_learning_set_idle_time(ofproto->ml, idle_time);
2761 mac_learning_set_max_entries(ofproto->ml, max_entries);
2762 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2767 static struct ofport_dpif *
2768 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2770 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2771 return ofport ? ofport_dpif_cast(ofport) : NULL;
2774 static struct ofport_dpif *
2775 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2777 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2778 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2782 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2783 struct ofproto_port *ofproto_port,
2784 struct dpif_port *dpif_port)
2786 ofproto_port->name = dpif_port->name;
2787 ofproto_port->type = dpif_port->type;
2788 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2792 ofport_update_peer(struct ofport_dpif *ofport)
2794 const struct ofproto_dpif *ofproto;
2795 struct dpif_backer *backer;
2798 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2802 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2803 backer->need_revalidate = REV_RECONFIGURE;
2806 ofport->peer->peer = NULL;
2807 ofport->peer = NULL;
2810 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2815 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2816 struct ofport *peer_ofport;
2817 struct ofport_dpif *peer;
2820 if (ofproto->backer != backer) {
2824 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2829 peer = ofport_dpif_cast(peer_ofport);
2830 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2831 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2833 ofport->peer = peer;
2834 ofport->peer->peer = ofport;
2844 port_run_fast(struct ofport_dpif *ofport)
2846 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2847 struct ofpbuf packet;
2849 ofpbuf_init(&packet, 0);
2850 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2851 ofproto_dpif_send_packet(ofport, &packet);
2852 ofpbuf_uninit(&packet);
2855 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2856 struct ofpbuf packet;
2858 ofpbuf_init(&packet, 0);
2859 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2860 ofproto_dpif_send_packet(ofport, &packet);
2861 ofpbuf_uninit(&packet);
2866 port_run(struct ofport_dpif *ofport)
2868 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2869 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2870 bool enable = netdev_get_carrier(ofport->up.netdev);
2871 bool cfm_enable = false;
2872 bool bfd_enable = false;
2874 ofport->carrier_seq = carrier_seq;
2876 port_run_fast(ofport);
2879 int cfm_opup = cfm_get_opup(ofport->cfm);
2881 cfm_run(ofport->cfm);
2882 cfm_enable = !cfm_get_fault(ofport->cfm);
2884 if (cfm_opup >= 0) {
2885 cfm_enable = cfm_enable && cfm_opup;
2890 bfd_run(ofport->bfd);
2891 bfd_enable = bfd_forwarding(ofport->bfd);
2894 if (ofport->bfd || ofport->cfm) {
2895 enable = enable && (cfm_enable || bfd_enable);
2898 if (ofport->bundle) {
2899 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2900 if (carrier_changed) {
2901 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2905 if (ofport->may_enable != enable) {
2906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2907 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2910 ofport->may_enable = enable;
2914 port_wait(struct ofport_dpif *ofport)
2917 cfm_wait(ofport->cfm);
2921 bfd_wait(ofport->bfd);
2926 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2927 struct ofproto_port *ofproto_port)
2929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2930 struct dpif_port dpif_port;
2933 if (sset_contains(&ofproto->ghost_ports, devname)) {
2934 const char *type = netdev_get_type_from_name(devname);
2936 /* We may be called before ofproto->up.port_by_name is populated with
2937 * the appropriate ofport. For this reason, we must get the name and
2938 * type from the netdev layer directly. */
2940 const struct ofport *ofport;
2942 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2943 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2944 ofproto_port->name = xstrdup(devname);
2945 ofproto_port->type = xstrdup(type);
2951 if (!sset_contains(&ofproto->ports, devname)) {
2954 error = dpif_port_query_by_name(ofproto->backer->dpif,
2955 devname, &dpif_port);
2957 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2963 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2965 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2966 const char *devname = netdev_get_name(netdev);
2967 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2968 const char *dp_port_name;
2970 if (netdev_vport_is_patch(netdev)) {
2971 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2975 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2976 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2977 odp_port_t port_no = ODPP_NONE;
2980 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2984 if (netdev_get_tunnel_config(netdev)) {
2985 simap_put(&ofproto->backer->tnl_backers,
2986 dp_port_name, odp_to_u32(port_no));
2990 if (netdev_get_tunnel_config(netdev)) {
2991 sset_add(&ofproto->ghost_ports, devname);
2993 sset_add(&ofproto->ports, devname);
2999 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3002 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3009 sset_find_and_delete(&ofproto->ghost_ports,
3010 netdev_get_name(ofport->up.netdev));
3011 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3012 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3013 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3015 /* The caller is going to close ofport->up.netdev. If this is a
3016 * bonded port, then the bond is using that netdev, so remove it
3017 * from the bond. The client will need to reconfigure everything
3018 * after deleting ports, so then the slave will get re-added. */
3019 bundle_remove(&ofport->up);
3026 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3028 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3033 error = netdev_get_stats(ofport->up.netdev, stats);
3035 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3036 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3038 /* ofproto->stats.tx_packets represents packets that we created
3039 * internally and sent to some port (e.g. packets sent with
3040 * ofproto_dpif_send_packet()). Account for them as if they had
3041 * come from OFPP_LOCAL and got forwarded. */
3043 if (stats->rx_packets != UINT64_MAX) {
3044 stats->rx_packets += ofproto->stats.tx_packets;
3047 if (stats->rx_bytes != UINT64_MAX) {
3048 stats->rx_bytes += ofproto->stats.tx_bytes;
3051 /* ofproto->stats.rx_packets represents packets that were received on
3052 * some port and we processed internally and dropped (e.g. STP).
3053 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3055 if (stats->tx_packets != UINT64_MAX) {
3056 stats->tx_packets += ofproto->stats.rx_packets;
3059 if (stats->tx_bytes != UINT64_MAX) {
3060 stats->tx_bytes += ofproto->stats.rx_bytes;
3067 struct port_dump_state {
3072 struct ofproto_port port;
3077 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3079 *statep = xzalloc(sizeof(struct port_dump_state));
3084 port_dump_next(const struct ofproto *ofproto_, void *state_,
3085 struct ofproto_port *port)
3087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3088 struct port_dump_state *state = state_;
3089 const struct sset *sset;
3090 struct sset_node *node;
3092 if (state->has_port) {
3093 ofproto_port_destroy(&state->port);
3094 state->has_port = false;
3096 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3097 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3100 error = port_query_by_name(ofproto_, node->name, &state->port);
3102 *port = state->port;
3103 state->has_port = true;
3105 } else if (error != ENODEV) {
3110 if (!state->ghost) {
3111 state->ghost = true;
3114 return port_dump_next(ofproto_, state_, port);
3121 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3123 struct port_dump_state *state = state_;
3125 if (state->has_port) {
3126 ofproto_port_destroy(&state->port);
3133 port_poll(const struct ofproto *ofproto_, char **devnamep)
3135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3137 if (ofproto->port_poll_errno) {
3138 int error = ofproto->port_poll_errno;
3139 ofproto->port_poll_errno = 0;
3143 if (sset_is_empty(&ofproto->port_poll_set)) {
3147 *devnamep = sset_pop(&ofproto->port_poll_set);
3152 port_poll_wait(const struct ofproto *ofproto_)
3154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3155 dpif_port_poll_wait(ofproto->backer->dpif);
3159 port_is_lacp_current(const struct ofport *ofport_)
3161 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3162 return (ofport->bundle && ofport->bundle->lacp
3163 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3167 /* Upcall handling. */
3169 struct flow_miss_op {
3170 struct dpif_op dpif_op;
3172 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3173 struct xlate_out xout;
3174 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3176 struct ofpbuf mask; /* Flow mask for "put" ops. */
3177 struct odputil_keybuf maskbuf;
3179 /* If this is a "put" op, then a pointer to the subfacet that should
3180 * be marked as uninstalled if the operation fails. */
3181 struct subfacet *subfacet;
3184 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3185 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3186 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3187 * return value of true). However, for short flows the cost of bookkeeping is
3188 * much higher than the benefits, so when the datapath holds a large number of
3189 * flows we impose some heuristics to decide which flows are likely to be worth
3192 flow_miss_should_make_facet(struct flow_miss *miss)
3194 struct dpif_backer *backer = miss->ofproto->backer;
3197 switch (flow_miss_model) {
3198 case OFPROTO_HANDLE_MISS_AUTO:
3200 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3202 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3206 if (!backer->governor) {
3209 n_subfacets = hmap_count(&backer->subfacets);
3210 if (n_subfacets * 2 <= flow_eviction_threshold) {
3214 backer->governor = governor_create();
3217 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3218 return governor_should_install_flow(backer->governor, hash,
3219 miss->stats.n_packets);
3222 /* Handles 'miss', which matches 'facet'. May add any required datapath
3223 * operations to 'ops', incrementing '*n_ops' for each new op.
3225 * All of the packets in 'miss' are considered to have arrived at time
3226 * 'miss->stats.used'. This is really important only for new facets: if we
3227 * just called time_msec() here, then the new subfacet or its packets could
3228 * look (occasionally) as though it was used some time after the facet was
3229 * used. That can make a one-packet flow look like it has a nonzero duration,
3230 * which looks odd in e.g. NetFlow statistics. */
3232 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3233 struct flow_miss_op *ops, size_t *n_ops)
3235 enum subfacet_path want_path;
3236 struct subfacet *subfacet;
3238 facet->packet_count += miss->stats.n_packets;
3239 facet->prev_packet_count += miss->stats.n_packets;
3240 facet->byte_count += miss->stats.n_bytes;
3241 facet->prev_byte_count += miss->stats.n_bytes;
3243 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3245 /* Don't install the flow if it's the result of the "userspace"
3246 * action for an already installed facet. This can occur when a
3247 * datapath flow with wildcards has a "userspace" action and flows
3248 * sent to userspace result in a different subfacet, which will then
3249 * be rejected as overlapping by the datapath. */
3250 if (miss->upcall_type == DPIF_UC_ACTION
3251 && !list_is_empty(&facet->subfacets)) {
3255 subfacet = subfacet_create(facet, miss);
3256 if (subfacet->path != want_path) {
3257 struct flow_miss_op *op = &ops[(*n_ops)++];
3258 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3260 subfacet->path = want_path;
3262 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3263 if (enable_megaflows) {
3264 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3265 &miss->flow, UINT32_MAX);
3268 op->xout_garbage = false;
3269 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3270 op->subfacet = subfacet;
3271 put->flags = DPIF_FP_CREATE;
3272 put->key = miss->key;
3273 put->key_len = miss->key_len;
3274 put->mask = op->mask.data;
3275 put->mask_len = op->mask.size;
3277 if (want_path == SF_FAST_PATH) {
3278 put->actions = facet->xout.odp_actions.data;
3279 put->actions_len = facet->xout.odp_actions.size;
3281 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3282 op->slow_stub, sizeof op->slow_stub,
3283 &put->actions, &put->actions_len);
3289 /* Handles flow miss 'miss'. May add any required datapath operations
3290 * to 'ops', incrementing '*n_ops' for each new op. */
3292 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3295 struct facet *facet;
3297 miss->ofproto->n_missed += miss->stats.n_packets;
3299 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3301 /* There does not exist a bijection between 'struct flow' and datapath
3302 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3303 * assumption used throughout the facet and subfacet handling code.
3304 * Since we have to handle these misses in userspace anyway, we simply
3305 * skip facet creation, avoiding the problem altogether. */
3306 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3307 || !flow_miss_should_make_facet(miss)) {
3311 facet = facet_create(miss);
3313 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3316 static struct drop_key *
3317 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3320 struct drop_key *drop_key;
3322 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3323 &backer->drop_keys) {
3324 if (drop_key->key_len == key_len
3325 && !memcmp(drop_key->key, key, key_len)) {
3333 drop_key_clear(struct dpif_backer *backer)
3335 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3336 struct drop_key *drop_key, *next;
3338 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3341 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3343 if (error && !VLOG_DROP_WARN(&rl)) {
3344 struct ds ds = DS_EMPTY_INITIALIZER;
3345 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3346 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3347 ovs_strerror(error), ds_cstr(&ds));
3351 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3352 drop_key_destroy(drop_key);
3355 udpif_drop_key_clear(backer->udpif);
3359 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3361 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3362 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3363 struct flow_miss *miss;
3366 /* Process each element in the to-do list, constructing the set of
3367 * operations to batch. */
3369 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3370 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3372 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3374 /* Execute batch. */
3375 for (i = 0; i < n_ops; i++) {
3376 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3378 dpif_operate(backer->dpif, dpif_ops, n_ops);
3380 for (i = 0; i < n_ops; i++) {
3381 if (dpif_ops[i]->error != 0
3382 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3383 && flow_miss_ops[i].subfacet) {
3384 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3386 COVERAGE_INC(subfacet_install_fail);
3388 /* Zero-out subfacet counters when installation failed, but
3389 * datapath reported hits. This should not happen and
3390 * indicates a bug, since if the datapath flow exists, we
3391 * should not be attempting to create a new subfacet. A
3392 * buggy datapath could trigger this, so just zero out the
3393 * counters and log an error. */
3394 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3395 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3396 "datapath reported hits");
3397 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3400 subfacet->path = SF_NOT_INSTALLED;
3406 handle_upcalls(struct dpif_backer *backer)
3408 struct flow_miss_batch *fmb;
3411 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3412 struct drop_key *drop_key = drop_key_next(backer->udpif);
3417 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3418 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3419 hash_bytes(drop_key->key, drop_key->key_len, 0));
3420 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3421 drop_key->key, drop_key->key_len,
3422 NULL, 0, NULL, 0, NULL);
3424 drop_key_destroy(drop_key);
3428 fmb = flow_miss_batch_next(backer->udpif);
3430 handle_flow_misses(backer, fmb);
3431 flow_miss_batch_destroy(fmb);
3435 /* Flow expiration. */
3437 static int subfacet_max_idle(const struct dpif_backer *);
3438 static void update_stats(struct dpif_backer *);
3439 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3440 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3442 /* This function is called periodically by run(). Its job is to collect
3443 * updates for the flows that have been installed into the datapath, most
3444 * importantly when they last were used, and then use that information to
3445 * expire flows that have not been used recently.
3447 * Returns the number of milliseconds after which it should be called again. */
3449 expire(struct dpif_backer *backer)
3451 struct ofproto_dpif *ofproto;
3455 /* Periodically clear out the drop keys in an effort to keep them
3456 * relatively few. */
3457 drop_key_clear(backer);
3459 /* Update stats for each flow in the backer. */
3460 update_stats(backer);
3462 n_subfacets = hmap_count(&backer->subfacets);
3464 struct subfacet *subfacet;
3465 long long int total, now;
3469 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3470 total += now - subfacet->created;
3472 backer->avg_subfacet_life += total / n_subfacets;
3474 backer->avg_subfacet_life /= 2;
3476 backer->avg_n_subfacet += n_subfacets;
3477 backer->avg_n_subfacet /= 2;
3479 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3481 max_idle = subfacet_max_idle(backer);
3482 expire_subfacets(backer, max_idle);
3484 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3485 struct rule *rule, *next_rule;
3487 if (ofproto->backer != backer) {
3491 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3493 ovs_mutex_lock(&ofproto_mutex);
3494 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3495 &ofproto->up.expirable) {
3496 rule_expire(rule_dpif_cast(rule));
3498 ovs_mutex_unlock(&ofproto_mutex);
3500 /* All outstanding data in existing flows has been accounted, so it's a
3501 * good time to do bond rebalancing. */
3502 if (ofproto->has_bonded_bundles) {
3503 struct ofbundle *bundle;
3505 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3507 bond_rebalance(bundle->bond);
3513 return MIN(max_idle, 1000);
3516 /* Updates flow table statistics given that the datapath just reported 'stats'
3517 * as 'subfacet''s statistics. */
3519 update_subfacet_stats(struct subfacet *subfacet,
3520 const struct dpif_flow_stats *stats)
3522 struct facet *facet = subfacet->facet;
3523 struct dpif_flow_stats diff;
3525 diff.tcp_flags = stats->tcp_flags;
3526 diff.used = stats->used;
3528 if (stats->n_packets >= subfacet->dp_packet_count) {
3529 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3531 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3535 if (stats->n_bytes >= subfacet->dp_byte_count) {
3536 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3538 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3542 facet->ofproto->n_hit += diff.n_packets;
3543 subfacet->dp_packet_count = stats->n_packets;
3544 subfacet->dp_byte_count = stats->n_bytes;
3545 subfacet_update_stats(subfacet, &diff);
3547 if (facet->accounted_bytes < facet->byte_count) {
3549 facet_account(facet);
3550 facet->accounted_bytes = facet->byte_count;
3554 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3555 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3557 delete_unexpected_flow(struct dpif_backer *backer,
3558 const struct nlattr *key, size_t key_len)
3560 if (!VLOG_DROP_WARN(&rl)) {
3564 odp_flow_key_format(key, key_len, &s);
3565 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3569 COVERAGE_INC(facet_unexpected);
3570 dpif_flow_del(backer->dpif, key, key_len, NULL);
3573 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3575 * This function also pushes statistics updates to rules which each facet
3576 * resubmits into. Generally these statistics will be accurate. However, if a
3577 * facet changes the rule it resubmits into at some time in between
3578 * update_stats() runs, it is possible that statistics accrued to the
3579 * old rule will be incorrectly attributed to the new rule. This could be
3580 * avoided by calling update_stats() whenever rules are created or
3581 * deleted. However, the performance impact of making so many calls to the
3582 * datapath do not justify the benefit of having perfectly accurate statistics.
3584 * In addition, this function maintains per ofproto flow hit counts. The patch
3585 * port is not treated specially. e.g. A packet ingress from br0 patched into
3586 * br1 will increase the hit count of br0 by 1, however, does not affect
3587 * the hit or miss counts of br1.
3590 update_stats(struct dpif_backer *backer)
3592 const struct dpif_flow_stats *stats;
3593 struct dpif_flow_dump dump;
3594 const struct nlattr *key, *mask;
3595 size_t key_len, mask_len;
3597 dpif_flow_dump_start(&dump, backer->dpif);
3598 while (dpif_flow_dump_next(&dump, &key, &key_len,
3599 &mask, &mask_len, NULL, NULL, &stats)) {
3600 struct subfacet *subfacet;
3603 key_hash = odp_flow_key_hash(key, key_len);
3604 subfacet = subfacet_find(backer, key, key_len, key_hash);
3605 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3607 update_subfacet_stats(subfacet, stats);
3611 /* Stats are updated per-packet. */
3614 case SF_NOT_INSTALLED:
3616 delete_unexpected_flow(backer, key, key_len);
3621 dpif_flow_dump_done(&dump);
3624 /* Calculates and returns the number of milliseconds of idle time after which
3625 * subfacets should expire from the datapath. When a subfacet expires, we fold
3626 * its statistics into its facet, and when a facet's last subfacet expires, we
3627 * fold its statistic into its rule. */
3629 subfacet_max_idle(const struct dpif_backer *backer)
3632 * Idle time histogram.
3634 * Most of the time a switch has a relatively small number of subfacets.
3635 * When this is the case we might as well keep statistics for all of them
3636 * in userspace and to cache them in the kernel datapath for performance as
3639 * As the number of subfacets increases, the memory required to maintain
3640 * statistics about them in userspace and in the kernel becomes
3641 * significant. However, with a large number of subfacets it is likely
3642 * that only a few of them are "heavy hitters" that consume a large amount
3643 * of bandwidth. At this point, only heavy hitters are worth caching in
3644 * the kernel and maintaining in userspaces; other subfacets we can
3647 * The technique used to compute the idle time is to build a histogram with
3648 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3649 * that is installed in the kernel gets dropped in the appropriate bucket.
3650 * After the histogram has been built, we compute the cutoff so that only
3651 * the most-recently-used 1% of subfacets (but at least
3652 * flow_eviction_threshold flows) are kept cached. At least
3653 * the most-recently-used bucket of subfacets is kept, so actually an
3654 * arbitrary number of subfacets can be kept in any given expiration run
3655 * (though the next run will delete most of those unless they receive
3658 * This requires a second pass through the subfacets, in addition to the
3659 * pass made by update_stats(), because the former function never looks at
3660 * uninstallable subfacets.
3662 enum { BUCKET_WIDTH = 100 };
3663 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3664 int buckets[N_BUCKETS] = { 0 };
3665 int total, subtotal, bucket;
3666 struct subfacet *subfacet;
3670 total = hmap_count(&backer->subfacets);
3671 if (total <= flow_eviction_threshold) {
3672 return N_BUCKETS * BUCKET_WIDTH;
3675 /* Build histogram. */
3677 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3678 long long int idle = now - subfacet->used;
3679 int bucket = (idle <= 0 ? 0
3680 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3681 : (unsigned int) idle / BUCKET_WIDTH);
3685 /* Find the first bucket whose flows should be expired. */
3686 subtotal = bucket = 0;
3688 subtotal += buckets[bucket++];
3689 } while (bucket < N_BUCKETS &&
3690 subtotal < MAX(flow_eviction_threshold, total / 100));
3692 if (VLOG_IS_DBG_ENABLED()) {
3696 ds_put_cstr(&s, "keep");
3697 for (i = 0; i < N_BUCKETS; i++) {
3699 ds_put_cstr(&s, ", drop");
3702 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3705 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3709 return bucket * BUCKET_WIDTH;
3713 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3715 /* Cutoff time for most flows. */
3716 long long int normal_cutoff = time_msec() - dp_max_idle;
3718 /* We really want to keep flows for special protocols around, so use a more
3719 * conservative cutoff. */
3720 long long int special_cutoff = time_msec() - 10000;
3722 struct subfacet *subfacet, *next_subfacet;
3723 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3727 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3728 &backer->subfacets) {
3729 long long int cutoff;
3731 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3735 if (subfacet->used < cutoff) {
3736 if (subfacet->path != SF_NOT_INSTALLED) {
3737 batch[n_batch++] = subfacet;
3738 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3739 subfacet_destroy_batch(backer, batch, n_batch);
3743 subfacet_destroy(subfacet);
3749 subfacet_destroy_batch(backer, batch, n_batch);
3753 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3754 * then delete it entirely. */
3756 rule_expire(struct rule_dpif *rule)
3757 OVS_REQUIRES(ofproto_mutex)
3759 uint16_t idle_timeout, hard_timeout;
3760 long long int now = time_msec();
3763 ovs_assert(!rule->up.pending);
3765 /* Has 'rule' expired? */
3766 ovs_mutex_lock(&rule->up.mutex);
3767 hard_timeout = rule->up.hard_timeout;
3768 idle_timeout = rule->up.idle_timeout;
3769 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3770 reason = OFPRR_HARD_TIMEOUT;
3771 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3772 reason = OFPRR_IDLE_TIMEOUT;
3776 ovs_mutex_unlock(&rule->up.mutex);
3779 COVERAGE_INC(ofproto_dpif_expired);
3780 ofproto_rule_expire(&rule->up, reason);
3786 /* Creates and returns a new facet based on 'miss'.
3788 * The caller must already have determined that no facet with an identical
3789 * 'miss->flow' exists in 'miss->ofproto'.
3791 * 'rule' and 'xout' must have been created based on 'miss'.
3793 * 'facet'' statistics are initialized based on 'stats'.
3795 * The facet will initially have no subfacets. The caller should create (at
3796 * least) one subfacet with subfacet_create(). */
3797 static struct facet *
3798 facet_create(const struct flow_miss *miss)
3800 struct ofproto_dpif *ofproto = miss->ofproto;
3801 struct facet *facet;
3804 COVERAGE_INC(facet_create);
3805 facet = xzalloc(sizeof *facet);
3806 facet->ofproto = miss->ofproto;
3807 facet->used = miss->stats.used;
3808 facet->flow = miss->flow;
3809 facet->learn_rl = time_msec() + 500;
3811 list_init(&facet->subfacets);
3812 netflow_flow_init(&facet->nf_flow);
3813 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3815 xlate_out_copy(&facet->xout, &miss->xout);
3817 match_init(&match, &facet->flow, &facet->xout.wc);
3818 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3819 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3820 classifier_insert(&ofproto->facets, &facet->cr);
3821 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3823 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3828 facet_free(struct facet *facet)
3831 xlate_out_uninit(&facet->xout);
3836 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3837 * 'packet', which arrived on 'in_port'. */
3839 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3840 const struct nlattr *odp_actions, size_t actions_len,
3841 struct ofpbuf *packet)
3843 struct odputil_keybuf keybuf;
3847 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3848 odp_flow_key_from_flow(&key, flow,
3849 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3851 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3852 odp_actions, actions_len, packet);
3856 /* Remove 'facet' from its ofproto and free up the associated memory:
3858 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3859 * rule's statistics, via subfacet_uninstall().
3861 * - Removes 'facet' from its rule and from ofproto->facets.
3864 facet_remove(struct facet *facet)
3866 struct subfacet *subfacet, *next_subfacet;
3868 COVERAGE_INC(facet_remove);
3869 ovs_assert(!list_is_empty(&facet->subfacets));
3871 /* First uninstall all of the subfacets to get final statistics. */
3872 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3873 subfacet_uninstall(subfacet);
3876 /* Flush the final stats to the rule.
3878 * This might require us to have at least one subfacet around so that we
3879 * can use its actions for accounting in facet_account(), which is why we
3880 * have uninstalled but not yet destroyed the subfacets. */
3881 facet_flush_stats(facet);
3883 /* Now we're really all done so destroy everything. */
3884 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3885 &facet->subfacets) {
3886 subfacet_destroy__(subfacet);
3888 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3889 classifier_remove(&facet->ofproto->facets, &facet->cr);
3890 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3891 cls_rule_destroy(&facet->cr);
3895 /* Feed information from 'facet' back into the learning table to keep it in
3896 * sync with what is actually flowing through the datapath. */
3898 facet_learn(struct facet *facet)
3900 long long int now = time_msec();
3902 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3906 facet->learn_rl = now + 500;
3908 if (!facet->xout.has_learn
3909 && !facet->xout.has_normal
3910 && (!facet->xout.has_fin_timeout
3911 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3915 facet_push_stats(facet, true);
3919 facet_account(struct facet *facet)
3921 const struct nlattr *a;
3926 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3929 n_bytes = facet->byte_count - facet->accounted_bytes;
3931 /* This loop feeds byte counters to bond_account() for rebalancing to use
3932 * as a basis. We also need to track the actual VLAN on which the packet
3933 * is going to be sent to ensure that it matches the one passed to
3934 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3937 * We use the actions from an arbitrary subfacet because they should all
3938 * be equally valid for our purpose. */
3939 vlan_tci = facet->flow.vlan_tci;
3940 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3941 facet->xout.odp_actions.size) {
3942 const struct ovs_action_push_vlan *vlan;
3943 struct ofport_dpif *port;
3945 switch (nl_attr_type(a)) {
3946 case OVS_ACTION_ATTR_OUTPUT:
3947 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3948 if (port && port->bundle && port->bundle->bond) {
3949 bond_account(port->bundle->bond, &facet->flow,
3950 vlan_tci_to_vid(vlan_tci), n_bytes);
3954 case OVS_ACTION_ATTR_POP_VLAN:
3955 vlan_tci = htons(0);
3958 case OVS_ACTION_ATTR_PUSH_VLAN:
3959 vlan = nl_attr_get(a);
3960 vlan_tci = vlan->vlan_tci;
3966 /* Returns true if the only action for 'facet' is to send to the controller.
3967 * (We don't report NetFlow expiration messages for such facets because they
3968 * are just part of the control logic for the network, not real traffic). */
3970 facet_is_controller_flow(struct facet *facet)
3973 struct ofproto_dpif *ofproto = facet->ofproto;
3974 const struct ofpact *ofpacts;
3975 struct rule_actions *actions;
3976 struct rule_dpif *rule;
3980 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
3981 actions = rule_dpif_get_actions(rule);
3982 rule_dpif_unref(rule);
3984 ofpacts_len = actions->ofpacts_len;
3985 ofpacts = actions->ofpacts;
3986 is_controller = ofpacts_len > 0
3987 && ofpacts->type == OFPACT_CONTROLLER
3988 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
3989 rule_actions_unref(actions);
3991 return is_controller;
3996 /* Folds all of 'facet''s statistics into its rule. Also updates the
3997 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3998 * 'facet''s statistics in the datapath should have been zeroed and folded into
3999 * its packet and byte counts before this function is called. */
4001 facet_flush_stats(struct facet *facet)
4003 struct ofproto_dpif *ofproto = facet->ofproto;
4004 struct subfacet *subfacet;
4006 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4007 ovs_assert(!subfacet->dp_byte_count);
4008 ovs_assert(!subfacet->dp_packet_count);
4011 facet_push_stats(facet, false);
4012 if (facet->accounted_bytes < facet->byte_count) {
4013 facet_account(facet);
4014 facet->accounted_bytes = facet->byte_count;
4017 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4018 struct ofexpired expired;
4019 expired.flow = facet->flow;
4020 expired.packet_count = facet->packet_count;
4021 expired.byte_count = facet->byte_count;
4022 expired.used = facet->used;
4023 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4026 /* Reset counters to prevent double counting if 'facet' ever gets
4028 facet_reset_counters(facet);
4030 netflow_flow_clear(&facet->nf_flow);
4031 facet->tcp_flags = 0;
4034 /* Searches 'ofproto''s table of facets for one which would be responsible for
4035 * 'flow'. Returns it if found, otherwise a null pointer.
4037 * The returned facet might need revalidation; use facet_lookup_valid()
4038 * instead if that is important. */
4039 static struct facet *
4040 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4042 struct cls_rule *cr;
4044 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4045 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4046 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4047 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4050 /* Searches 'ofproto''s table of facets for one capable that covers
4051 * 'flow'. Returns it if found, otherwise a null pointer.
4053 * The returned facet is guaranteed to be valid. */
4054 static struct facet *
4055 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4057 struct facet *facet;
4059 facet = facet_find(ofproto, flow);
4061 && ofproto->backer->need_revalidate
4062 && !facet_revalidate(facet)) {
4070 facet_check_consistency(struct facet *facet)
4072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4074 struct xlate_out xout;
4075 struct xlate_in xin;
4077 struct rule_dpif *rule;
4080 /* Check the datapath actions for consistency. */
4081 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4082 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4083 xlate_actions(&xin, &xout);
4084 rule_dpif_unref(rule);
4086 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4087 && facet->xout.slow == xout.slow;
4088 if (!ok && !VLOG_DROP_WARN(&rl)) {
4089 struct ds s = DS_EMPTY_INITIALIZER;
4091 flow_format(&s, &facet->flow);
4092 ds_put_cstr(&s, ": inconsistency in facet");
4094 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4095 ds_put_cstr(&s, " (actions were: ");
4096 format_odp_actions(&s, facet->xout.odp_actions.data,
4097 facet->xout.odp_actions.size);
4098 ds_put_cstr(&s, ") (correct actions: ");
4099 format_odp_actions(&s, xout.odp_actions.data,
4100 xout.odp_actions.size);
4101 ds_put_char(&s, ')');
4104 if (facet->xout.slow != xout.slow) {
4105 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4110 xlate_out_uninit(&xout);
4115 /* Re-searches the classifier for 'facet':
4117 * - If the rule found is different from 'facet''s current rule, moves
4118 * 'facet' to the new rule and recompiles its actions.
4120 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4121 * where it is and recompiles its actions anyway.
4123 * - If any of 'facet''s subfacets correspond to a new flow according to
4124 * xlate_receive(), 'facet' is removed.
4126 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4128 facet_revalidate(struct facet *facet)
4130 struct ofproto_dpif *ofproto = facet->ofproto;
4131 struct rule_dpif *new_rule;
4132 struct subfacet *subfacet;
4133 struct flow_wildcards wc;
4134 struct xlate_out xout;
4135 struct xlate_in xin;
4137 COVERAGE_INC(facet_revalidate);
4139 /* Check that child subfacets still correspond to this facet. Tunnel
4140 * configuration changes could cause a subfacet's OpenFlow in_port to
4142 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4143 struct ofproto_dpif *recv_ofproto;
4144 struct flow recv_flow;
4147 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4148 subfacet->key_len, &recv_flow, NULL,
4149 &recv_ofproto, NULL);
4151 || recv_ofproto != ofproto
4152 || facet != facet_find(ofproto, &recv_flow)) {
4153 facet_remove(facet);
4158 flow_wildcards_init_catchall(&wc);
4159 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4161 /* Calculate new datapath actions.
4163 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4164 * emit a NetFlow expiration and, if so, we need to have the old state
4165 * around to properly compose it. */
4166 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4167 xlate_actions(&xin, &xout);
4168 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4170 /* A facet's slow path reason should only change under dramatic
4171 * circumstances. Rather than try to update everything, it's simpler to
4172 * remove the facet and start over.
4174 * More importantly, if a facet's wildcards change, it will be relatively
4175 * difficult to figure out if its subfacets still belong to it, and if not
4176 * which facet they may belong to. Again, to avoid the complexity, we
4177 * simply give up instead. */
4178 if (facet->xout.slow != xout.slow
4179 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4180 facet_remove(facet);
4181 xlate_out_uninit(&xout);
4182 rule_dpif_unref(new_rule);
4186 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4187 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4188 if (subfacet->path == SF_FAST_PATH) {
4189 struct dpif_flow_stats stats;
4191 subfacet_install(subfacet, &xout.odp_actions, &stats);
4192 subfacet_update_stats(subfacet, &stats);
4196 facet_flush_stats(facet);
4198 ofpbuf_clear(&facet->xout.odp_actions);
4199 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4200 xout.odp_actions.size);
4203 /* Update 'facet' now that we've taken care of all the old state. */
4204 facet->xout.slow = xout.slow;
4205 facet->xout.has_learn = xout.has_learn;
4206 facet->xout.has_normal = xout.has_normal;
4207 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4208 facet->xout.nf_output_iface = xout.nf_output_iface;
4209 facet->xout.mirrors = xout.mirrors;
4210 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4212 ovs_mutex_lock(&new_rule->up.mutex);
4213 facet->used = MAX(facet->used, new_rule->up.created);
4214 ovs_mutex_unlock(&new_rule->up.mutex);
4216 xlate_out_uninit(&xout);
4217 rule_dpif_unref(new_rule);
4222 facet_reset_counters(struct facet *facet)
4224 facet->packet_count = 0;
4225 facet->byte_count = 0;
4226 facet->prev_packet_count = 0;
4227 facet->prev_byte_count = 0;
4228 facet->accounted_bytes = 0;
4232 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4233 struct dpif_flow_stats *stats, bool may_learn)
4235 struct ofport_dpif *in_port;
4236 struct rule_dpif *rule;
4237 struct xlate_in xin;
4239 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4240 if (in_port && in_port->is_tunnel) {
4241 netdev_vport_inc_rx(in_port->up.netdev, stats);
4244 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4245 rule_dpif_credit_stats(rule, stats);
4246 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4247 xin.resubmit_stats = stats;
4248 xin.may_learn = may_learn;
4249 xlate_actions_for_side_effects(&xin);
4250 rule_dpif_unref(rule);
4254 facet_push_stats(struct facet *facet, bool may_learn)
4256 struct dpif_flow_stats stats;
4258 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4259 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4260 ovs_assert(facet->used >= facet->prev_used);
4262 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4263 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4264 stats.used = facet->used;
4265 stats.tcp_flags = facet->tcp_flags;
4267 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4268 facet->prev_packet_count = facet->packet_count;
4269 facet->prev_byte_count = facet->byte_count;
4270 facet->prev_used = facet->used;
4272 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4274 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4275 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4276 stats.n_packets, stats.n_bytes);
4277 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4282 push_all_stats__(bool run_fast)
4284 static long long int rl = LLONG_MIN;
4285 struct ofproto_dpif *ofproto;
4287 if (time_msec() < rl) {
4291 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4292 struct cls_cursor cursor;
4293 struct facet *facet;
4295 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4296 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4297 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4298 facet_push_stats(facet, false);
4303 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4306 rl = time_msec() + 100;
4310 push_all_stats(void)
4312 push_all_stats__(true);
4316 rule_dpif_credit_stats(struct rule_dpif *rule,
4317 const struct dpif_flow_stats *stats)
4319 ovs_mutex_lock(&rule->stats_mutex);
4320 rule->packet_count += stats->n_packets;
4321 rule->byte_count += stats->n_bytes;
4322 rule->up.used = MAX(rule->up.used, stats->used);
4323 ovs_mutex_unlock(&rule->stats_mutex);
4327 rule_dpif_fail_open(const struct rule_dpif *rule)
4329 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4333 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4334 OVS_REQUIRES(rule->up.mutex)
4336 return rule->up.flow_cookie;
4340 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4341 uint16_t hard_timeout)
4343 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4346 /* Returns 'rule''s actions. The caller owns a reference on the returned
4347 * actions and must eventually release it (with rule_actions_unref()) to avoid
4349 struct rule_actions *
4350 rule_dpif_get_actions(const struct rule_dpif *rule)
4352 return rule_get_actions(&rule->up);
4357 static struct subfacet *
4358 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4359 size_t key_len, uint32_t key_hash)
4361 struct subfacet *subfacet;
4363 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4364 &backer->subfacets) {
4365 if (subfacet->key_len == key_len
4366 && !memcmp(key, subfacet->key, key_len)) {
4374 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4375 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4376 * existing subfacet if there is one, otherwise creates and returns a
4378 static struct subfacet *
4379 subfacet_create(struct facet *facet, struct flow_miss *miss)
4381 struct dpif_backer *backer = miss->ofproto->backer;
4382 const struct nlattr *key = miss->key;
4383 size_t key_len = miss->key_len;
4385 struct subfacet *subfacet;
4387 key_hash = odp_flow_key_hash(key, key_len);
4389 if (list_is_empty(&facet->subfacets)) {
4390 subfacet = &facet->one_subfacet;
4392 subfacet = subfacet_find(backer, key, key_len, key_hash);
4394 if (subfacet->facet == facet) {
4398 /* This shouldn't happen. */
4399 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4400 subfacet_destroy(subfacet);
4403 subfacet = xmalloc(sizeof *subfacet);
4406 COVERAGE_INC(subfacet_create);
4407 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4408 list_push_back(&facet->subfacets, &subfacet->list_node);
4409 subfacet->facet = facet;
4410 subfacet->key = xmemdup(key, key_len);
4411 subfacet->key_len = key_len;
4412 subfacet->used = miss->stats.used;
4413 subfacet->created = subfacet->used;
4414 subfacet->dp_packet_count = 0;
4415 subfacet->dp_byte_count = 0;
4416 subfacet->path = SF_NOT_INSTALLED;
4417 subfacet->backer = backer;
4422 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4423 * its facet within 'ofproto', and frees it. */
4425 subfacet_destroy__(struct subfacet *subfacet)
4427 struct facet *facet = subfacet->facet;
4429 COVERAGE_INC(subfacet_destroy);
4430 subfacet_uninstall(subfacet);
4431 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4432 list_remove(&subfacet->list_node);
4433 free(subfacet->key);
4434 if (subfacet != &facet->one_subfacet) {
4439 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4440 * last remaining subfacet in its facet destroys the facet too. */
4442 subfacet_destroy(struct subfacet *subfacet)
4444 struct facet *facet = subfacet->facet;
4446 if (list_is_singleton(&facet->subfacets)) {
4447 /* facet_remove() needs at least one subfacet (it will remove it). */
4448 facet_remove(facet);
4450 subfacet_destroy__(subfacet);
4455 subfacet_destroy_batch(struct dpif_backer *backer,
4456 struct subfacet **subfacets, int n)
4458 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4459 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4460 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4463 for (i = 0; i < n; i++) {
4464 ops[i].type = DPIF_OP_FLOW_DEL;
4465 ops[i].u.flow_del.key = subfacets[i]->key;
4466 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4467 ops[i].u.flow_del.stats = &stats[i];
4471 dpif_operate(backer->dpif, opsp, n);
4472 for (i = 0; i < n; i++) {
4473 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4474 subfacets[i]->path = SF_NOT_INSTALLED;
4475 subfacet_destroy(subfacets[i]);
4480 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4481 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4482 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4483 * since 'subfacet' was last updated.
4485 * Returns 0 if successful, otherwise a positive errno value. */
4487 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4488 struct dpif_flow_stats *stats)
4490 struct facet *facet = subfacet->facet;
4491 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4492 const struct nlattr *actions = odp_actions->data;
4493 size_t actions_len = odp_actions->size;
4494 struct odputil_keybuf maskbuf;
4497 uint64_t slow_path_stub[128 / 8];
4498 enum dpif_flow_put_flags flags;
4501 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4504 flags |= DPIF_FP_ZERO_STATS;
4507 if (path == SF_SLOW_PATH) {
4508 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4509 slow_path_stub, sizeof slow_path_stub,
4510 &actions, &actions_len);
4513 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4514 if (enable_megaflows) {
4515 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4516 &facet->flow, UINT32_MAX);
4519 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4520 subfacet->key_len, mask.data, mask.size,
4521 actions, actions_len, stats);
4524 subfacet_reset_dp_stats(subfacet, stats);
4528 COVERAGE_INC(subfacet_install_fail);
4530 subfacet->path = path;
4535 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4537 subfacet_uninstall(struct subfacet *subfacet)
4539 if (subfacet->path != SF_NOT_INSTALLED) {
4540 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4541 struct dpif_flow_stats stats;
4544 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4545 subfacet->key_len, &stats);
4546 subfacet_reset_dp_stats(subfacet, &stats);
4548 subfacet_update_stats(subfacet, &stats);
4550 subfacet->path = SF_NOT_INSTALLED;
4552 ovs_assert(subfacet->dp_packet_count == 0);
4553 ovs_assert(subfacet->dp_byte_count == 0);
4557 /* Resets 'subfacet''s datapath statistics counters. This should be called
4558 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4559 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4560 * was reset in the datapath. 'stats' will be modified to include only
4561 * statistics new since 'subfacet' was last updated. */
4563 subfacet_reset_dp_stats(struct subfacet *subfacet,
4564 struct dpif_flow_stats *stats)
4567 && subfacet->dp_packet_count <= stats->n_packets
4568 && subfacet->dp_byte_count <= stats->n_bytes) {
4569 stats->n_packets -= subfacet->dp_packet_count;
4570 stats->n_bytes -= subfacet->dp_byte_count;
4573 subfacet->dp_packet_count = 0;
4574 subfacet->dp_byte_count = 0;
4577 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4579 * Because of the meaning of a subfacet's counters, it only makes sense to do
4580 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4581 * represents a packet that was sent by hand or if it represents statistics
4582 * that have been cleared out of the datapath. */
4584 subfacet_update_stats(struct subfacet *subfacet,
4585 const struct dpif_flow_stats *stats)
4587 if (stats->n_packets || stats->used > subfacet->used) {
4588 struct facet *facet = subfacet->facet;
4590 subfacet->used = MAX(subfacet->used, stats->used);
4591 facet->used = MAX(facet->used, stats->used);
4592 facet->packet_count += stats->n_packets;
4593 facet->byte_count += stats->n_bytes;
4594 facet->tcp_flags |= stats->tcp_flags;
4600 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4601 * the fields that were relevant as part of the lookup. */
4603 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4604 struct flow_wildcards *wc, struct rule_dpif **rule)
4606 struct ofport_dpif *port;
4608 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4611 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4613 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4614 flow->in_port.ofp_port);
4617 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4618 ofproto->no_packet_in_rule, rule);
4622 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4623 const struct flow *flow, struct flow_wildcards *wc,
4624 uint8_t table_id, struct rule_dpif **rule)
4626 const struct cls_rule *cls_rule;
4627 struct classifier *cls;
4631 if (table_id >= N_TABLES) {
4636 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4637 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4640 cls = &ofproto->up.tables[table_id].cls;
4641 ovs_rwlock_rdlock(&cls->rwlock);
4642 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4643 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4644 /* We must pretend that transport ports are unavailable. */
4645 struct flow ofpc_normal_flow = *flow;
4646 ofpc_normal_flow.tp_src = htons(0);
4647 ofpc_normal_flow.tp_dst = htons(0);
4648 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4649 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4650 cls_rule = &ofproto->drop_frags_rule->up.cr;
4652 flow_wildcards_init_exact(wc);
4655 cls_rule = classifier_lookup(cls, flow, wc);
4658 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4659 rule_dpif_ref(*rule);
4660 ovs_rwlock_unlock(&cls->rwlock);
4662 return *rule != NULL;
4665 /* Given a port configuration (specified as zero if there's no port), chooses
4666 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4667 * flow table miss. */
4669 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4670 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4672 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4673 rule_dpif_ref(*rule);
4677 rule_dpif_ref(struct rule_dpif *rule)
4680 ofproto_rule_ref(&rule->up);
4685 rule_dpif_unref(struct rule_dpif *rule)
4688 ofproto_rule_unref(&rule->up);
4693 complete_operation(struct rule_dpif *rule)
4694 OVS_REQUIRES(ofproto_mutex)
4696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4698 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4699 ofoperation_complete(rule->up.pending, 0);
4702 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4704 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4707 static struct rule *
4710 struct rule_dpif *rule = xmalloc(sizeof *rule);
4715 rule_dealloc(struct rule *rule_)
4717 struct rule_dpif *rule = rule_dpif_cast(rule_);
4722 rule_construct(struct rule *rule_)
4724 struct rule_dpif *rule = rule_dpif_cast(rule_);
4725 ovs_mutex_init(&rule->stats_mutex);
4726 ovs_mutex_lock(&rule->stats_mutex);
4727 rule->packet_count = 0;
4728 rule->byte_count = 0;
4729 ovs_mutex_unlock(&rule->stats_mutex);
4734 rule_insert(struct rule *rule_)
4735 OVS_REQUIRES(ofproto_mutex)
4737 struct rule_dpif *rule = rule_dpif_cast(rule_);
4738 complete_operation(rule);
4742 rule_delete(struct rule *rule_)
4743 OVS_REQUIRES(ofproto_mutex)
4745 struct rule_dpif *rule = rule_dpif_cast(rule_);
4746 complete_operation(rule);
4750 rule_destruct(struct rule *rule_)
4752 struct rule_dpif *rule = rule_dpif_cast(rule_);
4753 ovs_mutex_destroy(&rule->stats_mutex);
4757 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4759 struct rule_dpif *rule = rule_dpif_cast(rule_);
4761 /* push_all_stats() can handle flow misses which, when using the learn
4762 * action, can cause rules to be added and deleted. This can corrupt our
4763 * caller's datastructures which assume that rule_get_stats() doesn't have
4764 * an impact on the flow table. To be safe, we disable miss handling. */
4765 push_all_stats__(false);
4767 /* Start from historical data for 'rule' itself that are no longer tracked
4768 * in facets. This counts, for example, facets that have expired. */
4769 ovs_mutex_lock(&rule->stats_mutex);
4770 *packets = rule->packet_count;
4771 *bytes = rule->byte_count;
4772 ovs_mutex_unlock(&rule->stats_mutex);
4776 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4777 struct ofpbuf *packet)
4779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4780 struct dpif_flow_stats stats;
4781 struct xlate_out xout;
4782 struct xlate_in xin;
4784 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4785 rule_dpif_credit_stats(rule, &stats);
4787 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4788 xin.resubmit_stats = &stats;
4789 xlate_actions(&xin, &xout);
4791 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4792 xout.odp_actions.size, packet);
4794 xlate_out_uninit(&xout);
4798 rule_execute(struct rule *rule, const struct flow *flow,
4799 struct ofpbuf *packet)
4801 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4802 ofpbuf_delete(packet);
4807 rule_modify_actions(struct rule *rule_, bool reset_counters)
4808 OVS_REQUIRES(ofproto_mutex)
4810 struct rule_dpif *rule = rule_dpif_cast(rule_);
4812 if (reset_counters) {
4813 ovs_mutex_lock(&rule->stats_mutex);
4814 rule->packet_count = 0;
4815 rule->byte_count = 0;
4816 ovs_mutex_unlock(&rule->stats_mutex);
4819 complete_operation(rule);
4822 /* Sends 'packet' out 'ofport'.
4823 * May modify 'packet'.
4824 * Returns 0 if successful, otherwise a positive errno value. */
4826 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4828 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4831 error = xlate_send_packet(ofport, packet);
4833 ofproto->stats.tx_packets++;
4834 ofproto->stats.tx_bytes += packet->size;
4838 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4839 * The action will state 'slow' as the reason that the action is in the slow
4840 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4841 * dump-flows" output to see why a flow is in the slow path.)
4843 * The 'stub_size' bytes in 'stub' will be used to store the action.
4844 * 'stub_size' must be large enough for the action.
4846 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4849 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4850 enum slow_path_reason slow,
4851 uint64_t *stub, size_t stub_size,
4852 const struct nlattr **actionsp, size_t *actions_lenp)
4854 union user_action_cookie cookie;
4857 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4858 cookie.slow_path.unused = 0;
4859 cookie.slow_path.reason = slow;
4861 ofpbuf_use_stack(&buf, stub, stub_size);
4862 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4863 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4865 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4867 odp_port_t odp_port;
4870 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4871 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4872 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4874 *actionsp = buf.data;
4875 *actions_lenp = buf.size;
4879 set_frag_handling(struct ofproto *ofproto_,
4880 enum ofp_config_flags frag_handling)
4882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4883 if (frag_handling != OFPC_FRAG_REASM) {
4884 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4892 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4893 const struct flow *flow,
4894 const struct ofpact *ofpacts, size_t ofpacts_len)
4896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4897 struct odputil_keybuf keybuf;
4898 struct dpif_flow_stats stats;
4899 struct xlate_out xout;
4900 struct xlate_in xin;
4904 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4905 odp_flow_key_from_flow(&key, flow,
4906 ofp_port_to_odp_port(ofproto,
4907 flow->in_port.ofp_port));
4909 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4911 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
4912 xin.resubmit_stats = &stats;
4913 xin.ofpacts_len = ofpacts_len;
4914 xin.ofpacts = ofpacts;
4916 xlate_actions(&xin, &xout);
4917 dpif_execute(ofproto->backer->dpif, key.data, key.size,
4918 xout.odp_actions.data, xout.odp_actions.size, packet);
4919 xlate_out_uninit(&xout);
4927 set_netflow(struct ofproto *ofproto_,
4928 const struct netflow_options *netflow_options)
4930 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4932 if (netflow_options) {
4933 if (!ofproto->netflow) {
4934 ofproto->netflow = netflow_create();
4935 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4937 return netflow_set_options(ofproto->netflow, netflow_options);
4938 } else if (ofproto->netflow) {
4939 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4940 netflow_destroy(ofproto->netflow);
4941 ofproto->netflow = NULL;
4948 get_netflow_ids(const struct ofproto *ofproto_,
4949 uint8_t *engine_type, uint8_t *engine_id)
4951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4953 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4957 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
4959 if (!facet_is_controller_flow(facet) &&
4960 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4961 struct subfacet *subfacet;
4962 struct ofexpired expired;
4964 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4965 if (subfacet->path == SF_FAST_PATH) {
4966 struct dpif_flow_stats stats;
4968 subfacet_install(subfacet, &facet->xout.odp_actions,
4970 subfacet_update_stats(subfacet, &stats);
4974 expired.flow = facet->flow;
4975 expired.packet_count = facet->packet_count;
4976 expired.byte_count = facet->byte_count;
4977 expired.used = facet->used;
4978 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4983 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
4985 struct cls_cursor cursor;
4986 struct facet *facet;
4988 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4989 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4990 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4991 send_active_timeout(ofproto, facet);
4993 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4996 static struct ofproto_dpif *
4997 ofproto_dpif_lookup(const char *name)
4999 struct ofproto_dpif *ofproto;
5001 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5002 hash_string(name, 0), &all_ofproto_dpifs) {
5003 if (!strcmp(ofproto->up.name, name)) {
5011 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5012 const char *argv[], void *aux OVS_UNUSED)
5014 struct ofproto_dpif *ofproto;
5017 ofproto = ofproto_dpif_lookup(argv[1]);
5019 unixctl_command_reply_error(conn, "no such bridge");
5022 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5023 mac_learning_flush(ofproto->ml);
5024 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5026 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5027 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5028 mac_learning_flush(ofproto->ml);
5029 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5033 unixctl_command_reply(conn, "table successfully flushed");
5036 static struct ofport_dpif *
5037 ofbundle_get_a_port(const struct ofbundle *bundle)
5039 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5044 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5045 const char *argv[], void *aux OVS_UNUSED)
5047 struct ds ds = DS_EMPTY_INITIALIZER;
5048 const struct ofproto_dpif *ofproto;
5049 const struct mac_entry *e;
5051 ofproto = ofproto_dpif_lookup(argv[1]);
5053 unixctl_command_reply_error(conn, "no such bridge");
5057 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5058 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5059 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5060 struct ofbundle *bundle = e->port.p;
5061 char name[OFP_MAX_PORT_NAME_LEN];
5063 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5065 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5066 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5067 mac_entry_age(ofproto->ml, e));
5069 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5070 unixctl_command_reply(conn, ds_cstr(&ds));
5075 struct xlate_out xout;
5076 struct xlate_in xin;
5082 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5084 struct rule_actions *actions;
5087 ds_put_char_multiple(result, '\t', level);
5089 ds_put_cstr(result, "No match\n");
5093 ovs_mutex_lock(&rule->up.mutex);
5094 cookie = rule->up.flow_cookie;
5095 ovs_mutex_unlock(&rule->up.mutex);
5097 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5098 rule ? rule->up.table_id : 0, ntohll(cookie));
5099 cls_rule_format(&rule->up.cr, result);
5100 ds_put_char(result, '\n');
5102 actions = rule_dpif_get_actions(rule);
5104 ds_put_char_multiple(result, '\t', level);
5105 ds_put_cstr(result, "OpenFlow ");
5106 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5107 ds_put_char(result, '\n');
5109 rule_actions_unref(actions);
5113 trace_format_flow(struct ds *result, int level, const char *title,
5114 struct trace_ctx *trace)
5116 ds_put_char_multiple(result, '\t', level);
5117 ds_put_format(result, "%s: ", title);
5118 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5119 ds_put_cstr(result, "unchanged");
5121 flow_format(result, &trace->xin.flow);
5122 trace->flow = trace->xin.flow;
5124 ds_put_char(result, '\n');
5128 trace_format_regs(struct ds *result, int level, const char *title,
5129 struct trace_ctx *trace)
5133 ds_put_char_multiple(result, '\t', level);
5134 ds_put_format(result, "%s:", title);
5135 for (i = 0; i < FLOW_N_REGS; i++) {
5136 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5138 ds_put_char(result, '\n');
5142 trace_format_odp(struct ds *result, int level, const char *title,
5143 struct trace_ctx *trace)
5145 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5147 ds_put_char_multiple(result, '\t', level);
5148 ds_put_format(result, "%s: ", title);
5149 format_odp_actions(result, odp_actions->data, odp_actions->size);
5150 ds_put_char(result, '\n');
5154 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5156 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5157 struct ds *result = trace->result;
5159 ds_put_char(result, '\n');
5160 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5161 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5162 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5163 trace_format_rule(result, recurse + 1, rule);
5167 trace_report(struct xlate_in *xin, const char *s, int recurse)
5169 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5170 struct ds *result = trace->result;
5172 ds_put_char_multiple(result, '\t', recurse);
5173 ds_put_cstr(result, s);
5174 ds_put_char(result, '\n');
5178 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5179 void *aux OVS_UNUSED)
5181 const struct dpif_backer *backer = NULL;
5182 struct ofproto_dpif *ofproto;
5183 struct ofpbuf odp_key, odp_mask;
5184 struct ofpbuf *packet;
5187 struct simap port_names;
5193 ofpbuf_init(&odp_key, 0);
5194 ofpbuf_init(&odp_mask, 0);
5195 simap_init(&port_names);
5197 /* Handle "-generate" or a hex string as the last argument. */
5198 if (!strcmp(argv[argc - 1], "-generate")) {
5199 packet = ofpbuf_new(0);
5202 const char *error = eth_from_hex(argv[argc - 1], &packet);
5205 } else if (argc == 4) {
5206 /* The 3-argument form must end in "-generate' or a hex string. */
5207 unixctl_command_reply_error(conn, error);
5212 /* odp_flow can have its in_port specified as a name instead of port no.
5213 * We do not yet know whether a given flow is a odp_flow or a br_flow.
5214 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5215 * we need to create a simap of name to port no. */
5217 const char *dp_type;
5218 if (!strncmp(argv[1], "ovs-", 4)) {
5219 dp_type = argv[1] + 4;
5223 backer = shash_find_data(&all_dpif_backers, dp_type);
5225 struct shash_node *node;
5226 if (shash_count(&all_dpif_backers) == 1) {
5227 node = shash_first(&all_dpif_backers);
5228 backer = node->data;
5231 if (backer && backer->dpif) {
5232 struct dpif_port dpif_port;
5233 struct dpif_port_dump port_dump;
5234 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5235 simap_put(&port_names, dpif_port.name,
5236 odp_to_u32(dpif_port.port_no));
5240 /* Parse the flow and determine whether a datapath or
5241 * bridge is specified. If function odp_flow_key_from_string()
5242 * returns 0, the flow is a odp_flow. If function
5243 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5244 if (!odp_flow_from_string(argv[argc - 1], &port_names, &odp_key, &odp_mask)) {
5246 unixctl_command_reply_error(conn, "Cannot find the datapath");
5250 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5251 NULL, &ofproto, NULL)) {
5252 unixctl_command_reply_error(conn, "Invalid datapath flow");
5255 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5256 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5258 unixctl_command_reply_error(conn, "Must specify bridge name");
5262 ofproto = ofproto_dpif_lookup(argv[1]);
5264 unixctl_command_reply_error(conn, "Unknown bridge name");
5268 unixctl_command_reply_error(conn, "Bad flow syntax");
5272 /* Generate a packet, if requested. */
5274 if (!packet->size) {
5275 flow_compose(packet, &flow);
5277 union flow_in_port in_port_;
5279 in_port_ = flow.in_port;
5280 ds_put_cstr(&result, "Packet: ");
5281 s = ofp_packet_to_string(packet->data, packet->size);
5282 ds_put_cstr(&result, s);
5285 /* Use the metadata from the flow and the packet argument
5286 * to reconstruct the flow. */
5287 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5292 ofproto_trace(ofproto, &flow, packet, &result);
5293 unixctl_command_reply(conn, ds_cstr(&result));
5296 ds_destroy(&result);
5297 ofpbuf_delete(packet);
5298 ofpbuf_uninit(&odp_key);
5299 ofpbuf_uninit(&odp_mask);
5300 simap_destroy(&port_names);
5304 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5305 const struct ofpbuf *packet, struct ds *ds)
5307 struct rule_dpif *rule;
5308 struct flow_wildcards wc;
5310 ds_put_cstr(ds, "Flow: ");
5311 flow_format(ds, flow);
5312 ds_put_char(ds, '\n');
5314 flow_wildcards_init_catchall(&wc);
5315 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5317 trace_format_rule(ds, 0, rule);
5318 if (rule == ofproto->miss_rule) {
5319 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5320 } else if (rule == ofproto->no_packet_in_rule) {
5321 ds_put_cstr(ds, "\nNo match, packets dropped because "
5322 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5323 } else if (rule == ofproto->drop_frags_rule) {
5324 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5325 "and the fragment handling mode is \"drop\".\n");
5329 uint64_t odp_actions_stub[1024 / 8];
5330 struct ofpbuf odp_actions;
5331 struct trace_ctx trace;
5335 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5338 ofpbuf_use_stub(&odp_actions,
5339 odp_actions_stub, sizeof odp_actions_stub);
5340 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5341 trace.xin.resubmit_hook = trace_resubmit;
5342 trace.xin.report_hook = trace_report;
5344 xlate_actions(&trace.xin, &trace.xout);
5345 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5347 ds_put_char(ds, '\n');
5348 trace_format_flow(ds, 0, "Final flow", &trace);
5350 match_init(&match, flow, &trace.xout.wc);
5351 ds_put_cstr(ds, "Relevant fields: ");
5352 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5353 ds_put_char(ds, '\n');
5355 ds_put_cstr(ds, "Datapath actions: ");
5356 format_odp_actions(ds, trace.xout.odp_actions.data,
5357 trace.xout.odp_actions.size);
5359 if (trace.xout.slow) {
5360 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5361 "slow path because it:");
5362 switch (trace.xout.slow) {
5364 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5367 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5370 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5373 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5375 case SLOW_CONTROLLER:
5376 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5377 "to the OpenFlow controller.");
5384 xlate_out_uninit(&trace.xout);
5387 rule_dpif_unref(rule);
5390 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5391 * 'reply' describing the results. */
5393 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5395 struct cls_cursor cursor;
5396 struct facet *facet;
5400 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5401 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5402 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5403 if (!facet_check_consistency(facet)) {
5407 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5409 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5413 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5414 ofproto->up.name, errors);
5416 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5421 ofproto_dpif_self_check(struct unixctl_conn *conn,
5422 int argc, const char *argv[], void *aux OVS_UNUSED)
5424 struct ds reply = DS_EMPTY_INITIALIZER;
5425 struct ofproto_dpif *ofproto;
5428 ofproto = ofproto_dpif_lookup(argv[1]);
5430 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5431 "ofproto/list for help)");
5434 ofproto_dpif_self_check__(ofproto, &reply);
5436 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5437 ofproto_dpif_self_check__(ofproto, &reply);
5441 unixctl_command_reply(conn, ds_cstr(&reply));
5445 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5446 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5447 * to destroy 'ofproto_shash' and free the returned value. */
5448 static const struct shash_node **
5449 get_ofprotos(struct shash *ofproto_shash)
5451 const struct ofproto_dpif *ofproto;
5453 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5454 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5455 shash_add_nocopy(ofproto_shash, name, ofproto);
5458 return shash_sort(ofproto_shash);
5462 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5463 const char *argv[] OVS_UNUSED,
5464 void *aux OVS_UNUSED)
5466 struct ds ds = DS_EMPTY_INITIALIZER;
5467 struct shash ofproto_shash;
5468 const struct shash_node **sorted_ofprotos;
5471 shash_init(&ofproto_shash);
5472 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5473 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5474 const struct shash_node *node = sorted_ofprotos[i];
5475 ds_put_format(&ds, "%s\n", node->name);
5478 shash_destroy(&ofproto_shash);
5479 free(sorted_ofprotos);
5481 unixctl_command_reply(conn, ds_cstr(&ds));
5486 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5488 const struct shash_node **ofprotos;
5489 struct ofproto_dpif *ofproto;
5490 struct shash ofproto_shash;
5491 uint64_t n_hit, n_missed;
5494 n_hit = n_missed = 0;
5495 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5496 if (ofproto->backer == backer) {
5497 n_missed += ofproto->n_missed;
5498 n_hit += ofproto->n_hit;
5502 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5503 dpif_name(backer->dpif), n_hit, n_missed);
5504 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5505 " life span: %lldms\n", hmap_count(&backer->subfacets),
5506 backer->avg_n_subfacet, backer->max_n_subfacet,
5507 backer->avg_subfacet_life);
5509 shash_init(&ofproto_shash);
5510 ofprotos = get_ofprotos(&ofproto_shash);
5511 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5512 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5513 const struct shash_node **ports;
5516 if (ofproto->backer != backer) {
5520 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5521 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5523 ports = shash_sort(&ofproto->up.port_by_name);
5524 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5525 const struct shash_node *node = ports[j];
5526 struct ofport *ofport = node->data;
5528 odp_port_t odp_port;
5530 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5533 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5534 if (odp_port != ODPP_NONE) {
5535 ds_put_format(ds, "%"PRIu32":", odp_port);
5537 ds_put_cstr(ds, "none:");
5540 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5543 if (!netdev_get_config(ofport->netdev, &config)) {
5544 const struct smap_node **nodes;
5547 nodes = smap_sort(&config);
5548 for (i = 0; i < smap_count(&config); i++) {
5549 const struct smap_node *node = nodes[i];
5550 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5551 node->key, node->value);
5555 smap_destroy(&config);
5557 ds_put_char(ds, ')');
5558 ds_put_char(ds, '\n');
5562 shash_destroy(&ofproto_shash);
5567 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5568 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5570 struct ds ds = DS_EMPTY_INITIALIZER;
5571 const struct shash_node **backers;
5574 backers = shash_sort(&all_dpif_backers);
5575 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5576 dpif_show_backer(backers[i]->data, &ds);
5580 unixctl_command_reply(conn, ds_cstr(&ds));
5584 /* Dump the megaflow (facet) cache. This is useful to check the
5585 * correctness of flow wildcarding, since the same mechanism is used for
5586 * both xlate caching and kernel wildcarding.
5588 * It's important to note that in the output the flow description uses
5589 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5591 * This command is only needed for advanced debugging, so it's not
5592 * documented in the man page. */
5594 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5595 int argc OVS_UNUSED, const char *argv[],
5596 void *aux OVS_UNUSED)
5598 struct ds ds = DS_EMPTY_INITIALIZER;
5599 const struct ofproto_dpif *ofproto;
5600 long long int now = time_msec();
5601 struct cls_cursor cursor;
5602 struct facet *facet;
5604 ofproto = ofproto_dpif_lookup(argv[1]);
5606 unixctl_command_reply_error(conn, "no such bridge");
5610 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5611 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5612 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5613 cls_rule_format(&facet->cr, &ds);
5614 ds_put_cstr(&ds, ", ");
5615 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5616 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5617 ds_put_cstr(&ds, "Datapath actions: ");
5618 if (facet->xout.slow) {
5619 uint64_t slow_path_stub[128 / 8];
5620 const struct nlattr *actions;
5623 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5624 slow_path_stub, sizeof slow_path_stub,
5625 &actions, &actions_len);
5626 format_odp_actions(&ds, actions, actions_len);
5628 format_odp_actions(&ds, facet->xout.odp_actions.data,
5629 facet->xout.odp_actions.size);
5631 ds_put_cstr(&ds, "\n");
5633 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5635 ds_chomp(&ds, '\n');
5636 unixctl_command_reply(conn, ds_cstr(&ds));
5640 /* Disable using the megaflows.
5642 * This command is only needed for advanced debugging, so it's not
5643 * documented in the man page. */
5645 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5646 int argc OVS_UNUSED,
5647 const char *argv[] OVS_UNUSED,
5648 void *aux OVS_UNUSED)
5650 struct ofproto_dpif *ofproto;
5652 enable_megaflows = false;
5654 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5655 flush(&ofproto->up);
5658 unixctl_command_reply(conn, "megaflows disabled");
5661 /* Re-enable using megaflows.
5663 * This command is only needed for advanced debugging, so it's not
5664 * documented in the man page. */
5666 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5667 int argc OVS_UNUSED,
5668 const char *argv[] OVS_UNUSED,
5669 void *aux OVS_UNUSED)
5671 struct ofproto_dpif *ofproto;
5673 enable_megaflows = true;
5675 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5676 flush(&ofproto->up);
5679 unixctl_command_reply(conn, "megaflows enabled");
5683 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5684 int argc OVS_UNUSED, const char *argv[],
5685 void *aux OVS_UNUSED)
5687 struct ds ds = DS_EMPTY_INITIALIZER;
5688 const struct ofproto_dpif *ofproto;
5689 struct subfacet *subfacet;
5691 ofproto = ofproto_dpif_lookup(argv[1]);
5693 unixctl_command_reply_error(conn, "no such bridge");
5697 update_stats(ofproto->backer);
5699 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5700 struct facet *facet = subfacet->facet;
5701 struct odputil_keybuf maskbuf;
5704 if (facet->ofproto != ofproto) {
5708 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5709 if (enable_megaflows) {
5710 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5711 &facet->flow, UINT32_MAX);
5714 odp_flow_format(subfacet->key, subfacet->key_len,
5715 mask.data, mask.size, NULL, &ds, false);
5717 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5718 subfacet->dp_packet_count, subfacet->dp_byte_count);
5719 if (subfacet->used) {
5720 ds_put_format(&ds, "%.3fs",
5721 (time_msec() - subfacet->used) / 1000.0);
5723 ds_put_format(&ds, "never");
5725 if (subfacet->facet->tcp_flags) {
5726 ds_put_cstr(&ds, ", flags:");
5727 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5730 ds_put_cstr(&ds, ", actions:");
5731 if (facet->xout.slow) {
5732 uint64_t slow_path_stub[128 / 8];
5733 const struct nlattr *actions;
5736 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5737 slow_path_stub, sizeof slow_path_stub,
5738 &actions, &actions_len);
5739 format_odp_actions(&ds, actions, actions_len);
5741 format_odp_actions(&ds, facet->xout.odp_actions.data,
5742 facet->xout.odp_actions.size);
5744 ds_put_char(&ds, '\n');
5747 unixctl_command_reply(conn, ds_cstr(&ds));
5752 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5753 int argc OVS_UNUSED, const char *argv[],
5754 void *aux OVS_UNUSED)
5756 struct ds ds = DS_EMPTY_INITIALIZER;
5757 struct ofproto_dpif *ofproto;
5759 ofproto = ofproto_dpif_lookup(argv[1]);
5761 unixctl_command_reply_error(conn, "no such bridge");
5765 flush(&ofproto->up);
5767 unixctl_command_reply(conn, ds_cstr(&ds));
5772 ofproto_dpif_unixctl_init(void)
5774 static bool registered;
5780 unixctl_command_register(
5782 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5783 1, 3, ofproto_unixctl_trace, NULL);
5784 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5785 ofproto_unixctl_fdb_flush, NULL);
5786 unixctl_command_register("fdb/show", "bridge", 1, 1,
5787 ofproto_unixctl_fdb_show, NULL);
5788 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5789 ofproto_dpif_self_check, NULL);
5790 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5791 ofproto_unixctl_dpif_dump_dps, NULL);
5792 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5794 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5795 ofproto_unixctl_dpif_dump_flows, NULL);
5796 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5797 ofproto_unixctl_dpif_del_flows, NULL);
5798 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5799 ofproto_unixctl_dpif_dump_megaflows, NULL);
5800 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5801 ofproto_unixctl_dpif_disable_megaflows, NULL);
5802 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5803 ofproto_unixctl_dpif_enable_megaflows, NULL);
5806 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5808 * This is deprecated. It is only for compatibility with broken device drivers
5809 * in old versions of Linux that do not properly support VLANs when VLAN
5810 * devices are not used. When broken device drivers are no longer in
5811 * widespread use, we will delete these interfaces. */
5814 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5816 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5817 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5819 if (realdev_ofp_port == ofport->realdev_ofp_port
5820 && vid == ofport->vlandev_vid) {
5824 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5826 if (ofport->realdev_ofp_port) {
5829 if (realdev_ofp_port && ofport->bundle) {
5830 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5831 * themselves be part of a bundle. */
5832 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5835 ofport->realdev_ofp_port = realdev_ofp_port;
5836 ofport->vlandev_vid = vid;
5838 if (realdev_ofp_port) {
5839 vsp_add(ofport, realdev_ofp_port, vid);
5846 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5848 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5852 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5853 OVS_EXCLUDED(ofproto->vsp_mutex)
5857 ovs_mutex_lock(&ofproto->vsp_mutex);
5858 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
5859 ovs_mutex_unlock(&ofproto->vsp_mutex);
5864 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5865 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5866 OVS_REQUIRES(ofproto->vsp_mutex)
5868 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5869 int vid = vlan_tci_to_vid(vlan_tci);
5870 const struct vlan_splinter *vsp;
5872 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5873 hash_realdev_vid(realdev_ofp_port, vid),
5874 &ofproto->realdev_vid_map) {
5875 if (vsp->realdev_ofp_port == realdev_ofp_port
5876 && vsp->vid == vid) {
5877 return vsp->vlandev_ofp_port;
5881 return realdev_ofp_port;
5884 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5885 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5886 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5887 * 'vlan_tci' 9, it would return the port number of eth0.9.
5889 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5890 * function just returns its 'realdev_ofp_port' argument. */
5892 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5893 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5894 OVS_EXCLUDED(ofproto->vsp_mutex)
5898 ovs_mutex_lock(&ofproto->vsp_mutex);
5899 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5900 ovs_mutex_unlock(&ofproto->vsp_mutex);
5904 static struct vlan_splinter *
5905 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5907 struct vlan_splinter *vsp;
5909 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5910 hash_ofp_port(vlandev_ofp_port),
5911 &ofproto->vlandev_map) {
5912 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5920 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5921 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5922 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5923 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5924 * eth0 and store 9 in '*vid'.
5926 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5927 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5930 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5931 ofp_port_t vlandev_ofp_port, int *vid)
5932 OVS_REQUIRES(ofproto->vsp_mutex)
5934 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5935 const struct vlan_splinter *vsp;
5937 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5942 return vsp->realdev_ofp_port;
5948 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5949 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5950 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5951 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
5952 * always the case unless VLAN splinters are enabled), returns false without
5953 * making any changes. */
5955 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
5956 OVS_EXCLUDED(ofproto->vsp_mutex)
5961 ovs_mutex_lock(&ofproto->vsp_mutex);
5962 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5963 ovs_mutex_unlock(&ofproto->vsp_mutex);
5968 /* Cause the flow to be processed as if it came in on the real device with
5969 * the VLAN device's VLAN ID. */
5970 flow->in_port.ofp_port = realdev;
5971 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5976 vsp_remove(struct ofport_dpif *port)
5978 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5979 struct vlan_splinter *vsp;
5981 ovs_mutex_lock(&ofproto->vsp_mutex);
5982 vsp = vlandev_find(ofproto, port->up.ofp_port);
5984 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5985 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5988 port->realdev_ofp_port = 0;
5990 VLOG_ERR("missing vlan device record");
5992 ovs_mutex_unlock(&ofproto->vsp_mutex);
5996 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6000 ovs_mutex_lock(&ofproto->vsp_mutex);
6001 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6002 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6003 == realdev_ofp_port)) {
6004 struct vlan_splinter *vsp;
6006 vsp = xmalloc(sizeof *vsp);
6007 vsp->realdev_ofp_port = realdev_ofp_port;
6008 vsp->vlandev_ofp_port = port->up.ofp_port;
6011 port->realdev_ofp_port = realdev_ofp_port;
6013 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6014 hash_ofp_port(port->up.ofp_port));
6015 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6016 hash_realdev_vid(realdev_ofp_port, vid));
6018 VLOG_ERR("duplicate vlan device record");
6020 ovs_mutex_unlock(&ofproto->vsp_mutex);
6024 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6026 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6027 return ofport ? ofport->odp_port : ODPP_NONE;
6030 struct ofport_dpif *
6031 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6033 struct ofport_dpif *port;
6035 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6036 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6037 &backer->odp_to_ofport_map) {
6038 if (port->odp_port == odp_port) {
6039 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6044 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6049 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6051 struct ofport_dpif *port;
6053 port = odp_port_to_ofport(ofproto->backer, odp_port);
6054 if (port && &ofproto->up == port->up.ofproto) {
6055 return port->up.ofp_port;
6061 const struct ofproto_class ofproto_dpif_class = {
6096 port_is_lacp_current,
6097 NULL, /* rule_choose_table */
6106 rule_modify_actions,
6120 get_stp_port_status,
6127 is_mirror_output_bundle,
6128 forward_bpdu_changed,
6129 set_mac_table_config,
6131 NULL, /* meter_get_features */
6132 NULL, /* meter_set */
6133 NULL, /* meter_get */
6134 NULL, /* meter_del */
6135 NULL, /* group_alloc */
6136 NULL, /* group_construct */
6137 NULL, /* group_destruct */
6138 NULL, /* group_dealloc */
6139 NULL, /* group_modify */
6140 NULL, /* group_get_stats */